# AI+Education Summit 2026

https://www.youtube.com/watch?v=EqouaCgSo-k
Translation: zh-CN

[00:07] All right, let's see.
  好的，我们来看看。

[00:07] Are we on?
  我们开始了吗？

[00:11] Hey.
  嘿。

[00:13] All right.
  好的。

[00:13] Okay.
  好的。

[00:13] Good morning, everybody.
  大家早上好。

[00:19] Wait a second here.
  等一下。

[00:19] All right.
  好的。

[00:23] All right.
  好的。

[00:23] Well, good morning again.
  嗯，再次祝大家早上好。

[00:25] Um, my name is Patrick Hines.
  嗯，我叫帕特里克·海因斯。

[00:27] I am senior manager of research communities at Stanford HAI.
  我是斯坦福HAI研究社区的高级经理。

[00:32] And it's my honor to welcome everyone to this wonderful new space at the graduate school of education.
  我很荣幸欢迎大家来到教育研究生院这个很棒的新空间。

[00:35] And um I just like to start by thanking all of you for joining us today.
  嗯，我想首先感谢大家今天加入我们。

[00:40] Um so last year in my opening remarks I I made a point of highlighting the velocity of change that we're seeing as a result of the current advances in AI.
  嗯，去年在我开幕致辞中，我特别强调了由于当前人工智能的进步，我们所看到的变革的速度。

[00:52] And it's fair to say that the speed of change has has has has really only continued over the past year.
  可以说，在过去的一年里，变革的速度确实只增不减。

[01:00] You know, last year I made a quick note of the fact that from the first time we had this summit, we were talking in hypotheticals to last year reaching a
  你知道，去年我曾简要地提到，从我们第一次举办这次峰会开始，我们一直在谈论假设，直到去年达到一个

[01:09] point where the majority of K through 2 students in the United States had noted that they used a generative AI tool in the past week.
  美国大多数K-2年级学生都指出，他们在过去一周使用过生成式人工智能工具。

[01:16] And so it's really fair to say that that speed has only continued over the past year.
  因此，可以肯定地说，这种速度在过去一年中只增不减。

[01:23] And so, yes, that means we're here again because addressing rapid change requires conversation.
  所以，是的，这意味着我们又聚在这里，因为应对快速变化需要对话。

[01:28] It requires convening and it requires a commitment to shaping a future in which AI is developed and deployed so that its benefits are broadly shared.
  它需要聚集，需要致力于塑造一个人工智能得到开发和部署，从而使其惠益得到广泛分享的未来。

[01:40] But we're also here because everyone in this room shares a few common characteristics.
  但我们之所以也在这里，是因为在座的每个人都具备一些共同的特质。

[01:45] We care.
  我们关心。

[01:45] We're curious.
  我们充满好奇。

[01:49] And we're invested in creating community.
  我们致力于建设社区。

[01:52] So, let me unpack those just quickly.
  那么，让我快速地解释一下。

[01:54] When I say that we care, well, yes, we all care about something.
  当我提到我们关心时，是的，我们都关心某件事。

[02:00] Um, but for today though, what I'm referencing is the fact that we all care deeply about building a future for educators and for learners that is made better by the integration of AI.
  嗯，但就今天而言，我所指的是我们都深切关注为教育者和学习者建设一个更美好的未来，这个未来将通过人工智能的整合而变得更好。

[02:08] And we care about
  我们关心

[02:10] making thoughtful decisions about where, when, why, and how we use AI.
  就如何使用人工智能做出深思熟虑的决定。

[02:13] Let me scroll the screen.
  让我滚动屏幕。

[02:17] And when I say that we're curious, well, yes, we're curious about the world that we're building, but we're also curious about how the integration of AI is going to affect the future for teachers, learners, and others.
  当我们说我们感到好奇时，是的，我们对我们正在构建的世界感到好奇，但我们也对人工智能的整合将如何影响教师、学习者和其他人的未来感到好奇。

[02:32] And I also think that that curiosity is a reflection of not just the interest, but also the excitement that we have about hearing from learn leaders in the field and also about best-in-class research that's emerged over the past year or so.
  而且，我认为这种好奇心不仅反映了兴趣，也反映了我们从该领域的学习领导者那里听到消息以及过去一年左右出现的最佳研究的兴奋之情。

[02:47] And so finally that brings me to my final point which is that of creating community.
  所以，最后，这让我想到我的最后一点，那就是创建社区。

[02:50] So as we start the day um I'd like to thank everyone again for your curiosity and for your care.
  所以，当我们开始这一天的时候，我想再次感谢大家的求知欲和关心。

[02:58] Um, and it's precisely because we have those shared values that we can have this space for the emergence of a multistakeholder, multidisciplinary community consisting of people that are passionate about ensuring that AI
  正是因为我们拥有这些共同的价值观，我们才能为多方利益相关者、多学科社区的出现提供空间，这个社区由那些热衷于确保人工智能的人组成。

[03:11] delivers on its promise while also allowing us to work together to mitigate its risks.
  兑现了其承诺，同时也使我们能够共同努力来减轻其风险。

[03:18] So, thank you all again for being here and I look forward to learning and sharing with all of you today.
  所以，再次感谢大家的光临，我期待着今天与大家一起学习和分享。

[03:25] Isabelle, over to you.
  伊莎贝尔，交给你了。

[03:25] Thank you, Patrick, and thank you to all of you.
  谢谢你，帕特里克，也谢谢你们大家。

[03:29] Welcome to the 2026 AI and Education Summit.
  欢迎来到2026人工智能与教育峰会。

[03:34] Yay.
  耶。

[03:40] So, uh it's morning, so we are going to do a little gym exercise.
  那么，呃，现在是早上，所以我们要做一个小小的健身运动。

[03:46] Okay.
  好的。

[03:47] First uh I would like to ask PE for people for whom this is the first AI and education summit to stand up.
  首先，呃，我想请一下，对于第一次参加人工智能与教育峰会的人，请站起来。

[04:01] Sit back.
  坐下。

[04:03] So we have about you know 40 to 50% of the room and if you start looking around I'm really thrilled to invite welcome
  所以我们有大约，你知道，房间里有40%到50%的人，如果你开始环顾四周，我真的很高兴邀请欢迎

[04:16] um a lot of K12 teachers and their students high schoolers and middle schoolers.
  有很多K12教师和他们的学生，高中生和初中生。

[04:25] So look at look around. They were with us for an incredible day yesterday and uh they are also here for the summit uh today.
  所以看看，看看周围。他们昨天和我们一起度过了难以置信的一天，嗯，他们今天也在这里参加峰会。

[04:30] asked some questions.
  问了一些问题。

[04:33] They were so incredibly thoughtful.
  他们非常体贴。

[04:35] I learned so much from the group yesterday.
  我从昨天的小组中学到了很多。

[04:38] We also have a different group that also was with us yesterday.
  我们还有一个不同的团体，他们昨天也和我们在一起。

[04:42] Um all the winners of the great AI challenge and you will hear from some of them later on this afternoon.
  嗯，所有伟大人工智能挑战赛的获胜者，今天下午晚些时候你会听到他们中的一些人的发言。

[04:53] Uh many of them are also first timers uh for this event.
  嗯，他们中的许多人也是第一次参加这个活动。

[04:56] So congratulations and welcome.
  所以祝贺你们，欢迎你们。

[05:00] Stay up.
  保持清醒。

[05:00] No no no no.
  不不不不。

[05:03] The gym is not done.
  健身房还没结束。

[05:07] Now I'm going to ask people to get up if at your second AI and education summit.
  现在我要请大家起立，如果在你们的第二届人工智能与教育峰会上。

[05:16] Few more people.
  还有几个人。

[05:17] Okay.
  好的。

[05:18] You know where this is going.
  你知道这是怎么回事。

[05:21] People who have attended free AI and education summits.
  参加过免费人工智能和教育峰会的人。

[05:26] All right.
  好吧。

[05:27] All right.
  好吧。

[05:29] Okay.
  好的。

[05:29] And the really really special people who have attended all four, please get up.
  还有那些参加了全部四次活动，非常非常特别的人，请起立。

[05:42] So if we can move maybe and thank you, we can now sit down and welcome everyone.
  所以，如果我们能稍微移动一下，谢谢，我们现在可以坐下并欢迎大家。

[05:48] Um, welcome everyone.
  嗯，欢迎大家。

[05:48] So this is our fourth um AI and education summit.
  这是我们的第四届人工智能和教育峰会。

[05:57] Um, it has evolved quite a bit and um um this is the first time that we are also welcoming such a large group in our new graduate school of education building that we in inaugurated back in November last year.
  嗯，它已经发展了很多，嗯嗯，这是我们第一次在去年十一月落成的新的教育研究生院大楼里欢迎如此大的一个团体。

[06:13] So let me just say a few months ago um this incredible beautiful
  所以，让我说几个月前，嗯，这个令人难以置信的美丽的

[06:19] space that uh I think is very well deserved for education.
  我认为这是教育界非常应得的空间。

[06:27] Uh um please also I will ask for all for all of you for your grace because this is the first time we have such a large group in uh this building uh including I think the first time we have actually AV for example uh with us.
  呃嗯，请大家也请你们大家多多包涵，因为这是我们第一次有这么多人聚集在这个建筑里，包括我认为也是第一次有AV设备，例如，和我们在一起。

[06:44] So uh I will uh if something is not completely working thank you for your uh patience and grace with all of you.
  所以，如果有什么不完全正常，感谢你们大家的耐心和包容。

[06:51] So the title of this year is the inflection point.
  所以今年的主题是拐点。

[06:55] Why the inflection point?
  为什么是拐点？

[06:55] Because simply um not only research and practice but certainly research where we sit here at at Stanford University is actually quite divided on um the impact of AI in education.
  因为，简单来说，不仅是研究和实践，当然，我们斯坦福大学的研究也相当分歧，关于人工智能在教育中的影响。

[07:11] Next slide please.
  下一张幻灯片，请。

[07:14] And that's a lot of words but just to say that's what you're going to hear today.
  这有很多话，但只是想说，这就是你们今天将要听到的。

[07:18] you're going to hear and on each
  你们将要听到，并且在每一个

[07:20] panel we made sure actually that they are different voices different voices on where we see gains from research and practice on where AI can actually advance learning for all learners but we'll also hear about the opposite views where there are meaningful risks um uh from this new technology um in our education systems.
  小组我们实际上确保了它们是不同的声音，不同的声音，关于我们在研究和实践中看到的收益，关于人工智能如何真正地促进所有学习者的学习，但我们也将听到相反的观点，即存在有意义的风险，嗯，嗯，来自这项新技术，嗯，在我们的教育体系中。

[07:48] And this divide I think is very healthy.
  我认为这种分歧非常有益。

[07:51] Um and uh this convenience I think stands at at this particular juncture where to me this divide suggests that the outcome of this technology and the use of it will depends not only on technology for sure on its progress but really on pedagogy and how we use uh this technology in our education systems as well as something that we at the accelerator for learning in close partnership with our friends at
  嗯，嗯，我认为这种便利性处于这个特殊的关节点，对我来说，这种分歧表明这项技术的成果及其使用不仅取决于技术，当然取决于其进步，而且真正取决于教学法以及我们如何在教育体系中使用这项技术，以及我们加速学习中心与我们的朋友们密切合作所做的事情。

[08:21] Stanford for AI are working on which is ensuring that learning sciences are at the forefront of this wave of innovation.
  斯坦福人工智能团队正在致力于确保学习科学处于这场创新浪潮的前沿。

[08:29] And ensuring that learning science are embedded in everything what we do both in practice in innovation and technology.
  并确保学习科学融入我们所做的一切，包括实践、创新和技术。

[08:38] And last not least ensuring that human relationships and context remain central to everything that we think as education is a human endeavor.
  最后但同样重要的是，确保人际关系和背景始终是我们所认为的教育——一项人类事业——的核心。

[08:49] So with all of us the divide here on this inflection point is ultimately a design choice.
  因此，在我们所有人看来，这个转折点上的分歧最终是一个设计选择。

[08:56] And I think this is why we are all gathered today because it's not a predetermined outcome.
  我想这就是我们今天聚集在这里的原因，因为它不是一个预先确定的结果。

[09:02] It's one that we all together can actually help shape.
  这是我们可以共同塑造的一个结果。

[09:08] And I believe this is what we are going to do today together.
  我相信这正是在我们今天将要共同做的事情。

[09:14] With this uh let me open the floor to their first incredible session which is inviting Dennis Pope um who is one of a senior um.
  那么，请允许我将会议交给他们的第一个精彩环节，邀请丹尼斯·波普，他是一位资深人士。

[09:24] lecturer at the graduate school of education and um leader of challenge success who is going to interview Randy Michelle.
  教育研究生院的讲师，也是挑战成功项目的负责人，他将采访兰迪·米歇尔。

[09:35] is works on AI um AI broadly including education for governor nuome here in California.
  他从事人工智能研究，广泛涉及人工智能，包括在加利福尼亚州诺姆州长的人工智能教育。

[09:44] So with that let's welcome our first session. Thank you.
  那么，让我们开始我们的第一次会议。谢谢。

[10:00] Okay.
  好的。

[10:04] All right. Well, thanks Isabelle. I am so excited to be here in this beautiful new school of education building speaking to all of you.
  好的。嗯，谢谢伊莎贝尔。我非常激动能来到这座美丽的教育学院新大楼，和大家在这里交流。

[10:14] Um, as Isabelle mentioned, I'm Denise Pope. I'm a senior lecturer here and I happen to be at the right time at the right place to start researching AI even though my kids think
  嗯，正如伊莎贝尔提到的，我是丹妮丝·波普。我是这里的资深讲师，恰好在正确的时间、正确的地点开始研究人工智能，尽管我的孩子们认为

[10:25] That's hysterical because I still need their help sometimes to do it work.
  这太搞笑了，因为我有时仍然需要他们的帮助来完成这项工作。

[10:30] Um I co-founded Challenge Success over 20 years ago to partner with schools and communities to really center the student experience of school and to improve uh K12.
  我于20多年前共同创立了Challenge Success，与学校和社区合作，真正以学校的学生体验为中心，并改善K12教育。

[10:42] And one of the things I studied was student stress and academic integrity issues in schools.
  我研究的一个方面是学校里的学生压力和学术诚信问题。

[10:47] So I had lots and lots of longitudinal data to um think about and think about its connection to student engagement and student well-being and student belonging.
  因此，我有很多很多纵向数据可以思考，并思考它与学生参与度、学生福祉和学生归属感之间的联系。

[10:58] And so when chat GPT was widely released in November 2022, I decided to partner with Victor Lee and the Stanford Accelerator for Learning um to study the effect of this new thing uh in school not just on academic integrity but also how it relates to how student engage um or not with their learning.
  因此，当ChatGPT于2022年11月广泛发布时，我决定与Victor Lee和斯坦福学习加速器合作，研究这个新事物在学校里对学术诚信的影响，以及它如何影响学生参与学习的程度。

[11:19] And so since then, Victor and his team and Challenge Success and my team have been looking really closely at how teachers and students are using AI at the classroom.
  从那时起，Victor和他的团队，以及Challenge Success和我的团队一直在密切关注教师和学生在课堂上如何使用人工智能。

[11:27] Level.
  层面。

[11:29] And we work with schools to help them consider effective interventions and policies.
  我们与学校合作，帮助他们考虑有效的干预措施和政策。

[11:35] And now this morning, I'm excited to talk with Randy to get an even broader picture of AI policy and implementation at the state level here in California.
  今天上午，我很高兴能与兰迪交谈，以便更全面地了解加州在此层面的AI政策和实施情况。

[11:44] So I am really excited to introduce Randy.
  所以我非常激动地介绍兰迪。

[11:46] We are honored to have Randy here.
  我们很荣幸能请到兰迪。

[11:49] She serves as senior adviser for technology to Governor Gavin Nuome.
  她担任州长Gavin Nuome的技术高级顾问。

[11:52] Uh in this role, she leads Nuome's tech policy, including AI, digital infrastructure, cyber security, child online safety, and other related topics, which is big.
  在此职位上，她领导Nuome的技术政策，包括人工智能、数字基础设施、网络安全、儿童在线安全以及其他相关主题，这很重要。

[12:04] Prior to joining the governor's office, she served in the Biden administration at the White House National Security Council as director for technology and democracy as well as director for international cyber policy.
  在加入州长办公室之前，她曾在拜登政府的国家安全委员会担任技术与民主事务主管，以及国际网络政策事务主管。

[12:15] Randy previously worked as an election security adviser at the US Department of State, including tours in DC, Kenya, and the Philippines.
  兰迪此前曾在Such as Department of State担任选举安全顾问，包括在华盛顿特区、肯尼亚和菲律宾的任职经历。

[12:24] So, Randy really knows what she's talking about.
  所以，兰迪确实知道她在说什么。

[12:25] She received her JD from Yale Law School and her BA
  她获得了耶鲁法学院的法学博士学位和文学学士学位

[12:28] from Harvard, that little university across the way.
  来自哈佛，那个在路对面的小大学。

[12:30] Um, so Randy, thank you for being here.
  嗯，兰迪，谢谢你来这里。

[12:33] Welcome. Welcome.
  欢迎。欢迎。

[12:36] Since the title of this year's summit is the inflection point, my first question to you is, is this moment fundamentally different from past education or technology waves in your mind?
  既然今年峰会的主题是拐点，我的第一个问题是，在你看来，这个时刻与过去的教育或技术浪潮有根本性的不同吗？

[12:48] And if so, why?
  如果是的话，为什么？

[12:49] Well, first of all, thank you so much for having me.
  嗯，首先，非常感谢你们邀请我来。

[12:50] It's a real thrill to be here.
  很高兴来到这里。

[12:52] Um, as even though my bio may suggest otherwise, I am a big fan of Stanford.
  嗯，尽管我的简历可能表明并非如此，但我非常喜欢斯坦福。

[12:58] So, thank you for having.
  所以，谢谢你的邀请。

[12:59] We're happy to hear that.
  我们很高兴听到这个。

[13:01] That's good.
  那很好。

[13:01] Um, and I also want to just preface this by saying that while I'm thrilled to share the governor's vision on AI and education, I really am here to hear from you all and especially the educators and students that are really on the front lines of this conversation and uh tackling the challenges and taking advantage of the opportunities that AI poses in classrooms.
  嗯，而且我想先说一下，虽然我很高兴能分享州长关于人工智能和教育的愿景，但我来这里主要是想听听你们所有人的意见，特别是那些真正处于这场对话前沿的教育工作者和学生，他们正在应对挑战，并利用人工智能在课堂上带来的机遇。

[13:28] having me. Thank you for sharing your wisdom here.
  感谢你在这里分享你的智慧。

[13:32] Um and hopefully I can help provide some broad uh sort of strategic framework to contextualize the rest of today's conversation.
  嗯，希望我能提供一个广泛的、战略性的框架，以 the rest of today's conversation 为背景。

[13:40] Uh to your question about the inflection point today, I think in a similar way that you know I'm not don't have to tell anyone here this that um AI poses something fundamentally different than prior technologies.
  呃，关于你今天关于拐点的提问，我认为以一种类似的方式，你知道我不用告诉这里的任何人，人工智能比以往任何技术都提出了根本性的不同。

[13:55] when we had the invention of the blackboard, the chalkboard, the invention of the calculator, the invention of internet, and now of AI, they each have fundamentally changed the way that learning and teaching works, but in very different ways.
  当我们发明了黑板、粉笔板、计算器、互联网，以及现在的人工智能时，它们都从根本上改变了学习和教学的方式，但方式却大不相同。

[14:15] So, if you start with just the fundamental chalkboard, you switched from individualized education to a broad-based classroom teaching calculator.
  所以，如果你从最基本的黑板开始，你就从个性化教育转向了广泛的课堂教学计算器。

[14:24] You know, there was a time when we were all saying, "Everyone's going to forget how to do math.
  你知道，曾经有一段时间我们都在说，“每个人都会忘记如何做数学。

[14:29] not going to be able to do any math anymore.
  将无法进行任何数学运算了。

[14:30] It's being all done for you.
  一切都为你做好了。

[14:31] But it switched from just understanding how to add or calculate numbers to really understanding the theory and the reasons behind it, having to show those proofs that I really hated doing in high school math.
  但它从仅仅理解如何加减或计算数字，转变为真正理解其背后的理论和原因，不得不展示那些我非常讨厌在高中数学中做的证明。

[14:41] Um, and then you shift to internet and I think all of us here remember there was an era where people said Wikipedia is going to ruin education.
  嗯，然后你转向互联网，我想我们这里所有人都记得曾有一个时代，人们说维基百科将毁掉教育。

[14:53] we're not going to be able to do research anymore.
  我们将无法再进行研究了。

[14:55] We're not going to be able to uh teach our children.
  我们将无法教我们的孩子。

[14:57] They're just going to have all the answers right in front of us.
  他们将直接在我们面前获得所有答案。

[15:01] But really, what it did is shift the type of teaching from being able to identify facts to being able to really analyze and consolidate and apply um and the real sort of cognitive thinking that we uh and that that you all here at Stanford are so good at teaching.
  但实际上，它所做的就是将教学类型从能够识别事实转变为能够真正地分析、巩固和应用，嗯，以及我们在这里斯坦福的各位都非常擅长教授的那种真正的认知思维。

[15:24] Um, and it no longer is helpful to know the first the year that a war started or to
  嗯，知道战争开始的年份或者

[15:29] know the um birth date of a president because you can look that up so easily.
  知道总统的出生日期，因为你可以很容易地查到。

[15:35] Uh and I think AI as you all know presents something fundamentally different while also continuing that thread where there there has been concerns over the years around the impacts of technology on teaching and those concerns have in many ways been addressed and amilarated by the very thoughtful educators and policy makers in this room.
  嗯，我想你们都知道，人工智能提出了根本不同的东西，同时也延续了这些年来围绕技术对教学影响的担忧的思路，而这些担忧在很多方面已经被这个房间里非常周到的教育工作者和政策制定者所解决和缓和。

[16:00] And so I do think that it's important not to catastrophize what we have here today.
  所以，我认为我们不应该对今天我们所拥有的东西感到恐慌，这一点很重要。

[16:05] But at the same time, AI is capable of a degree of analysis and cognitive capabilities that we didn't see previously.
  但与此同时，人工智能能够进行我们以前未曾见过的程度的分析和认知能力。

[16:15] Which begs the question of what's next?
  这就引出了一个问题：接下来是什么？

[16:17] if all of these inventions to date have brought us to the point where we've really been focusing on the analytics and the um sort of cognitive load then what is the
  如果迄今为止的所有这些发明都让我们达到了一个真正关注分析和认知负荷的程度，那么什么是

[16:33] Next step if that's not a purely human function.
  下一步，如果那不是纯粹人类的功能。

[16:37] And I think we I mean that is what I'm I'm curious to hear from you all.
  我想我们，我的意思是，我很想听听你们所有人的看法。

[16:39] But I do think there's a question about what kind of um sort of checks do we need to have.
  但我确实认为有一个问题，关于我们需要什么样的嗯，某种程度的检查。

[16:42] Where humans need to be able to understand is this thing that the AI is telling me is it right?
  人类需要能够理解的是，人工智能告诉我的这个东西是对的吗？

[16:50] Is it biased in some way?
  它是否存在某种偏见？

[16:54] What are the um broad what's the broader context here?
  嗯，更广泛的，这里的更广泛的背景是什么？

[16:58] Is it missing something?
  它是否遗漏了什么？

[17:00] But then what does that mean in terms of preparing our students to be able to flourish and thrive in the working world?
  但这对我们如何培养学生在工作世界中茁壮成长意味着什么？

[17:07] Does it mean you know even just a few months or years ago we were saying we need to do more training for prompt engineering?
  这是否意味着，你知道，就在几个月或几年前，我们还在说我们需要为提示工程做更多的培训？

[17:14] And now most of the prompt engineering is done by AI.
  而现在，大部分的提示工程都是由人工智能完成的。

[17:16] And so thinking about what does that mean on a practical level?
  所以，从实际层面思考，这意味着什么？

[17:22] What are the skills that are going to be most useful?
  哪些技能将是最有用的？

[17:24] Perhaps it's the interpersonal skills.
  也许是人际交往能力。

[17:25] It's the public speaking skills.
  是公开演讲的技巧。

[17:28] It's the human relationship side which would fundamentally change some of the
  是人际关系方面，这将从根本上改变一些

[17:34] STEM focus for example that we've had over the years and So I think these are still evolving conversations that need to be had.
  例如，我们多年来一直关注STEM。所以，我认为这些仍然是需要进行的不断发展的对话。

[17:41] U and part of the reason why events like today are so important.
  U，而像今天这样的活动如此重要的部分原因在于此。

[17:47] I hear you.
  我明白了。

[17:47] Yes.
  是的。

[17:47] Yes to all.
  都同意。

[17:51] Larry Cuban would be very proud of that walk through history from the chalkboard on up.
  拉里·库班会为从粉笔开始的那段历史回顾感到非常自豪。

[17:55] He's absolutely saying the same thing.
  他绝对是这么说的。

[17:57] He's a professor of meritis here at Stanford.
  他是斯坦福大学的荣誉教授。

[17:59] So maybe I'm just trying to justify my own liberal arts degree.
  所以，也许我只是想为我自己的文科学位辩护。

[18:02] I think maybe.
  我想是的。

[18:02] What was your major?
  你的专业是什么？

[18:05] uh social studies which is uh like a social theory.
  呃，社会学，也就是社会理论。

[18:08] Uh Larry Cuban would be even more proud.
  呃，拉里·库班会更自豪。

[18:10] Yes.
  是的。

[18:10] So okay what so as we think about this walk us through some of the things you're doing and working on with the governor and your team to think about the role of policy and and you know given that things are moving at the pace that they're moving and we're sort of building the plane while we're flying it which typically happens after a new innovation.
  那么，好的，当我们考虑这一点时，请向我们介绍您正在与州长和您的团队一起做的一些事情，以及您正在处理的一些事情，以思考政策的作用，以及，你知道，考虑到事情正在以它们正在发展的速度发展，而我们有点像在飞行中建造飞机，这通常发生在新的创新之后。

[18:32] Uh, how's California leading
  呃，加州是如何领先的

[18:35] In defining responsible human-centered AI in education for the country?
  在定义该国教育领域负责任的人性化人工智能方面？

[18:43] The governor really likes to talk about how we are both promoting innovation and protecting the safety and well-being of Californians.
  州长非常喜欢谈论我们如何在促进创新和保护加州人安全福祉之间取得平衡。

[18:53] I think it's really important to note that those two things are not mutually incompatible.
  我认为需要注意的是，这两件事并非相互排斥。

[18:58] I think there's often a conversation where you have to choose one or the other, but in California, we have shown and proven that you can in fact do both.
  我认为人们常常认为必须二选一，但在加州，我们已经表明并证明了实际上可以兼顾两者。

[19:06] And there are actually regulations that are common sense guardrails that in fact promote innovation.
  实际上，有一些法规就像常识性的保障措施，它们实际上促进了创新。

[19:15] I think one really important uh historical example here is that one of the reasons why Silicon Valley where we are here today in addition to the presence of Stanford and other um educational institutions is that California had restrictions on non-compete clauses and contracts.
  我认为一个非常重要的历史例子是，我们今天所在的硅谷之所以能发展起来，除了斯坦福大学和其他教育机构的存在之外，还有一个原因是加州对竞业禁止条款和合同进行了限制。

[19:34] And so employees that had a an idea from
  因此，有了一个想法的员工

[19:39] Their prior business were able to leave those previous professions and start a business here because they didn't have the non-compete contracts in their employment contracts.
  他们之前的企业能够离开那些以前的职业，在这里创业，因为他们的雇佣合同中没有竞业禁止条款。

[19:50] And so that led to this amazing innovation ecosystem in Silicon Valley where again because of among other things that regulation and so that is the sort of ethos that we think about these issues but that regulation has to be smart and it has to be deliberate and it has to find that balance and and that's challenging.
  因此，这导致了硅谷这个令人惊叹的创新生态系统，再次因为包括监管在内的各种因素，这就是我们思考这些问题的精神，但监管必须是明智的，必须是深思熟虑的，必须找到那个平衡，而这是具有挑战性的。

[20:19] Um I also want to just give a broad overview about how we're thinking about AI policy at large.
  嗯，我还想大致概述一下我们对人工智能政策的总体看法。

[20:23] Um we're really follow uh operating under sort of three core pillars.
  嗯，我们确实遵循着大约三个核心支柱在运作。

[20:28] The first is modernizing government service delivery leveraging AI to improve the way that government both uh does the sort of back
  第一个是现代化政府服务交付，利用人工智能来改进政府在后台所做的工作方式。

[20:41] of house operations. How do we do our

[20:43] own permits? How do we do our own sort

[20:46] of filing like all all the nonsexy

[20:49] things but the things that really make

[20:50] government work better? Uh two, how do

[20:53] we use AI to improve customer experience

[20:56] in call centers where for example we

[20:59] issued a pilot on um leveraging AI to

[21:03] support the tax call center where it

[21:06] helps um the the individuals at the call

[21:10] center be able to get the answer for the

[21:12] the customer as quickly as possible. Um,

[21:15] and so we're really delving as deeply as

[21:18] possible into exploring, again, not all

[21:21] use cases need AI, but there are some

[21:24] certain instances, and we are continuing

[21:26] to explore what those look like. Um, and

[21:29] doing that again with rights and safety

[21:32] in mind where we have extensive risk

[21:35] mitigation measures in place and we have

[21:38] extensive collaboration with labor

[21:41] through that process. The the second

[21:44] sort of bucket that we think about is on

[21:46] promoting the innovation economy and

[21:49] also empowering uh workers to thrive in

[21:52] an economic and technological

[21:55] transition. Um part of this work has

[21:59] included for example uh the governor

[22:01] last summer announced new partnerships

[22:04] with um Google, Microsoft, IBM and Adobe

[22:09] and the year prior with Nvidia to

[22:12] support AI training u both in the CSU

[22:16] systems and community college systems

[22:18] and in Adobe's case 9 through 12

[22:21] education. And the idea here is that we

[22:25] need to make sure that our workers and

[22:29] our students are set up to excel and to

[22:34] thrive in an economy that really will

[22:36] require understanding how to leverage

[22:38] AI. And then the third area is in

[22:42] protecting vulnerable communities,

[22:44] protecting rights and safety, um, and

[22:46] making sure that especially that we're

[22:48] we're protecting the well-being of

[22:51] Californians. This includes extensive

[22:53] work on protecting child safety. The

[22:56] governor signed a number of bills last

[22:58] year, including a new companion chatbot

[23:01] bill that requires um certain

[23:04] disclosures of uh suicide risk and self

[23:07] harm by companion chatbot companies. It

[23:10] requires uh reporting to the California

[23:12] Department of Public Health and a number

[23:15] of other restrictions. We're also very

[23:17] focused on privacy. The governor uh last

[23:19] year signed the first ever um

[23:22] browserwide opt out bill which will

[23:25] require browsers to have essentially a

[23:28] single click where you can rather than

[23:30] having all of those very annoying popups

[23:33] that come every single website. You can

[23:35] just toggle your browser once to say

[23:37] that you opt out of thirdparty cookie

[23:40] sharing. Um, and we actually just

[23:42] launched a new drop uh platform where

[23:46] you can again with a single click opt

[23:48] out of data brokers uh tracking and

[23:51] selling your data. Um, so really

[23:54] thinking about this from the the broad

[23:56] spectrum and then as it applies to

[23:58] education, I think it really cuts across

[24:01] all three of those categories where we

[24:04] need to empower our educational

[24:07] institutions to leverage AI to improve

[24:10] all of their sort of operational

[24:12] efficiencies. some of the things that

[24:13] the administrators and teachers here

[24:15] know far better than I do, but I'm sure

[24:17] are not fun parts of your job that we've

[24:19] really seen AI support some of these

[24:22] back of house operations to making sure

[24:24] that um we are empowering uh students

[24:29] and teachers and workers to understand

[24:31] how to use AI. Um and then also that we

[24:34] are making sure that we're protecting

[24:36] rights and safety. And I'm happy to talk

[24:39] about some more examples here, but

[24:40] that's the sort of broad framework we're

[24:42] thinking about.

[24:44] On behalf of everyone here, I'm saying

[24:46] thank you. Thank you for some of the

[24:47] little things like those opt outs,

[24:49] right? And thank you for thinking like

[24:52] um but broader than just how are we

[24:55] using it in the government, how you know

[24:57] really understanding that we need we

[24:59] need your help. So we appreciate that.

[25:01] And and that's kind of a segue to the

[25:03] role of research. So you've got

[25:05] researchers in the room. um how are how

[25:08] are you or are you and your team and if

[25:10] so how are you using research and

[25:13] learning science to help shape some of

[25:15] these uh policies and some of the

[25:17] strategy? So the governor feels very

[25:20] strongly that our policy needs to be

[25:22] rooted in expertise. In fact, it's one

[25:26] of the reasons why um last I guess a

[25:30] couple of years ago now the governor

[25:31] commissioned a report by Stanford,

[25:34] Berkeley, Carnegie Endowment and others

[25:38] um on frontier AI safety and that report

[25:41] led by Dr. Fe Lee of Stanford um among

[25:45] others has laid the foundation for SP53

[25:49] which is a um as many of you know the

[25:52] first in the country AI safety

[25:55] legislation that includes various

[25:58] transparency and disclosure requirements

[26:01] uh for large language models and for

[26:04] frontier AI um systems. And so we really

[26:10] value the expertise of that

[26:13] evidence-based policym. Um we also the

[26:17] governor launched um just a few months

[26:20] ago back in November the governor's

[26:23] innovation council which brings together

[26:26] a range of academic experts, think

[26:29] tanks, um advocates who really are are

[26:34] deeply um have a deep expertise in uh

[26:39] four key areas. is modernizing

[26:41] government service delivery, um

[26:44] promoting the innovation economy,

[26:45] combating tech- enabled fraud, and

[26:48] protecting child online safety. Um and

[26:51] so we are thrilled to have a number of

[26:53] Stanford professors as part of that

[26:55] group and really leaning on them to be

[26:58] able to advise policy makers to

[27:01] collaborate with us on a range of

[27:03] initiatives that I'm happy to delve more

[27:06] more deeply into. Um, and finally, just

[27:10] I have a number of colleagues here today

[27:13] at events like this. We really value

[27:15] this kind of input. And so again, just

[27:18] very grateful for your collaboration

[27:20] both today and moving forward.

[27:23] If you have some research that you want

[27:25] to do or ideas for research, right, this

[27:27] is this is your opportunity to talk to

[27:29] Randy. What if as it comes to mind, is

[27:31] there some big piece that's missing that

[27:33] you would love to see us kind of dig

[27:35] into as researchers? I think as we are

[27:38] still figuring out how all of this

[27:41] works, how AI can best be leveraged in

[27:44] the classroom, how we protect our

[27:46] students privacy, our students

[27:47] well-being.

[27:49] I would love to see more pilots that

[27:52] then we have actual clinical data of how

[27:56] this works in the real world. and it's

[27:58] still early, so it's understandable that

[27:59] that hasn't um been as readily

[28:03] available. But we actually um are

[28:06] working with the California Department

[28:08] for Public for Public Health on a pilot

[28:12] that um will deliver an AI literacy

[28:15] curriculum that includes both how

[28:18] parents and excuse me, how teachers and

[28:21] school administrators should teach about

[28:23] sort of understanding how to use AI, but

[28:25] also the sort of rights and safety

[28:27] components of it. How how do we help

[28:30] classrooms deal with mental health

[28:33] impacts, with cyber bullying, with the

[28:36] issues around cognitive decline, with

[28:39] the issues around deep fake pornography

[28:41] and um non-conentual intimate imagery

[28:44] and child sexual abuse material. And so

[28:47] we're looking to pilot that in

[28:49] classrooms in the state um hopefully in

[28:52] the next few at some at some point in

[28:55] this year. Um and we'll we're really

[28:58] excited to get more data. Um and if

[29:00] there are other pilots or other sort of

[29:03] data from actually implementing

[29:06] um interventions that would be extremely

[29:09] valuable to us.

[29:10] I think there's many folks in here who

[29:12] who are nodding and saying yes and we're

[29:14] here to help and and and talk to you. So

[29:17] that that's great. Um Isabelle listed a

[29:20] whole bunch of things that we should be

[29:22] worried about. Um, you just named a

[29:24] whole bunch of things that we should be

[29:25] worried about around mental health,

[29:27] around safety, uh, you know, protection,

[29:31] cognitive offloading,

[29:33] we have environmental issues, we have

[29:35] bias, we have, you know, so many things.

[29:36] What what keeps you personally up at

[29:38] night? Like what's the number one like

[29:40] if we do anything, I I got to make sure

[29:42] this doesn't happen. I think personally

[29:45] sorry and this is not necessarily

[29:47] reflective of of the govern I don't know

[29:50] what out of all of these different

[29:51] issues is the governor's necessarily

[29:53] priority and I think the answer is

[29:54] probably we are addressing all of these

[29:57] issues and they're all very important at

[30:00] the same time while balancing the

[30:01] opportunities that these tools provide.

[30:03] Um, but I think one of the issues that

[30:07] doesn't get talked about as much in the

[30:09] educational context in particular is

[30:11] just the proliferation of non-consentual

[30:13] intimate imagery in schools. Um, I I

[30:17] think the latest data from a 2023 Thorn

[30:20] survey said that one in eight of young

[30:23] people know someone who was personally

[30:26] victimized. I think it's one in 16 were

[30:29] personally victimized by non-consensual

[30:31] intimate imagery. And that was almost

[30:34] three years ago. And as we know, the

[30:37] technology has rapidly evolved and um

[30:41] these tools are increasingly

[30:43] easy to use. It's part of the reason why

[30:46] the governor signed last year new

[30:49] legislation allowing for private right

[30:52] of action um against non-conensual

[30:54] intimate imagery. Also, um, in

[30:57] California, CSAM, AI generated CSAM is

[31:01] illegal just like any other form of C

[31:03] SAM. But, um, it's not something that I

[31:06] think we've we talk about enough in

[31:08] these contexts. I I would like to do

[31:11] better to empower schools to sort of

[31:13] understand how to deal with these kinds

[31:16] of issues, particularly when the

[31:18] perpetrator is also a minor, and how

[31:20] what is the appropriate recourse. Um,

[31:23] and if you are a victim, um, as again, I

[31:27] don't need to tell all of you, but um,

[31:29] if you're a victim to cyber bullying

[31:31] through deep fake pornography, it it not

[31:35] only is a well-being and mental health

[31:38] risk, but it also impedes learning. And

[31:42] um it's so so important that we uh

[31:45] protect our children from this kind of

[31:49] both illegal and um I think unethical

[31:52] behavior.

[31:53] Yeah, we know I mean we know so much

[31:55] that you can't learn if you don't feel

[31:57] safe. Um you know it's all

[32:00] neurologically intertwined, right?

[32:01] Social, emotional, cognitive links. So,

[32:04] uh, that's huge and I'm really glad

[32:06] that, um, it's on the radar screen, uh,

[32:09] for policy and hopefully maybe with that

[32:11] new partnership with the Department of

[32:13] Health, too, as they're thinking about

[32:14] rolling out the lessons. Um, so what,

[32:18] you know,

[32:20] what should be that's regulated, what

[32:22] should be regulated now, as you think

[32:26] about the future? I think I think

[32:27] there's people who think, well, once the

[32:28] genie is let out of the bottle, it's

[32:30] really hard to bring it back in. What do

[32:33] you think?

[32:34] So I think this technology is rapidly

[32:37] changing. We are

[32:40] both trying to get ahead, trying to keep

[32:43] up and we have to be able to iterate as

[32:46] we get more data, as we better

[32:48] understand how to protect children, how

[32:51] to empower children. Uh, and so I don't

[32:56] I think it's a little nihilistic to

[32:58] think about like we have to do

[32:59] everything now and if we don't do it

[33:01] now, we'll never be able to do it in the

[33:02] future because we have to continue to

[33:04] iterate and to learn. And I think we saw

[33:06] this with phone free schools that even

[33:09] though people thought the genie was out

[33:10] of the bottle and that we're never going

[33:13] to be able to get phones out of the

[33:15] classroom, we've seen it take hold

[33:19] across the state and increasingly across

[33:21] the country. And at least from what I've

[33:24] heard and I'm look forward to hearing

[33:26] more from you all that it's been a a

[33:28] resounding success. And so I think it's

[33:32] it's um a little too fatalistic to say

[33:34] that, you know, there's a point when

[33:37] we've reached the the point past no

[33:39] return, the po point of no return, and

[33:42] that we should be consistently learning

[33:44] and consistently improving upon our

[33:47] approach both on the sort of education

[33:49] positive perspective and on the

[33:52] regulatory protective point of view. So

[33:56] both and I like the example of of the

[33:58] cell phones that you know they're not

[33:59] back in the bottle but they're we're

[34:01] working on on at least curbing and

[34:03] educating. Um so so who do you think

[34:07] should lead at this moment? We have the

[34:09] state, we have research, we have um uh

[34:12] the market

[34:14] whose voices are you paying can I say

[34:16] most attention to? I don't even know if

[34:18] that's fair question, but who should

[34:19] lead? I think the answer is all of the

[34:22] above. And I'd add a couple of other

[34:25] really key stakeholders. I think we

[34:26] really need to listen to educators and

[34:29] school administrators. We really need to

[34:31] listen to students themselves.

[34:33] Yay.

[34:33] And we really need to listen to parents

[34:35] that are going through this and trying

[34:37] to figure out how to adapt to new kinds

[34:40] of learning. Um and through the

[34:42] innovation council, we are working with

[34:44] some nonprofits to explore ways to um

[34:48] more sort of struct have a a better

[34:51] structure to feeding in those voices and

[34:53] particularly students voices into our

[34:56] efforts. And so hopefully more to come

[34:57] there shortly. But it's something that

[34:59] we are um very focused on to make sure

[35:02] that we're not just listening to um you

[35:05] know the the advocates or just listening

[35:07] to industry, but that it's really a a

[35:10] cohesive um effort that's grounded in

[35:12] expertise and in real world experience.

[35:15] Lots of students in the audience. Randy

[35:17] wants to hear from you. So don't mob her

[35:20] all at once at the end here. But um

[35:22] okay, we're going to start to wrap up

[35:23] here. what what guiding principle you

[35:26] think should anchor every AI policy

[35:28] decision? What's a what's a guiding

[35:30] principle that you're holding on to? I

[35:33] think it's just to reiterate the points

[35:35] I made earlier, which is one, we need a

[35:38] balance between opportunity and risk

[35:41] mitigation. It can't be all or nothing.

[35:45] And two, we have to learn from and

[35:49] engage with the students and educators

[35:51] themselves. And three, we have to be ev

[35:55] evidence-based and really understand

[35:57] what works and what doesn't. Uh, and so

[36:00] I think that is again why I'm really

[36:03] thrilled to be here today and look

[36:05] forward to hearing from you all. Um, and

[36:07] really do encourage you as said earlier

[36:10] that if there's either research that you

[36:12] want to share, if you have ideas of how

[36:14] policy makers can do better, please do

[36:17] not hesitate to reach out. We really do

[36:19] want to hear from you.

[36:22] What gives me hope is you clearly know

[36:24] what you're talking about. Uh you're

[36:26] using evidence. You're thinking this

[36:28] through. You don't seem to be um um sort

[36:33] of, you know, on this mission one way or

[36:36] the other. A lot of balance. One thing

[36:38] that you and I talked about before we

[36:40] came on stage is the governor's really

[36:42] knowledgeable

[36:43] in this area too. Uh and you said that

[36:46] makes your your job easy and hard,

[36:48] right? Um, so if you think about this, I

[36:51] do think about California as a leader.

[36:53] Um, how do you think about California as

[36:56] uh as a leader at this particular

[36:58] inflection point um uh around sort of

[37:01] this AI and and the future for learning?

[37:04] I mean, I think the proof is in the

[37:06] pudding that we really are leading the

[37:09] charge here, whether it's with the age

[37:13] signaling legislation that the governor

[37:17] signed last year that provided a a truly

[37:20] novel way of making sure that um parents

[37:26] have the opportunity to control the

[37:28] kinds of content that their kids see on

[37:30] their phones.

[37:32] to SB53 that provides the first in the

[37:36] country um guard rails on frontier AI

[37:40] systems to the AI and education working

[37:45] group that the state superintendent is

[37:47] leading that was based on legislation

[37:49] from last year where we're really

[37:51] soliciting input from um all

[37:54] stakeholders from industry from schools

[37:57] and educators from academic experts to

[38:01] issue um guidelines for schools, model

[38:05] policies to help local school districts

[38:09] uh be able to better both leverage the

[38:11] opportunities of AI again, which include

[38:14] everything from um being able to better

[38:18] address individualized learning

[38:20] disabilities to um making sure that we

[38:24] there's potential to bridge some of the

[38:27] digital divide and and sort of help

[38:30] support uh schools that maybe are

[38:32] underresourced to again being able to um

[38:36] support

[38:38] educators with some of the less um

[38:42] desirable parts of their job to things

[38:44] like performance-based

[38:46] grading um and evaluations. And so there

[38:50] really is a lot of opportunity here that

[38:52] we're excited to leverage in partnership

[38:55] with these companies that I mentioned as

[38:57] well. Um, and then we're also very

[39:00] focused on mitigating the potential

[39:02] risks that we've talked about. And so I

[39:04] think we're leading the charge on both

[39:06] sides of the coin. And uh, in in no

[39:10] small part because of institutions like

[39:12] Stanford that really the um I think the

[39:15] the thing that is the special sauce here

[39:18] in California is the are the

[39:21] partnerships that we have here. We have

[39:23] all the companies. We have the amazing

[39:26] world-renowned research institutions. I

[39:28] mean, this is the place where things are

[39:30] happening and we are so thrilled to be

[39:33] able to leverage those partnerships to

[39:36] um make sure that we continue to lead

[39:39] the country and the world in this space.

[39:42] I mean, I think that's why we're here

[39:43] today, too. So, um Randy, this has been

[39:46] a breath of fresh air to talk to a

[39:48] politician who knows what they're

[39:49] talking about, who looks at science, who

[39:51] wants to get lots of voices. uh in the

[39:54] room who wants to spend the day um

[39:56] listening and gathering input to make

[39:58] some really thoughtful uh legislation

[40:01] that both uh helps folks innovate and do

[40:05] what can be done with the new technology

[40:07] and yet also very carefully is looking

[40:09] at safety and and understand some of the

[40:11] other mitigating issues. So on behalf of

[40:13] everyone here, I just want to thank you

[40:15] so much. It's been lovely and we

[40:17] appreciate you and all that the governor

[40:19] is doing. Well, likewise. Thank you.

[40:21] Thank you for having me.

[40:35] Okay.

[40:37] All right. So, we will keep uh we'll

[40:41] keep pressing ahead here with our next

[40:42] panel which is titled scaling human-

[40:44] centered AI, what it takes to transform

[40:47] learning for all. Um, and with that, I'd

[40:50] like to introduce the moderator for this

[40:52] panel, Lewis Lebo, who is who leads K

[40:55] through2 research and development at the

[40:57] Gates Foundation. So, Lewis, over to

[40:59] you.

[41:03] Good morning.

[41:04] Morning. Morning.

[41:06] Come on up. Hi, everyone.

[41:09] Good morning.

[41:16] Uh, excited to be here. We've got a

[41:17] great panel on human- centered AI. I'm

[41:19] Lewis Lebo from the K12 team at the

[41:21] Gates Foundation. Just got in from

[41:23] Seattle last night. It was sunny there,

[41:26] by the way. Uh, and nowhere I would

[41:29] rather be than than with this amazing

[41:31] group of people. Uh, if I wasn't here, I

[41:34] would definitely not be with the uh

[41:36] million of my friends in downtown

[41:38] Seattle. Have a little parade this

[41:40] morning celebrating a victory. Sure what

[41:43] that's about. Um, but that's okay. It's

[41:45] okay. Um, I uh

[41:48] start the panel with guilt.

[41:49] I told I told my daughter I was going to

[41:51] make a couple Seahawks jokes and she was

[41:52] like, "That's a room of tech nerds. I

[41:54] would not suggest that." Um, but

[41:58] anyways, we have a fantastic panel here

[42:01] um this morning and as the speakers

[42:03] talked about this morning, we're really

[42:04] at a moment where AI and education has

[42:06] moved from possibility to reality. And

[42:10] the real question isn't whether AI is

[42:12] going to scale in learning, but how it

[42:13] scales, who it's designed for, and who

[42:17] actually benefits. And you know, this

[42:19] isn't the first wave of technology that

[42:21] we've seen with lots of promise in

[42:23] education. Uh there's been, you know,

[42:25] over the past couple decades, access has

[42:27] scaled dramatically. near universal

[42:29] internet in schools, widespread devices,

[42:32] digital tools that are embedded in daily

[42:34] instruction, but learning outcomes are

[42:38] still at early 2000 levels. And the

[42:41] trend lines are even worse for

[42:43] underserved students who need the

[42:44] support the most. Uh, now what's

[42:46] different this morning, of course,

[42:48] different with AI, of course, is the

[42:50] capability is amazing. The speed of

[42:52] adoption is incredible. Um, Patrick this

[42:55] morning referenced the rapid adoption.

[42:57] You all know tons about that. Um, of

[42:59] course, a lot of that's outside of

[43:00] school and without adult guidance, and

[43:03] that makes this moment promising, but

[43:05] also quite risky. On the promising side,

[43:08] we're seeing encouraging early evidence

[43:10] from AI tutoring studies, for example,

[43:13] uh, which is fitting to talk about here

[43:14] at Stanford, the the gold standard for

[43:17] evidence on human tutoring, which is

[43:19] still one of the most effective

[43:20] interventions we know in education, but

[43:22] it's expensive. It's hard to implement.

[43:25] Um just yesterday Susanna Loe who's out

[43:28] teaching at the moment but will be

[43:29] around later was talking to my team

[43:31] about um one of the recent studies about

[43:33] human tutoring and the results are

[43:36] complicated. It's it's challenging. Um

[43:39] but maybe AI tutoring can help with

[43:41] scale. Some of the recent trials we've

[43:43] seen have reported gains close to two to

[43:46] four months of additional learning. Uh

[43:48] but much of this is from early stage

[43:49] studies, limited samples and pretty

[43:52] short time horizons. So, it's a signal.

[43:55] We'll see if the panel agrees. Um, but

[43:57] not yet proof of durable system impact.

[44:00] And of course, we're seeing tons of work

[44:01] on how AI can support teachers as well,

[44:04] whether that's analyzing student work,

[44:05] surfacing misconceptions, providing

[44:08] instructional coaching, and some initial

[44:10] studies on this is really promising as

[44:12] well. So, tons of activity going on. Of

[44:15] course, I know we have an amazing group

[44:17] of people in this room who care deeply

[44:18] about all of this. We've got researchers

[44:20] and technologists and product developers

[44:23] and people doing the hardest job there

[44:24] is. The educators that are here and

[44:27] their students, which is fantastic. Um,

[44:29] I've been just so excited to see the

[44:31] many collaborations that I've learned

[44:32] about among those desperate, usually

[44:35] disperate groups of people. Um, and if

[44:39] this group uh this group here can can

[44:41] figure this out, but we have to stay

[44:43] focused. Um, we know that the benefits

[44:46] of the new technology will not

[44:47] automatically reach all students without

[44:50] intentional design and guard rails.

[44:52] Otherwise, AI could easily widen gaps.

[44:56] And so, that's why this session uh is

[44:58] focused on human- centered AI. It's not

[45:00] just a slogan, it's a real discipline

[45:02] and one that asks what evidence we

[45:04] require before scaling, how we design

[45:06] for real classrooms, and what guardrails

[45:09] are truly non-negotiable if these tools

[45:11] are going to serve all learners well.

[45:13] Um, so we've got a fantastic panel,

[45:15] Susan, James, Ian, and Amanda. Thanks in

[45:17] advance for sharing your wisdom. And I'm

[45:19] going to get started with, uh, one

[45:21] question for all of you, which is, what

[45:24] are some examples and themes of human-

[45:26] centered AI in education? It's a great

[45:29] term. Uh, tough to disagree that it

[45:31] sounds important, but what does it

[45:32] actually look like? What does it

[45:34] actually mean to you? Um, I'll start

[45:36] with Susan. Susan Ay. Um she's an econ

[45:40] economics of technology professor here

[45:41] at Stanford at the business school and

[45:43] is a leading scholar in causal inference

[45:45] and market design. Um and leads the Gol

[45:48] capital social impact lab which I hope

[45:50] we get to hear a lot about. Um and so

[45:52] set the stage for us. How do you think

[45:54] about this?

[45:55] Great. Well, so this is an amazing

[45:57] moment in time. We all know that. Um,

[46:00] one of the the things I've been doing

[46:03] over the last few years, beyond actually

[46:05] building and measuring AI education

[46:08] products directly, um, recently I've

[46:10] been the faculty adviser for the World

[46:12] Bank World Development Report for 2026

[46:14] on AI. So, I've been traveling all over

[46:16] the world trying to see, you know, what

[46:18] are people building, what are people

[46:20] adopting, and and what needs to happen

[46:23] to really realize our potential. And I

[46:26] think there's a few themes that have

[46:28] really come out for me. Um, first of

[46:30] all, like this is one reason this is an

[46:32] amazing moment is is not just that the

[46:35] AI inside the products are good, but

[46:38] this is a moment where we've kind of

[46:40] democratized the ability to create

[46:42] products. And so people who have ideas,

[46:46] even non-technical people who have ideas

[46:48] or people in developing countries who

[46:50] have ideas that pre previously didn't

[46:53] have access to engineering talent and so

[46:55] on. now can actually make their ideas a

[46:57] reality. So that's very exciting.

[47:01] But what that's already starting to look

[47:03] like on the ground and what it may look

[47:04] like even more in a year or two is

[47:07] actually now the bottleneck is no longer

[47:09] like engineers to build your ideas. The

[47:11] bottleneck is we have too many pilots

[47:14] actually and still not enough

[47:19] implementations that are actually

[47:21] effective. And so it's kind of obvious

[47:24] but I'll just say it like you know as an

[47:25] economist we think about scarce

[47:28] resources and so again the scarce

[47:30] resource was engineering but now there's

[47:32] going to be some other scarce resource

[47:35] some other bottleneck. A product has no

[47:37] value unless it's implemented and used.

[47:40] That's kind of obvious but we didn't

[47:42] focus on as much when we didn't have

[47:43] products. But if we have lots of

[47:45] products now the implemented and used is

[47:47] like everything. And so what did what

[47:50] does it take to make that happen? Well,

[47:53] that's where the human- centered AI

[47:55] comes in in the sense that we need

[47:57] educational science to make sure that

[47:59] the pilots are actually pilots of the

[48:01] useful things, but we also need the

[48:04] human- centered AI to understand the

[48:06] needs of the user, everything about it.

[48:08] What is the bandwidth constraints, the

[48:10] devices, the environment? If it's for

[48:12] teachers, what are their rubrics? What

[48:14] are their requirements? What is their

[48:17] context? And then it also needs

[48:20] economics. So what is the e why

[48:23] economics? Well,

[48:26] often if you have like a school district

[48:28] or a government adopting something, the

[48:31] history of governments and school

[48:32] districts adopting IT products is not

[48:35] great. There's a lot of waste. Um and

[48:38] and that so there's a procurement

[48:40] problem. There's an incentive problem.

[48:42] You adopt a software product, you get

[48:44] locked in and then it it's bad forever.

[48:46] And that's not just in education. Like

[48:47] I've asked people how many people are

[48:49] using Salesforce or something and

[48:50] they're stuck with it and they hate it.

[48:52] You know, you you get locked into

[48:54] software products because of that. It's

[48:55] terrifying to adopt them. You're going

[48:57] to take your teachers out of the

[48:58] classroom and train them on something.

[49:00] You're going to take your top government

[49:02] officials off of what they're doing to

[49:04] do this and there's just a limited

[49:05] amount of time. So that makes people

[49:08] scared to adopt things. So we need to do

[49:10] a really good job at thinking through

[49:12] that whole process and making sure that

[49:16] when you adopt you actually also have

[49:19] the ability and the incentive to achieve

[49:22] the potential. So that means we have to

[49:23] have a we have to be able to measure. We

[49:26] have to be able to improve these AI

[49:29] products are the most improvable things

[49:31] ever because changing software is as is

[49:34] as easy it has ever been. But then we

[49:36] need to think have frameworks also for

[49:38] thinking about what kinds of changes

[49:40] impose costs on the users. What can be

[49:42] done on the back end easily and how can

[49:45] we both um set up the the infrastructure

[49:49] and incentivize the providers of

[49:52] software to make them better achieving

[49:54] educational goals um for the users.

[49:58] That's great. Thanks. Um all right, I'm

[49:59] gonna thanks for setting the stage,

[50:00] Susan. I'm going to turn to James next

[50:03] um to to answer the same question in

[50:05] terms of examples and themes for human-

[50:07] centered AI uh right at the heart of

[50:09] your lot of work of your work. James is

[50:11] a professor of computer science here at

[50:13] Stanford also co-founder of the Stanford

[50:15] human computer interaction institute and

[50:17] is one of the leading voices in HCI and

[50:20] how design is applied to complex

[50:22] technologies including AI. Um so what do

[50:26] you think what do you think about what

[50:27] Susan said? What are you seeing in your

[50:28] work? Yeah. So, um, you know, Susan got

[50:31] out some of the issues here. I think one

[50:33] of the keys, uh, she mentioned is it's

[50:36] getting easier to build software, but

[50:39] understanding what to build is the

[50:41] problem. And to me, that's what this

[50:42] whole idea of human- centered AI is

[50:44] about. It's you can't just have the good

[50:47] intention to say, "Oh, I have this great

[50:49] idea." you actually have to go about

[50:51] designing it in a very different way

[50:53] than traditional software AI systems and

[50:56] especially in a a area like education

[51:00] more often have impact beyond the direct

[51:03] user of the tools. So we actually have

[51:05] to change how we think about design to

[51:07] go beyond what we would call user

[51:09] centered design which is you know taken

[51:12] uh maybe 30 years to uh become the

[51:15] standard in tech uh companies to think

[51:18] about how we evolve the user in the

[51:20] design process and we actually now need

[51:21] to say we need to go beyond user

[51:23] centered design to what I would call

[51:25] human- centered AI design which means we

[51:28] need to do user centered design but we

[51:30] also need to think about the broader

[51:31] community that is impacted by aa system.

[51:34] So you could think in an education

[51:36] context, I can't just design let's say

[51:38] for the teacher who might be using a

[51:39] system, I need to also understand how

[51:41] it's impacting the students, how it's

[51:43] impacting the families, how it's

[51:45] impacting other people in that

[51:47] community. Uh similarly, if we had a

[51:49] tutoring system for a a kid, we need to

[51:51] think about who those other stakeholders

[51:54] are and bring them into the design

[51:55] process. Also, we need to think about

[51:59] even higher level effects. If an AI

[52:01] system becomes very successful, let's

[52:03] say a large percentage of uh uh school

[52:06] districts were using a certain system,

[52:08] we start to see societal level effects.

[52:10] So think about the effects that AI in

[52:13] the news feed on Facebook or Instagram

[52:16] have had on society. Whether it's about

[52:18] what's true or false or whether it's

[52:20] young women uh feeling bad about their

[52:22] body image, those are society scale

[52:24] effects. So if we're really successful

[52:26] and do something that has that scale, we

[52:28] want to even think about what is the

[52:30] societal level potential effect of the

[52:32] thing I'm building and how do I bring

[52:34] that all into my design process. So it

[52:37] really requires a different way of

[52:38] thinking about design and it requires a

[52:41] different set of players on a design

[52:44] team. You can't we need technologists

[52:46] who understand AI. We need designers,

[52:48] but we actually need social scientists,

[52:50] humanists, and domain experts, whether

[52:53] it's education, finance, or law,

[52:55] depending on what you're building, to be

[52:56] in those teams from the start. You

[52:59] simply can't bring people in at the end,

[53:01] and say, "Hey, can you check whether

[53:03] this is safe or this is going to work?

[53:04] It's too late. That they need to be on

[53:06] the ground at the beginning, and that

[53:08] affects what we will build." So, human-

[53:10] centered AI is really a philosophy about

[53:12] how to actually create these systems to

[53:15] have positive impact for society. in

[53:18] areas like this in education.

[53:20] Great. Thanks. Your social media

[53:22] reference is interesting. I've been

[53:23] thinking about I mean it's it's quite a

[53:25] contrast to think about 70 80% whatever

[53:27] it is K12 students using chat GPT to

[53:30] help them with their homework lately.

[53:32] And at the same time 75% of elementary

[53:35] and middle schools banning cell phones

[53:38] in class. Amazing piece of technology

[53:40] but lots of harms, lots of distraction,

[53:43] lots of risks. tough to tough to sit

[53:46] with both of those at the same time and

[53:47] figure out what to what to do.

[53:49] Well, it's the early days and uh I think

[53:52] we need to uh give oursel a little time

[53:55] to try to get it right. Now, it is good

[53:56] that we figured out finally the negative

[54:00] effects of cell phones and social media

[54:02] and these things on our kids. Uh um and

[54:05] AI could have some of the same negative

[54:08] effects. Um, but we could also think

[54:10] about how to use AI to help all of us

[54:14] flourish in the long term. And so, how

[54:16] do we start to look at how to use this

[54:18] technology to have people reach life

[54:21] goals and behaviors, you know, that are

[54:23] long-term instead of these quick actions

[54:26] that the technology right now is really

[54:29] designed for because of the advertising

[54:32] related model. And I think there's a lot

[54:33] of interesting research on how AI

[54:35] actually can help move our kids towards

[54:38] the long-term goals, move all of us to

[54:40] uh be focused on being better in the

[54:43] ways that we all want to be. And so

[54:45] there's potential for it.

[54:47] Great. All right, Ian, I'm going to turn

[54:49] to you next. Ian is the managing is a

[54:51] managing partner at Owl Ventures, uh the

[54:54] largest education focused venture firm

[54:56] globally. Um, and you've backed uh many

[54:59] of the companies that are shaping how AI

[55:01] shows up in learning and hoping it all

[55:04] scales, I assume. Um, and so how do you

[55:07] approach this? What examples and themes

[55:08] are you seeing for human- centered AI?

[55:10] Yeah, first of all, I just want to thank

[55:11] Isabelle and team for hosting this

[55:13] wonderful event here. Um, it's just like

[55:16] a real full circle moment for me. uh

[55:18] having been on campus for you know seven

[55:20] years across uh undergrad and um you

[55:23] know grad school and now talking about

[55:25] education with you know three young kids

[55:27] I think is you know close to my heart

[55:28] you know every day and so you know I

[55:30] have um you know again as I mentioned

[55:32] sort of uh you know young children uh in

[55:34] public schools right now uh you know

[55:37] their teachers are grappling with uh you

[55:39] know with AI how to use it how to

[55:41] implement it I talk to my son you know

[55:42] every day about it you know what what's

[55:44] what's going on at school does your

[55:45] teacher you know use this you how do

[55:48] they how do they incorporate into the

[55:49] classroom and and what you're learning

[55:50] or do they just ignore it, you know? And

[55:52] so, you know, the the key thing for um

[55:55] you know, what we're seeing at Owl is

[55:56] really around, you know, ensuring that

[55:58] this AI is really kind of built around

[56:00] the core of the teacher. Um and in many

[56:03] ways uh you know we think about it as

[56:05] being a world-class almost you know TA

[56:08] and data analyst in a sense you know for

[56:10] the teacher to really augment and

[56:12] amplify the effectiveness of the teacher

[56:14] free up time uh and allowing them to do

[56:17] you know what they do best uh you know

[56:19] which is to inspire and to engage and to

[56:21] really make that human connection with

[56:23] the student but you know informing them

[56:25] and arming them with all of the data uh

[56:28] as well as some of the again real- time

[56:30] sort of feedback loops uh you you know,

[56:32] in a you know, very heterogeneous, uh,

[56:34] you know, classroom environment, right?

[56:36] And so, um, you know, we have, you know,

[56:38] 100 portfolio companies that span across

[56:40] K12 all the way through to workforce.

[56:42] And, you know, this is a slightly more,

[56:44] I think, K12 oriented, uh, audience.

[56:46] But, uh, I think the same applies to

[56:48] adults and adults learning as well is

[56:50] that, you know, the learner and the

[56:52] teacher, you know, at the center of it

[56:53] all is really, really critical. Um you

[56:56] know a couple of examples I think that

[56:58] uh I can kind of pull from uh you know

[57:00] we have a portfolio company called

[57:01] Newella uh you know which is used by has

[57:04] been used by over four million teachers

[57:05] and a big part of that is that you know

[57:07] the teacher remains in the driver's

[57:09] seat. The teacher you know has control

[57:11] over what content uh they want to use

[57:13] but the AI actually is helping with the

[57:15] scaffolding and the um you know

[57:17] appropriate sort of reading levels for

[57:19] the kids. And so that sort of

[57:20] differentiation at scale you know is

[57:23] helpful uh and it allows the teacher to

[57:25] do again what they do best which is to

[57:26] meet the students you know where they

[57:28] are. Um one other example is a company

[57:30] called Kdum uh you know which really

[57:33] kind of integrates uh instruction

[57:35] assessment uh and curriculum all

[57:37] together into one platform and really

[57:39] again helps the teacher sort of see the

[57:42] learning trends uh not only in terms of

[57:44] you know student by student but also you

[57:46] know across the classroom. think, you

[57:48] know, we've started to hear a little bit

[57:50] more about some of the, you know,

[57:52] multiplayer, if you will, sort of AI

[57:54] elements or how do you sort of get

[57:55] people into groupings and skill levels.

[57:58] Um, you know, KDUm, as an example, fits

[58:00] into district workflows, fits into

[58:03] curriculum and really sort of allows the

[58:05] teacher to in real time, uh, adjust uh,

[58:09] you know, effectively kind of the

[58:10] learning and teaching curriculum to map

[58:12] what the needs of the students and the

[58:13] misconceptions that are being faced, uh,

[58:15] you know, by the students in the

[58:16] classroom. So, um, a lot more examples,

[58:19] but I know, uh, we want to make sure

[58:20] everybody has a chance to, uh, to share

[58:23] thoughts.

[58:23] All right, great. Thanks, Ian. Um, all

[58:25] right, and we'll turn to Amanda. Amanda

[58:26] mentioned there's lots of educators in

[58:28] the room. You can tell me whether you

[58:29] agree that that, uh, they have the

[58:31] hardest job. I know you taught high

[58:32] school biology. Um, Amanda's the CEO and

[58:35] co-founder of AI for Education and work

[58:38] very hands-on with public school systems

[58:40] across the country as they figure out

[58:42] what does responsible AI adoption

[58:44] actually mean and look like. Um, so a

[58:47] bit where the rubber meets the road. Um,

[58:49] how does this all land with you and what

[58:50] what are you seeing in your work?

[58:52] Absolutely. Well, hi everyone. I'm

[58:53] excited to be here. This is a full

[58:55] circle moment for me as well. In January

[58:56] of or February 2024, this is the first

[59:00] big panel I ever did. um had five

[59:02] minutes to talk about AI and education

[59:04] and I watched it recently and not that

[59:06] much has changed which I don't know if

[59:07] that's really a very positive statement

[59:09] to start with but I think that um it's

[59:12] really important to think about human-

[59:13] centered AI with the question of you

[59:16] know we have an arrival technology

[59:17] that's as disruptive of any technology

[59:19] ever um and not only disruptive but

[59:22] faster in that disruption that is

[59:24] starting to really force us to look at

[59:26] what is human and not to get too uh

[59:28] philosophical to start but skills you

[59:31] talked about software engineering no

[59:32] longer being a barrier to ideiation,

[59:35] development, and creating products that

[59:37] we are looking at a technology that is

[59:40] now outstripping abilities across pretty

[59:43] much every like human skill. In fact,

[59:45] there's research that shows that it's

[59:47] not just that, you know, AI is

[59:48] augmenting or is uh going to automate

[59:51] out specific jobs within jobs. Tasks are

[59:54] being both at the same time. And what we

[59:57] have to really think about is just how

[59:59] much disruption that's going to cause

[01:00:00] and who is making the decisions. Um we

[01:00:03] talk about some of the technology that's

[01:00:04] being built for schools like how many of

[01:00:06] those organizations are really working

[01:00:08] handinhand not just with educators with

[01:00:11] researchers with students with

[01:00:12] psychologists. One of the most

[01:00:14] fascinating parts of this is that um you

[01:00:16] know almost half of all generative AI

[01:00:18] users are under 25. So I just want to

[01:00:21] say that again. And so if you look at

[01:00:22] 800 million active users for chatbt,

[01:00:24] they've said that 40% of their users are

[01:00:26] under 25. That's over 300 million active

[01:00:29] users a month. Um we also see a really

[01:00:32] kind of interesting moment of

[01:00:33] synchronicity is that the first ever AI

[01:00:35] chatbot was in the 60s and I'm sure some

[01:00:37] of you know Eliza. Eliza was a mental

[01:00:40] health chatbot and now we look, you

[01:00:42] know, three years into the time and you

[01:00:43] talk about, you know, the number of

[01:00:44] students using AI. They're more using AI

[01:00:47] for mental health and well-being than

[01:00:48] they are for school work. And so we have

[01:00:50] this really fascinating moment where

[01:00:52] it's starting to be this mirror or

[01:00:54] reflection to what people want, but

[01:00:57] maybe also what is missing, you know,

[01:01:00] and I think that when we see young

[01:01:02] people and adults turning to these uh

[01:01:04] technologies for much more than we ever

[01:01:06] thought, um it's going to really change

[01:01:09] what's necessary around supporting uh

[01:01:12] all of us. I mean, we see not just

[01:01:14] cognitive offloading that's happening.

[01:01:15] we see uh we have mental health and like

[01:01:18] well-being being offloaded. There is a

[01:01:20] new paper came out about belief

[01:01:22] offloading is that like what are we

[01:01:23] doing if these tools have been designed

[01:01:25] in a way in which they're actually

[01:01:26] shifting us into ways of thinking. Um

[01:01:29] especially at a really big risk

[01:01:31] considering that you know we have four

[01:01:32] or five chatbot makers that are having

[01:01:34] such an outsized effect on what we use.

[01:01:37] And so I think when I think of human-

[01:01:38] centered AI you know I think that this

[01:01:40] is an inflection point. Um this is the

[01:01:43] moment in which the choices we make

[01:01:44] today will have such outsized effects on

[01:01:47] not just education but the ways in which

[01:01:49] we interact with each other um and what

[01:01:51] we prioritize and what we believe has

[01:01:53] value. And so I think for us when we

[01:01:55] think about AI and education the most

[01:01:57] important thing right now is that we

[01:01:59] need to equip people with the real like

[01:02:01] knowledge and skills and mindsets to

[01:02:03] understand how to use these tools and to

[01:02:05] know when not to use them. I think this

[01:02:08] is a huge thing. We're going to have to

[01:02:09] make decisions in the future not to use

[01:02:11] AI. We're going to have to decide that

[01:02:13] even if AI can do this better than I

[01:02:15] can, it's important for me to do it

[01:02:16] myself. And I'm speaking a little bit to

[01:02:18] the young people in this room as well.

[01:02:19] That's going to be a really big question

[01:02:21] going forward. So, I think for us, we

[01:02:23] were really like pleased by the work

[01:02:25] that's being done, but we think we're at

[01:02:26] the very early stages.

[01:02:27] Great. Thanks. All right. I'm gonna I'm

[01:02:29] gonna give you some uh quick hit or easy

[01:02:32] questions, Susan. How do we measure all

[01:02:33] of this? Um, you mentioned you mentioned

[01:02:35] the many pilots. Um, you mentioned that

[01:02:39] there's many pilots designing there's

[01:02:41] now infinite variations of planes if I'm

[01:02:44] following the the metaphor correctly

[01:02:46] with the democratization of software

[01:02:48] development that you mentioned and and

[01:02:51] it's all moving so quickly. Uh, so what

[01:02:54] counts as credible evidence of impact?

[01:02:57] uh how do we learn from the different

[01:02:59] feedback loops that are necessary and

[01:03:02] and what do you what are you seeing in

[01:03:04] terms of translating from all the other

[01:03:05] fields that you've worked in and how

[01:03:07] that applies to education?

[01:03:10] Yeah. So I I mentioned in my opening

[01:03:12] remarks about incentives and it's very

[01:03:14] hard to provide incentives for things

[01:03:16] you can't measure. Um and also we know

[01:03:19] that if you measure one thing the

[01:03:21] teaching to the test is the metaphor we

[01:03:23] use across all of science in fact. And

[01:03:26] so here we are in education where you

[01:03:28] know we understand the problems of

[01:03:29] teaching to the test and people vendors

[01:03:32] may also build to the test. So this

[01:03:35] challenge is something I've been working

[01:03:37] on for 20 years. Um I did a st as chief

[01:03:39] economist at Microsoft where I focus a

[01:03:42] lot on the search engine and there um

[01:03:44] you know it was a very big challenge.

[01:03:48] The first thing I realized is you've got

[01:03:50] thousands of engineers that are all

[01:03:51] building things. They're building for

[01:03:53] what you measure. And if you measure the

[01:03:55] wrong thing, you get the wrong thing.

[01:03:57] So, it's easy to measure clicks. You can

[01:03:59] do an experiment in a day and see the

[01:04:01] clicks and make a decision tomorrow,

[01:04:03] which is great. But it leads to spammy

[01:04:07] ads and other stuff. And so you actually

[01:04:10] have a whole science team set up to

[01:04:13] figure out what to measure in the

[01:04:16] specific context and how to fight the

[01:04:19] teaching to the test which includes

[01:04:21] things like peer review and human review

[01:04:24] in addition to these like measurable

[01:04:28] you know easily measurable numbers. As

[01:04:31] education moves to be more in this

[01:04:34] digital product space you are going to

[01:04:36] want to make fast decisions. You do want

[01:04:37] your vendors to see that the teachers

[01:04:40] can't this interface isn't working and

[01:04:42] you want to be able to fix it tomorrow.

[01:04:44] So you do want to measure things fast

[01:04:45] and make decisions fast yet you have the

[01:04:49] same problems that you know how what

[01:04:52] does it mean if the teacher is spending

[01:04:53] more time in the interface. Is it

[01:04:55] because it's terrible and they takes

[01:04:57] them forever to get something done or is

[01:04:58] it because it's great and they're

[01:05:00] learning a lot from the experience. So

[01:05:02] that requires a multifaceted approach

[01:05:04] and I think you know the Gates

[01:05:05] Foundation has been a pioneer and in a

[01:05:08] lot of my speaking I'm actually looking

[01:05:09] at some of Gates Foundation and other in

[01:05:12] contexts like financial inclusion where

[01:05:15] they've identified digital public goods

[01:05:18] that can have an outsized impact on

[01:05:21] ecosystems and country growth. I think

[01:05:24] in this space there are also lots of

[01:05:27] digital public goods that be could be

[01:05:29] provided here by university and

[01:05:31] philanthropy. They are the tools for

[01:05:34] testing. They're the tools for

[01:05:36] evaluation.

[01:05:38] That's it's it's multifaceted. It's

[01:05:40] going to be human review tools. It's

[01:05:42] going to be AI guinea pigs so that you

[01:05:45] don't have to make humans your guinea

[01:05:47] pigs. But actually making a good AI

[01:05:50] guinea pig is itself a science because

[01:05:53] if you have a fake student testing your

[01:05:54] product but it does dumb things that's

[01:05:56] going to lead your product in the wrong

[01:05:57] direction. So you have to evaluate your

[01:06:01] fake guinea pigs. So that's a little bit

[01:06:04] meta right so I see actually but this is

[01:06:08] not just like a small problem or

[01:06:10] specific to education. I actually see

[01:06:12] this as like the foundation of what it's

[01:06:15] going to take to make good AI products

[01:06:18] across the economy is investing in the

[01:06:21] tools that help you understand whether

[01:06:25] the AI is doing what you want it to do.

[01:06:27] And by the way, even inside Open AI,

[01:06:29] inside anthropical, that's a that's a

[01:06:32] huge huge scientific focus is just how

[01:06:35] do you know whether you've done the

[01:06:37] right thing for your user? So, I'm

[01:06:41] really excited about that science that's

[01:06:42] been a theme at Stamford High, but it

[01:06:45] but in the specific education context,

[01:06:49] it's like even more important because,

[01:06:52] you know, using students as guinea pigs

[01:06:54] is like or teachers as guinea pigs is is

[01:06:56] a, you know, we don't want to make

[01:06:58] mistakes. We want to have a really

[01:07:01] robust measurement, testing, evaluation

[01:07:03] pipeline. And then just coming all the

[01:07:05] way back to the incentive point. If we

[01:07:07] can figure out these multifaceted ways

[01:07:09] to measure, then we can do a much better

[01:07:12] job incentivizing the vendors to

[01:07:15] experiment, measure and improve.

[01:07:18] Great. Uh we are working on some of that

[01:07:21] evaluation stuff at the on our team. We

[01:07:23] usually say that the the health teams

[01:07:25] all those other they got it so they got

[01:07:27] much easier to measure over there.

[01:07:28] Education's very much more complicated

[01:07:30] to measure. Um uh great. Thank you. Let

[01:07:34] me I've got one more question for each

[01:07:35] of you and then and then we'll head to

[01:07:36] the audience if we have time for for

[01:07:38] probably just one or two questions. Ian,

[01:07:40] let me go to you since you also are

[01:07:42] thinking about measurement as well but

[01:07:44] from a bit of a different lens. Uh

[01:07:46] you're thinking about obviously you care

[01:07:48] about impact, you also care about scale,

[01:07:49] you care about financial returns. When

[01:07:51] you're when you're look I mean you

[01:07:52] mentioned Nuzella and Kdum. They're

[01:07:54] relatively scaled, I'd say, in a way.

[01:07:57] But like if you think about some of the

[01:07:58] earlier stage organizations, what are

[01:08:00] some of the signals you're looking for

[01:08:02] to try to figure out, hey, is there the

[01:08:05] potential for impact and scale? I'm sure

[01:08:07] we've got some early stage organizations

[01:08:09] in the room. Ian will be right over

[01:08:11] there afterwards if you want to talk to

[01:08:12] him. What um you know, how do you think

[01:08:14] about that?

[01:08:15] Yeah. Well, good news is all these uh

[01:08:18] companies like Newella and KDM at one

[01:08:20] point were early stage as well. So,

[01:08:21] they've been through that journey. Um

[01:08:23] and I think you know a good example

[01:08:25] another one just to kind of share with

[01:08:27] you um is a company called Amir Learning

[01:08:29] uh you know which does oral fluency and

[01:08:31] reading and literacy um and you know

[01:08:34] they uh had the opportunity to be sort

[01:08:36] of grounded in in research. So they had

[01:08:37] work that came out of Carnegie Melon uh

[01:08:39] you know they started to do uh you know

[01:08:42] essentially studies uh and independent

[01:08:44] evaluations in concert with state

[01:08:46] agencies uh education agencies as well.

[01:08:48] And so they did that across um you know

[01:08:50] Louisiana, across Utah, across Georgia.

[01:08:53] Um and you know in those ways this sort

[01:08:57] of collaboration I think is really

[01:08:59] important. So it's not simply about we

[01:09:02] we do see a lot of companies where um

[01:09:04] you know the first sort of uh points of

[01:09:07] data all are around usage and engagement

[01:09:09] and you know that's sort of 80% of the

[01:09:11] you know conversation but uh you know

[01:09:13] oftentimes there's a huge gap you know

[01:09:16] between usage and engagement and

[01:09:17] outcomes uh and so the ability to sort

[01:09:20] of collaborate with third party groups

[01:09:21] uh I think is really really important uh

[01:09:24] you know kdum right now is also doing

[01:09:26] the same thing in New York public

[01:09:27] schools so they had you know over 400

[01:09:29] teachers and educators uh you know

[01:09:31] implementing KDUm and and triing it with

[01:09:33] uh you know their students and you know

[01:09:35] they've put together a body of work now

[01:09:37] where essentially students in classrooms

[01:09:40] that have you know teachers that are

[01:09:42] using KDUm you know 30 or plus times

[01:09:44] have double digit gains uh and you know

[01:09:47] that sort of evidence driven sort of um

[01:09:51] uh fact pattern is really important

[01:09:53] before you make claims uh and I think a

[01:09:55] lot of companies right now especially

[01:09:56] early on struggle with that dynamic of,

[01:09:59] you know, how do I get that traction and

[01:10:01] how do I get the attention, you know, to

[01:10:03] make these claims but then not yet have

[01:10:04] the evidence. And so I think, you know,

[01:10:06] finding the right uh, you know, not only

[01:10:08] the pilots but then the right districts

[01:10:10] and, you know, ensuring that you're in

[01:10:11] that in that workflow and then working

[01:10:12] in collaboration with the right third

[01:10:14] party groups is is just critically

[01:10:16] important to to establishing that

[01:10:18] credibility and trust.

[01:10:19] Great. Great. Um, James, let's let's

[01:10:22] talk about teachers for a moment. Um,

[01:10:25] we've heard already a few times this

[01:10:26] morning, we want to make sure AI

[01:10:28] augments teachers. We we're not

[01:10:29] replacing teachers. We want to make sure

[01:10:30] we're supporting them. We're augmenting

[01:10:32] them. Great. Uh, makes sense. How do we

[01:10:35] actually do that from a design

[01:10:36] perspective to focus on that

[01:10:37] specifically?

[01:10:39] Well, um, I think it's also related to

[01:10:43] the question about evaluation. So, you

[01:10:45] know, there's a field of learning

[01:10:47] sciences. There's a lot of great folks

[01:10:49] in that field right here at Stanford

[01:10:50] from our dean Dan Schwarz on down. And

[01:10:53] so we need to make sure that we're

[01:10:55] applying those techniques to do this,

[01:10:56] not just ad hoc ways of measuring

[01:10:58] things. This is a field that with a lot

[01:11:00] of rich research on how to do it. Um,

[01:11:03] but I would say one thing supporting

[01:11:05] teachers or measuring things. What we

[01:11:08] really actually need to watch out for is

[01:11:10] uh almost to paraphrase Wayne Gretzky,

[01:11:12] we need to go where the puck's going to

[01:11:14] be, not where it is now. We can't be

[01:11:16] measuring the stuff that mattered for

[01:11:19] schools in the last century. This

[01:11:22] technology is going to change what we

[01:11:24] think kids need to know whether it's in

[01:11:27] K12 or in college or in lifelong. And we

[01:11:29] need to start rethinking what it is are

[01:11:32] the key skills. Now, some of them are

[01:11:33] going to go back to the classic skills

[01:11:34] that we might all think about like

[01:11:36] critical thinking, how to make an

[01:11:38] argument, the basics, but it's going to

[01:11:40] go away from a lot of uh testing that is

[01:11:44] really how good you are at regurgitating

[01:11:46] information. And we have to first think

[01:11:48] about what is going to change because a

[01:11:52] lot's going to have to change. And so

[01:11:54] once we know that then we understand

[01:11:55] what it is we need to measure properly

[01:11:57] and use the learning sciences to do

[01:12:00] that. And similarly, how do I augment

[01:12:02] teachers? Well, I got to know what it is

[01:12:04] that's important to teach and important

[01:12:06] for kids to learn because now if I have

[01:12:08] a smart great tutor for kids, maybe some

[01:12:11] of these other skills aren't where they

[01:12:13] need help. Maybe it's more on the social

[01:12:15] skills and the other parts of life that

[01:12:17] teachers are going to maybe focus more

[01:12:19] on. But we have to figure that question

[01:12:20] out first before jumping to a bunch of

[01:12:23] solutions that in some ways are teaching

[01:12:26] kids for the industrial revolution.

[01:12:29] Great. Um Amanda, let me finish with

[01:12:32] you. Could we talk about the beyond the

[01:12:34] AI inflection point? You you uh paper

[01:12:37] that you released recently. Um and would

[01:12:40] love to hear like a little bit more

[01:12:41] about it. But in that in that paper you

[01:12:43] talked about, you know, a fictional

[01:12:45] school district uh central schools I

[01:12:47] think it was called that if um fictional

[01:12:51] but maybe totally fictional in the sense

[01:12:54] that it was grounded in deep research

[01:12:55] that you did and thought about the three

[01:12:57] potential paths over the next five years

[01:12:59] that could happen um depending on the

[01:13:02] types of choices that those districts

[01:13:04] make. So tell us a little bit about what

[01:13:06] you learned from that research, but in

[01:13:08] particular, especially to close us out,

[01:13:10] what about that pathway where a lot the

[01:13:12] guardrails are put in place? Like what

[01:13:13] is that? What does that mean? What does

[01:13:15] that look like? What are the most

[01:13:16] important types of guardrails and safety

[01:13:19] considerations that everyone in the room

[01:13:21] needs to be thinking about right now?

[01:13:22] Absolutely. Well, I want to call out

[01:13:23] Jason from Imagine Learning. This was a

[01:13:25] gorilla project and Isabelle How and

[01:13:27] other we had Chris Agn from Stanford. We

[01:13:30] brought a group of almost 20 people over

[01:13:32] the summer. Uh, has anybody read AI

[01:13:34] 2027? It's this futurist piece about AGI

[01:13:37] or super intelligence being possible.

[01:13:39] And we were actually sitting on the

[01:13:40] floor of the AI show at ASUGSV and

[01:13:43] looking around uh all the point

[01:13:45] solutions that were measuring for today,

[01:13:47] not for tomorrow. To your point, James,

[01:13:49] and got a bit frustrated and I was like,

[01:13:52] well, what if we do an AI 2027 for

[01:13:54] education, which sounds like not that

[01:13:56] hard, guys. It was really hard. Uh we

[01:13:59] wrote a fictional story that has a

[01:14:01] choose your own adventure approach. Um

[01:14:03] which uh none of us are fiction writers.

[01:14:04] So that was fun. But what we did is we

[01:14:06] brought together a really crossunctional

[01:14:08] group including students and came up

[01:14:10] with central school district and it was

[01:14:12] I'm glad that you felt like it felt

[01:14:13] real. We've we've worked with hundreds

[01:14:16] of districts. And so from 2023 to 2025

[01:14:19] it should feel as real as possible. The

[01:14:21] complexity of even doing the good work

[01:14:23] of setting guardrails starting to think

[01:14:25] about AI literacy is just not enough.

[01:14:28] And so the big kind of um like piece of

[01:14:30] the the central district approach is

[01:14:32] starting a lab school in 2026 that

[01:14:34] becomes a skunk works for calculated

[01:14:36] risk takingaking for measurement

[01:14:37] evaluation for trying to create the like

[01:14:40] a future where we're not just creating

[01:14:42] you know kids for a future that used to

[01:14:45] exist or one we think will exist. And

[01:14:48] what we do is we branch in 2027 where

[01:14:50] things get really hard because anybody

[01:14:52] you know education is really good at not

[01:14:55] changing like we're just really really

[01:14:57] stubborn. Um and so it becomes this kind

[01:15:00] of choice and we have three paths that

[01:15:01] are followed. One is this kind of uh

[01:15:04] re-entrenchment of traditional methods.

[01:15:06] We go back to a place in which we're pen

[01:15:08] and paper which is happening right now.

[01:15:10] And what happens is young people kind of

[01:15:12] come out not knowing either not being AI

[01:15:14] literate and not also being like deep

[01:15:16] critical thinkers because they're using

[01:15:17] AI at home and hidden in in ways that

[01:15:20] are not very helpful. They're definitely

[01:15:21] not prepared to go get a job with uh

[01:15:23] with Ian. Uh so that's path A. Path B is

[01:15:26] focused on this this kind of algorithmic

[01:15:28] all-in. We're going to pick an AI

[01:15:30] system. It's going to it's going to do

[01:15:31] everything not just help students learn,

[01:15:33] but also it's going to surveil students,

[01:15:35] keep us safe. And what ends up happening

[01:15:37] is you have young people that are really

[01:15:39] like it's stripping out the human, you

[01:15:41] know, we of that kind of interaction of

[01:15:42] the human interaction and the durable

[01:15:44] skills that we want to see. And then

[01:15:46] pass C is kind of the the messy middle

[01:15:48] where the the lab school keeps going and

[01:15:50] kids are more prepared to be agile than

[01:15:53] they are necessarily prepared to be an

[01:15:55] engineer or a technologist or an

[01:15:58] economist. They're they're going to

[01:16:00] learn the skills to keep learning

[01:16:02] because it's never going to slow down

[01:16:04] from this moment on. And one of the

[01:16:05] things that I think is really important

[01:16:06] is that the guard rules are in place for

[01:16:08] all three of those, but the decisions of

[01:16:11] like how how much are we willing to look

[01:16:13] at our like what works and doesn't work

[01:16:15] and and make the change is really the

[01:16:18] more important piece. But I would say

[01:16:19] very much is that I just want to the

[01:16:21] last thing I'll say is that a core

[01:16:23] component of everything if there is one

[01:16:25] place that we should be thinking about

[01:16:26] in terms of guard rails, it's not

[01:16:28] responsible guidelines. It is AI

[01:16:29] literacy for young people. And it is

[01:16:31] technical AI literacy. It's

[01:16:33] understanding what these tools are and

[01:16:34] aren't, but it's moving from AI literacy

[01:16:36] to fluency. And if we do not do that in

[01:16:38] the next two to three years, starting

[01:16:40] that out, we are in a world of hurt. And

[01:16:42] so that to us is like the kind of the

[01:16:45] underlying for everything. We have to

[01:16:46] equip young people for the future that's

[01:16:48] coming.

[01:16:48] That's great. Thank you. There's a

[01:16:50] there's a red clock flashing and staring

[01:16:52] at me. And so we'll defer on audience

[01:16:54] questions, but this is a small enough

[01:16:56] group. Please come find us throughout

[01:16:57] the day. We'd love to hear your

[01:16:58] reactions to this. Final words,

[01:17:01] 10 seconds or less. What's one thing you

[01:17:04] would love for this audience to do this

[01:17:06] year to make sure that human centered AI

[01:17:10] becomes the norm?

[01:17:13] James, you sir.

[01:17:14] I'm just going to say this is one of the

[01:17:17] most important areas for AI. So we at

[01:17:21] the Stanford Institute for Human Center

[01:17:23] AI as we plan our next seven years, this

[01:17:26] is one of the three key areas that we

[01:17:28] are focused on with our friends at

[01:17:30] Stanford and other folks to really push

[01:17:33] forward because this is really one of

[01:17:34] the most important areas for AI and the

[01:17:36] future of our world.

[01:17:37] Awesome others.

[01:17:39] Yeah, I mean I'll just take a more

[01:17:40] personal uh stance at that question

[01:17:42] which is get you know be be involved and

[01:17:44] and be sort of integrated into the

[01:17:47] community. Um, you know, I think this is

[01:17:48] a very select group here of course, but

[01:17:50] uh it's obviously a much bigger world

[01:17:52] and just making sure each of us sort of

[01:17:54] touches four or five educators, touches

[01:17:56] students and really just involves

[01:17:59] everybody because I think this is

[01:18:00] certainly going to be um you know a

[01:18:01] group effort. Uh to your point,

[01:18:04] yeah, I would say speak up like this is

[01:18:06] the time to be loud like this is time to

[01:18:08] if you have you know Nella or others are

[01:18:10] coming in saying we want to do this say

[01:18:12] why or how. um work with each other. If

[01:18:16] you're a young person, advocate for AI

[01:18:17] literacy in your organizations. If

[01:18:19] you're a teacher, like talk to young

[01:18:21] people. I think that this is like

[01:18:23] everyone has a voice today if we're

[01:18:24] willing to do it. Um it's going to be

[01:18:26] like harder as we move forward. So, I

[01:18:28] just think that this is a time to do it.

[01:18:31] I would say that the the bottlenecks now

[01:18:34] are sort of non-technical

[01:18:37] in some ways. And so everybody who has

[01:18:40] an idea, who has domain knowledge, who

[01:18:43] knows about a problem has more kind of

[01:18:46] power and opportunity to

[01:18:49] do something that actually matters than

[01:18:51] almost any other time. Like if I have an

[01:18:53] idea that requires, you know, hundred

[01:18:55] billion dollars to build, well, like I

[01:18:58] can h it doesn't really matter whether I

[01:19:01] have the idea unless I have hundred

[01:19:03] billion dollars. But at this moment,

[01:19:05] there are so many ways that each

[01:19:07] individual human can actually take their

[01:19:09] ideas and help other humans. Um, and and

[01:19:12] so I'm I just this is a great moment to

[01:19:15] be alive, but it's there's no guarantee

[01:19:17] that we'll actually take it across the

[01:19:19] finish line. So that's really what I'm

[01:19:20] focused on.

[01:19:21] All right. Thanks for a great

[01:19:22] conversation. Let's thank our panelists.

[01:19:52] So before everyone takes a little

[01:19:54] stretch here, we are um we are about to

[01:19:58] uh start our next session. Um, and our

[01:20:02] next session is one that you want to

[01:20:05] stay for.

[01:20:08] It's one that has actually a lot of

[01:20:12] tension.

[01:20:14] Does AI make us more creative or not?

[01:20:19] And as it turns out, we have actually

[01:20:21] researchers uh that are showing both

[01:20:25] sides. and we are going to hear about

[01:20:29] both sides. So it's not a debate

[01:20:32] uh but it is uh very close to it. Um and

[01:20:36] I'm so excited to introduce a moderator

[01:20:39] for the session who has spent

[01:20:42] extraordinary amount of time thinking

[01:20:44] about this topic about creativity.

[01:20:48] Uh he leads um a big effort at Adobe.

[01:20:52] Adobe as a as a corporations has a big

[01:20:55] focus on creativity and is a man who is

[01:20:58] leading this for creativity and

[01:21:00] education. He's also a Stanford grad. So

[01:21:04] Brian welcome and thank you.

[01:21:14] Good morning everyone.

[01:21:17] I think it's safe to say creativity is

[01:21:19] having a moment right now. I've been in

[01:21:22] the space of creativity and education

[01:21:24] for quite a while and five years ago if

[01:21:27] we had created a top 10 list of what are

[01:21:29] the most burning urgent issues in

[01:21:32] education besides niche circles I don't

[01:21:35] think creativity would have been on that

[01:21:36] list but boy has that changed the

[01:21:39] questions what is creativity how do we

[01:21:42] teach it and why does it matter in every

[01:21:45] subject and discipline are echoing

[01:21:48] through the halls of schools and

[01:21:49] campuses louder than ever before. And I

[01:21:52] think our little friend generative AI is

[01:21:55] a large reason that is happening. But

[01:21:57] even outside of formal education, at a

[01:21:59] broader cultural level, as soon as Chat

[01:22:02] GBT came on the market, fall 2023, we

[01:22:05] all heard this sentence starter. But AI

[01:22:08] will never be able to do. And one of the

[01:22:11] key words that was rushed to finish that

[01:22:14] sentence was be creative. Do creativity.

[01:22:17] So it's an exciting time. It's an

[01:22:19] important time and I'm really excited to

[01:22:22] introduce four educators and researchers

[01:22:26] who from different disciplines and

[01:22:28] different areas are going to help us

[01:22:30] grapple with this big question. So I

[01:22:33] will introduce them one at a time. They

[01:22:35] have a hard task. They have five minutes

[01:22:37] to give you an overview of everything

[01:22:39] they're doing in this space. We'll bring

[01:22:41] them up one at a time and then we'll all

[01:22:44] uh in front of you have a discussion,

[01:22:47] maybe even a debate afterwards. So

[01:22:49] without further ado, um from the Center

[01:22:51] for Computer Research and Music and

[01:22:53] Acoustics, uh please join me in

[01:22:56] welcoming Chris Chafe.

[01:23:08] Okay. Well, thank you for taking a few

[01:23:11] minutes to listen to my everchanging

[01:23:14] life in presto tempo right now. It's a

[01:23:18] it's an amazing moment. Um, and I come

[01:23:23] at this from computer music and having

[01:23:25] worked as an artist but also a teacher

[01:23:28] for, you know, quite a long time.

[01:23:32] Most of where we end up now is as you

[01:23:35] know sort of sophisticated musicians

[01:23:37] using computers is often with shrink

[01:23:39] wrap software. So it's software that's

[01:23:41] been developed very special purpose

[01:23:43] maybe uh um we teach that but often at

[01:23:47] at at a lot of institutions especially

[01:23:49] Stanford we teach the you know how to

[01:23:52] make that stuff. And I I have a

[01:23:55] statistic that I just kind of mowled

[01:23:58] over recently, which is back when I

[01:24:00] started, the industry spec was five

[01:24:03] lines of code per day per programmer,

[01:24:08] debugged, tested, documented, and then,

[01:24:12] you know, kind of, you know, this is a

[01:24:14] 50-year span. Halfway through it, I'm up

[01:24:15] to about 50 lines of code. Last summer,

[01:24:18] 500. It just went, you know, exponential

[01:24:21] like that. and five or 10 lines, 500 or

[01:24:25] a thousand lines, doesn't matter. It's

[01:24:27] exponential. So, so we're in this place

[01:24:29] and I'd like to um show you how that

[01:24:31] translates and what we can do as an

[01:24:32] ensemble in music. So, go ahead and hit

[01:24:35] the uh clip that I'm providing you. So

[01:24:38] this is network music performance

[01:24:41] co-design ensemble and this performance

[01:24:43] from 10 days ago in the Bing auditorium

[01:24:46] with eight students

[01:24:49] and they're playing instruments and you

[01:24:51] can bring the volume up a bit. They're

[01:24:53] playing uh instruments and laptop

[01:24:56] instruments.

[01:25:35] So, in this case, everyone's sharing an

[01:25:37] instrument that we passed around in the

[01:25:38] ensemble in rehearsal, and it's part of

[01:25:41] a texture that we're working up. It's a

[01:25:43] longer piece. Um, I asked a class in

[01:25:47] September to do this

[01:25:49] just to find out if this works at all.

[01:25:52] That's from a prompt. 31 students did

[01:25:54] the same prompt. And here is the result.

[01:25:57] It's actually

[01:25:59] 31 usable web apps that run in browsers.

[01:26:03] This is a class they're making

[01:26:05] hydrophones and learning how to solder.

[01:26:08] And the next thing I was going to do in

[01:26:09] the class after this was um sort you

[01:26:11] know introducing um perceptual

[01:26:13] phenomena. So the Gestalt principle of

[01:26:17] common fate and this is around

[01:26:19] Halloween. So forgive me for what I this

[01:26:22] is a more complex program that I did.

[01:26:25] And all of the dots are individually

[01:26:29] programmed in the web app. And common

[01:26:33] fate is when the dots move together and

[01:26:35] you can pull out an identity. So that

[01:26:38] was quiet. Let's do it with the THX logo

[01:26:41] theme, which you may remember from

[01:26:44] theaters.

[01:26:55] So, scary stuff, Halloween compatible.

[01:26:59] And then uh try try something even

[01:27:02] scarier.

[01:27:12] Sort of like a facial therman is the

[01:27:15] sort of air instrument from from years

[01:27:17] and years ago. So these are all you know

[01:27:19] like like kind of developing technical

[01:27:22] things that are really tough to program.

[01:27:25] Um you know would take that 50 lines of

[01:27:27] code and it would take 10 days and so

[01:27:29] on. Um but what about doing

[01:27:32] new realms you know like imaginary

[01:27:34] things? Ask me later what spider

[01:27:37] ballooning is but we put this one prompt

[01:27:41] and developed a quartet

[01:27:44] that we're about to rehearse tomorrow.

[01:28:30] Spider loom ballooning is the scariest

[01:28:33] thing that happened to me in September,

[01:28:35] but I'll leave it there. And this this

[01:28:37] catches this totally captures it. A

[01:28:39] oneline

[01:28:40] prompt that we could modify and do

[01:28:43] something else. Anyway, thanks for

[01:28:44] listening and on to the next speaker or

[01:28:46] so. Yeah.

[01:28:52] All right. Our next panelist um is

[01:28:55] actually near and dear to my heart

[01:28:57] because I did my PhD here about 15 years

[01:29:00] ago in modern thought and literature,

[01:29:01] which is sort of yeah, choose your own

[01:29:04] adventure. Um but I was you know a

[01:29:06] humanities and education guy and then I

[01:29:08] took an intro to computer science course

[01:29:11] with Marin opened my mind to

[01:29:14] computational thinking and honestly

[01:29:16] changed uh my career trajectory because

[01:29:18] now here I am in tech. Um so please join

[01:29:20] me in welcoming from the department of

[01:29:22] engineering Tommy.

[01:29:24] Thanks very much and you're really too

[01:29:27] kind. The checks in the mail. Um, what I

[01:29:30] want to do is spend a little bit of time

[01:29:31] today talking to you about three

[01:29:33] provocations on creativity and

[01:29:34] generative AI. For about the last eight

[01:29:36] or nine years, I've been co-eing with a

[01:29:38] philosopher. And so, I've moved from the

[01:29:39] world of thinking about problem and

[01:29:40] solution to provocation. Uh, so first

[01:29:43] provocation or before we get there, what

[01:29:44] do we even talk about when we think

[01:29:46] about creativity? That's the kind of the

[01:29:47] computer scientist coming out, right? We

[01:29:48] need to define it. I like to think of it

[01:29:50] as novel approaches for problem solving.

[01:29:52] Being able to come to some new problem,

[01:29:54] bringing fresh ideas to that problem to

[01:29:56] be able to generate a solution. There's

[01:29:58] lots of different definitions we can

[01:29:59] think of. This is just kind of our

[01:30:00] working one for the next five minutes.

[01:30:02] So, here's provocation number one.

[01:30:04] Thinking about education as a product

[01:30:06] versus process. And this is something I

[01:30:08] probably don't need to tell most of the

[01:30:09] people in this room, but I'm going to

[01:30:10] tell you anyway because it's in the

[01:30:11] slides. What do we think of when we

[01:30:14] think of education? Education obviously

[01:30:16] is about learning. That's the goal. And

[01:30:18] learning is a process, right? There's a

[01:30:20] whole bunch of things like research and

[01:30:21] writing and problem solving, all these

[01:30:22] things that students do as part of the

[01:30:24] learning process. But what we do is we

[01:30:27] evaluate products. We ask students to

[01:30:29] turn in papers or a program or solutions

[01:30:31] to a problem set or whatever the case

[01:30:32] may be. And for many years basically our

[01:30:35] conceit in education is that a strong

[01:30:37] product that we assess is indicative of

[01:30:39] a strong process that the students went

[01:30:41] through to generate that product. And

[01:30:43] what generative AI has basically done is

[01:30:44] taken that assumption that we've had I

[01:30:46] don't know for a couple hundred years in

[01:30:47] education and just fundamentally broken

[01:30:49] it and said basically you can generate a

[01:30:51] strong product without engaging in a

[01:30:53] strong learning process. So what does

[01:30:55] this mean for us? Right? What it means

[01:30:57] for us is that we need to think about

[01:30:59] ways in which learning engages more with

[01:31:01] the actual process rather than just

[01:31:04] asking students to produce a product and

[01:31:06] expect that the product actually meant

[01:31:07] that a good process happened. Part of

[01:31:10] that is also realizing that even experts

[01:31:12] in technology can be lulled into a false

[01:31:14] sense of security by these processes or

[01:31:17] these algorithms that exist to basically

[01:31:19] shortcircuit processes. I'll give you

[01:31:21] one example that comes from my domain.

[01:31:22] Uh Dan Benet and some of his students.

[01:31:24] He's a computer network security

[01:31:25] researcher in the computer science

[01:31:27] department. Did this experiment a few

[01:31:28] years ago took a group of people, gave

[01:31:30] him generative AI tools and said, "Build

[01:31:32] some stuff that has to do with computer

[01:31:34] security, like a login authenticator,

[01:31:36] stuff like that." Another group of

[01:31:37] people, these are trained programmers,

[01:31:39] right? Had to do the same thing, but

[01:31:41] without generative AI tools. And then

[01:31:43] they checked how secure was the code and

[01:31:45] how strongly did they believe their code

[01:31:46] was secure. And here's the interesting

[01:31:48] thing. The folks that had access to

[01:31:49] generative AI believed that their code

[01:31:52] was more insecure. And when it was

[01:31:54] actually tested and reviewed, it was

[01:31:55] less secure. Right? So it gives a false

[01:31:58] sense of what we are actually producing

[01:32:01] with these technologies. And relative to

[01:32:02] the last panel, if we think about one of

[01:32:04] the critical factors, the critical

[01:32:06] factors is to have people be critical of

[01:32:09] AI, right? To think about how their

[01:32:11] learning process is actually impacted by

[01:32:14] the use of these tools. So that gets me

[01:32:16] to the second propagation, topics versus

[01:32:18] tools. Generally, when we think about

[01:32:20] generative AI in a curriculum, it

[01:32:22] requires a curricular organization. And

[01:32:23] you're like, Maron, that's tautological,

[01:32:25] right? You just told us it's part of a

[01:32:26] curriculum, so of course it needs a

[01:32:27] curricular organization. Yeah, seems

[01:32:29] obvious, doesn't it? Except we don't

[01:32:30] treat it that way. What we do is we

[01:32:32] think about a curriculum being organized

[01:32:34] by topics that a student should know.

[01:32:36] But we don't think of generative AI as a

[01:32:37] topic. We treat generative AI as a tool.

[01:32:40] And when we think about tools, the

[01:32:41] decisions around tools are often made by

[01:32:43] individual instructors and individual

[01:32:44] classes. and there isn't a building of a

[01:32:47] topic in the same way we might think

[01:32:48] about the ways you go through for

[01:32:50] example learning math building on

[01:32:52] previous classes and so I think of that

[01:32:54] as our challenge is to generative AI not

[01:32:56] to consider it as a tool but consider it

[01:32:57] as a topic and build a curriculum in

[01:33:00] which we teach how to use generative AI

[01:33:02] to foster things like creativity and

[01:33:04] deeper educational outcomes by actually

[01:33:07] treating it as a topic where for example

[01:33:08] here's just a stylized illustrative

[01:33:10] example so you get a sense of what I'm

[01:33:11] talking about to give an introduction we

[01:33:13] can tell students about what these tools

[01:33:15] tools are, where they come from, maybe

[01:33:16] what a generative AI model is. In a

[01:33:19] subsequent grade or class, we might tell

[01:33:21] them about how do these things actually

[01:33:22] hallucinate bad results, how they might

[01:33:25] incorporate bias into what they

[01:33:26] generate, and thinking about the notion

[01:33:28] of verifying references that these

[01:33:30] things produce, which now gets into a

[01:33:32] different learning process that they

[01:33:33] need to do to effectively use these

[01:33:35] tools. And ultimately, we might teach

[01:33:36] them how to be really effective users of

[01:33:38] the tools. For example, research that's

[01:33:40] been done on telling the models, for

[01:33:41] example, to think step by step, which

[01:33:43] causes them to solve problems that they

[01:33:45] don't solve when they're not asked to

[01:33:46] think that way. These are things that

[01:33:48] new research are revealing about how to

[01:33:50] use the tools effectively. And there are

[01:33:52] things that should be taught to students

[01:33:53] because if we don't teach them, they

[01:33:55] will teach themselves. And what they've

[01:33:56] already taught themselves to do for 70

[01:33:58] or 80% of them is use these tools

[01:34:00] ineffectively to shortcircuit their

[01:34:02] learning process. So, can we get to the

[01:34:04] final provocation, which is around

[01:34:05] creativity and thinking about novelty

[01:34:07] versus regurgitation? Can AI actually

[01:34:10] produce something novel? Now, here's the

[01:34:12] interesting thing. A lot of critics say

[01:34:13] AI is just trained on certain

[01:34:15] information and it can only regurgitate

[01:34:17] that information. Okay? What do we do

[01:34:19] for most of education,

[01:34:22] right? Exactly the same thing. And

[01:34:25] somehow we accept students, we expect

[01:34:27] students to be able to generate novel

[01:34:28] results. So the provocation there is can

[01:34:31] we actually help AI to produce novel

[01:34:34] results in concert with human beings. I

[01:34:36] think the answer there is yes. But if we

[01:34:38] want to be able to do that what it means

[01:34:39] is we need to teach creativity to both

[01:34:42] human beings and AI to be able to work

[01:34:44] in concert rather than just expecting

[01:34:47] students to generate yet another essay

[01:34:48] about the Civil War. We don't need more

[01:34:50] essays about the Civil War. What we need

[01:34:52] them is to think critically about

[01:34:54] engaging with the tools to do it. And so

[01:34:56] I'll leave you with the final points in

[01:34:58] terms of AI to spur creativity can AI

[01:35:00] generate new ideas. It's already been

[01:35:02] used by for example hard sciences to

[01:35:05] generate hypotheses that the scientists

[01:35:06] themselves had not actually considered

[01:35:08] before. And people in mathematics for

[01:35:10] example see the future of mathematics as

[01:35:12] working collaboratively between humans

[01:35:14] and AI to be able to generate larger

[01:35:16] results in math. So thanks very much for

[01:35:18] your attention.

[01:35:26] All right, our next panelist uh is here

[01:35:28] in the graduate school of education, but

[01:35:30] I just learned was uh long ago an intern

[01:35:33] at Adobe. So, no wonder creativity

[01:35:35] planted a seed. Um please join me in

[01:35:38] welcoming Hari Subramanium.

[01:35:46] Okay. Uh thanks Brian. Hello everyone.

[01:35:49] um very recently was asked this really

[01:35:51] interesting question. If AI could fix

[01:35:54] one thing about today's education

[01:35:56] system, what would it be? Now, before I

[01:35:59] give you my answer, why don't you take a

[01:36:01] second to think about how you might

[01:36:02] answer it? Although Maharan may already

[01:36:04] have given you some clues about it?

[01:36:10] Well, my answer is to equip students

[01:36:12] with the infrastructure to learn by

[01:36:14] creating things.

[01:36:17] You see, um, the ultimate effect that

[01:36:19] education has on a learner is to advance

[01:36:23] their capacity to create, to generate

[01:36:25] ideas, to solve problems, and to build

[01:36:29] things that didn't exist before. Across

[01:36:32] all these different domains, whether

[01:36:33] it's design or engineering or science or

[01:36:36] uh, media or craftsmanship, creation is

[01:36:40] the ultimate test of understanding.

[01:36:45] Yet in today's classrooms, a lot of

[01:36:48] learning is organized around what to

[01:36:50] know instead of what to create. Students

[01:36:54] learn facts and concepts, they memorize

[01:36:57] procedures, and they solve familiar

[01:36:59] problems.

[01:37:01] And we find that this is a little bit

[01:37:03] problematic in many ways because one, it

[01:37:07] introduces this purpose gap. When

[01:37:09] knowledge isn't acquired purposefully,

[01:37:12] when learners don't have the opportunity

[01:37:14] to create while learning, we find that

[01:37:16] the knowledge they acquire is often

[01:37:18] abstract and fragmented. You may have a

[01:37:21] student who aces like a physics test in

[01:37:23] fractions, but will not be able to

[01:37:25] explain why a sled slides differently on

[01:37:28] ice compared to concrete.

[01:37:32] And we we my argument is that sort of

[01:37:36] creation fundamentally changes how we

[01:37:39] learn.

[01:37:40] When we create we are forced to sort of

[01:37:43] organize knowledge and represent

[01:37:46] knowledge in purposeful ways. We are

[01:37:48] forced to make connection between ideas

[01:37:51] and we are forced to consolidate

[01:37:53] knowledge in a way passive consumption

[01:37:55] can no long can cannot do. So here is an

[01:37:59] example where imagine introducing like

[01:38:01] 3D geometric shapes to a student. And if

[01:38:04] you just do that, all they're going to

[01:38:06] remember are these different shapes. But

[01:38:08] instead, if you ask them to model a real

[01:38:10] world object, in this case a sneaker,

[01:38:12] the student now needs to understand how

[01:38:14] these different shapes combine, uh how

[01:38:16] do edges align, how to compute complex

[01:38:19] volumes, and so on. All of this is what

[01:38:22] makes knowledge that we acquire in K

[01:38:24] through2 education more purposeful.

[01:38:27] So if creativity and creation is

[01:38:30] essential uh for learning, how do we

[01:38:33] make it accessible to all learners?

[01:38:36] And this is where like AI can play a

[01:38:38] role.

[01:38:41] First of all, AI can sort of provide the

[01:38:43] infrastructure to lower the floor for

[01:38:45] creation to get students get started in

[01:38:48] creating things. What you're seeing here

[01:38:50] is an example where a child sketches an

[01:38:53] ordinary everyday scenario, a car on the

[01:38:55] street. AI recognizes the scenario and

[01:38:58] generates what we call situated physics

[01:39:01] brushes. In this case, like a physics

[01:39:03] brush for friction that the student can

[01:39:06] then paint over their scenario.

[01:39:08] Notice that we're not introducing

[01:39:10] friction as this abstract formula.

[01:39:12] Instead, friction is presented as a

[01:39:14] creative tool where the student can

[01:39:16] paint the street with ice and paint the

[01:39:18] street with gravel and see how the car

[01:39:20] moves on these different surfaces. They

[01:39:22] can explore and start asking questions

[01:39:24] like will the car move faster in ice?

[01:39:26] What if what happens if we combine like

[01:39:28] ice and gravel or how much friction or

[01:39:31] what material is needed to get the car

[01:39:33] to completely come to a stop? The

[01:39:35] concept of friction sort of like emerges

[01:39:38] through this process of exploration

[01:39:40] through this process of testing out

[01:39:41] their intuitive theories about physics

[01:39:43] based on everyday behavior they've

[01:39:45] already observed. And over here AI can

[01:39:48] actually help by sort of producing by

[01:39:50] analyzing these scenes and producing the

[01:39:52] necessary creative tools for learners to

[01:39:54] create but also automating away the more

[01:39:57] uh design like a difficult or mechanical

[01:40:00] process of creation in this case like

[01:40:02] simulating an animation.

[01:40:05] A second example is in addition to like

[01:40:08] you know lowering the floor floor for

[01:40:10] creation to get learners started it can

[01:40:12] also like raise the ceiling where

[01:40:14] learners can be more expressive and can

[01:40:17] expand based on their understanding.

[01:40:19] What you're seeing here is a creative

[01:40:21] tool for generative art. Over here the

[01:40:24] learner is using AI and natural language

[01:40:27] to combine concepts in new and bespoke

[01:40:29] ways. They are trying to map the

[01:40:31] jellyfish behavior, the how a jellyfish

[01:40:33] moves with abstract rigid fractal

[01:40:35] shapes. If we were doing this in a

[01:40:38] traditional programming uh paradigm,

[01:40:40] then the student would have to learn how

[01:40:42] to program these individual behaviors,

[01:40:44] how to sort of engage in like construct

[01:40:46] recursive patterns and how to apply

[01:40:48] these interactions. But instead over

[01:40:50] here the learner can focus on creative

[01:40:52] composition and thinking about

[01:40:54] conceptual blending and thinking about

[01:40:55] abstractions while AI can do this uh

[01:40:58] harder technical work that is not

[01:41:00] central to their learning.

[01:41:03] And finally you know AI can provide

[01:41:06] feedback and guidance in bridging this

[01:41:09] gap or like this going back and forth

[01:41:11] between knowledge and practice. On the

[01:41:13] left, what you see is a simulator where

[01:41:15] a learner can adjust the angle, force,

[01:41:18] and gravity to understand the trajectory

[01:41:20] of a basketball and thereby understand

[01:41:23] how the laws of motion works. An AI over

[01:41:26] here can help the learner in thinking

[01:41:28] about like how to run these experiments

[01:41:30] in understanding or interpreting the

[01:41:33] evidence or the observation that they

[01:41:34] generate through each runs. On the

[01:41:37] right, you're seeing a tool where blind

[01:41:39] and low vision learners can create

[01:41:41] accessible content and engage in spatial

[01:41:44] reasoning. Over here, AI can provide

[01:41:46] rich descriptions and helping them

[01:41:47] construct this understanding.

[01:41:51] So, in conclusion, I want to sort of um

[01:41:54] leave with this provocation. I find it a

[01:41:56] little bit ironic that we're using a

[01:41:59] technology that is somewhat powerful

[01:42:01] today in helping learners consume

[01:42:03] knowledge while leaving the act of

[01:42:05] creation to the AI. And I think we

[01:42:08] should sort of think a little bit

[01:42:09] differently and focus on how do we equip

[01:42:11] learners to be more creative and use

[01:42:14] creative work uh in engaging in sort of

[01:42:16] like their own um cognitive task of

[01:42:19] learning. So in summary, knowledge is

[01:42:21] powerful. um creation helps learners

[01:42:24] sort of organize and structure knowledge

[01:42:26] in more meaningful usable ways and AI

[01:42:29] should support this. Thank you.

[01:42:37] All right. And our final panelist also

[01:42:40] from the graduate school of education

[01:42:42] here. It's actually doing incredible

[01:42:43] research looking at what happens in real

[01:42:46] classrooms when we take a specific tool

[01:42:48] and a specific creative task. What

[01:42:50] happens with that learning? So please

[01:42:51] join me in welcoming Glish.

[01:42:54] Thank you.

[01:42:59] Thank you very much. Being an economist

[01:43:01] by training, you know, over the years

[01:43:03] I've often been confronted with these

[01:43:05] ideas that you know this boom is not

[01:43:08] going to be followed by a bust because

[01:43:10] this time is different or you know this

[01:43:13] depression is we're never going to

[01:43:14] recover from this because this time is

[01:43:16] different. So let me present you with my

[01:43:18] version of this time is different. Why I

[01:43:21] think Genai is not just like a

[01:43:24] calculator.

[01:43:26] I have to learn how to move it forward

[01:43:28] first.

[01:43:29] Oh, the green one.

[01:43:31] Oh, thank you.

[01:43:33] We've heard other metaphors for for

[01:43:35] Genai and I think there's at least two

[01:43:38] big differences between the two

[01:43:41] technologies and I think they are

[01:43:43] important for what we're discussing

[01:43:44] today, the relationship to creativity.

[01:43:47] So calculators, you know, they're

[01:43:49] available even in super low resource

[01:43:51] settings such as, you know, the ones

[01:43:52] that I often engage for my research.

[01:43:55] Whereas Genai, we know it depends on

[01:43:57] devices, internet, subscription

[01:43:59] services. What is free today might not

[01:44:01] be free tomorrow. So we might worry

[01:44:02] about that especially when it comes to

[01:44:04] equity. But there's one extra dimension

[01:44:06] that I think it's might be pretty

[01:44:08] fundamental at least at the moment,

[01:44:10] which is the extent to which these

[01:44:12] technologies automate tasks. So

[01:44:14] calculators never do the whole thing for

[01:44:16] you unless you're using a very

[01:44:17] sophisticated scientific calculator.

[01:44:20] They will do the repeatable kind of

[01:44:22] algorithmic components of the task and

[01:44:25] still leave it to the student to the

[01:44:27] user to interpret what the problem is to

[01:44:30] figure out what calculations have to be

[01:44:31] inputed into the machine such that you

[01:44:34] get the answer that you want. Whereas

[01:44:36] jai

[01:44:38] at least again at the moment without

[01:44:40] guard rails we'll talk more about that

[01:44:41] in principle can automate all the

[01:44:43] components of a task and that I want to

[01:44:46] argue can be pretty fundamental to how

[01:44:48] we might or might not encourage

[01:44:51] creativity and learning. So it's you

[01:44:55] know once you can automate everything

[01:44:57] it's pretty obvious that you know we

[01:44:59] need guardrails if we want to use it in

[01:45:01] the classroom especially in K through2

[01:45:03] and then once we start thinking about

[01:45:05] guard rails hard questions emerge like

[01:45:09] we want the technology to scaffold

[01:45:11] learning to help students you know

[01:45:13] acquire knowledge. So what's the right

[01:45:15] model to do it? uh what you know what is

[01:45:18] the until what point do we do we want

[01:45:21] the technology to do the task on for the

[01:45:23] students you know what what kind of help

[01:45:25] do are we willing to provide and then as

[01:45:27] a result of this gario capian supports

[01:45:30] uh are students learning more or not so

[01:45:33] we ran this study this is in Rio

[01:45:36] southeast Brazil uh with middle school

[01:45:39] students and we wanted to know that with

[01:45:41] whether with some guard rails uh student

[01:45:44] creativity will be helped or harmed by

[01:45:49] Genai assistance. We thought I want to

[01:45:51] say that there was a need to conduct a

[01:45:53] study like this because first if you

[01:45:55] just compare students that are already

[01:45:57] using genai versus those who are not and

[01:45:59] you try to ask who is more creative

[01:46:01] that's not necessarily you know the what

[01:46:04] the technology is causing right in terms

[01:46:07] of the its impacts of on creativity

[01:46:08] because who uses J might be very

[01:46:10] different and as I say at the student

[01:46:12] level at the school level at the

[01:46:14] district level so you want some more

[01:46:16] rigorous evidence that you can trust so

[01:46:19] we're going to run experiments. The

[01:46:21] second thing is you've seen studies like

[01:46:25] uh the famous now I think people call it

[01:46:27] the MIT brain rot study where you know

[01:46:31] some university students had access to a

[01:46:33] tool they some some wrote a text with ji

[01:46:38] some did not and then their brain

[01:46:40] engaged less in it they remember it less

[01:46:41] if they had ji assistance well to some

[01:46:44] extent a lot of those studies they kind

[01:46:46] of give you the to and they take it away

[01:46:48] and then perhaps it's no surprise that

[01:46:49] you get much worse wants to no longer

[01:46:51] have that tool. So it's a a big question

[01:46:52] is the education is about transfer. What

[01:46:54] can you do not only in the same task

[01:46:56] that you had the assistance but when the

[01:46:58] tool is taken away in a new task? Can

[01:47:01] you transfer knowledge? You know if

[01:47:03] you're better at first you continue to

[01:47:05] be better or maybe you get worse. So

[01:47:07] that's what we're going to look at. We

[01:47:08] measure creativity because we're

[01:47:10] interested in this uh in creating novel

[01:47:14] and useful ideas by focusing on

[01:47:16] divergent thinking which is the ability

[01:47:17] to come up with new ideas and we use two

[01:47:21] common tasks in the literature. One is

[01:47:23] the alternative uses task. So we're

[01:47:25] going to give you an object and ask what

[01:47:26] else could you use it for other than the

[01:47:28] most straightforward use. The more valid

[01:47:30] uses you come up with the better the

[01:47:32] higher your score. Then the other test

[01:47:34] so these are some of the objects they

[01:47:35] will do it for. And another task is the

[01:47:38] diver association task. So I give you a

[01:47:40] seat word and I say come up with a list

[01:47:42] of 10 words that are as different from

[01:47:43] each other as you can and you start with

[01:47:46] the first word. So you know this is not

[01:47:48] an easy task. Here's just some guidance

[01:47:50] on how to do it. And here's our creative

[01:47:52] buddy. So we created some AI assistance

[01:47:55] but we put guard rails in it. Here's the

[01:47:57] prompt. I just want to call your

[01:47:58] attention to the guardrail. So it says

[01:48:01] if you know if the user asks you for 10

[01:48:03] words just give it three. So give some

[01:48:05] assistance but don't do the task

[01:48:07] entirely for the student. Okay. So we're

[01:48:09] gonna take you through the results very

[01:48:11] quickly because of my time. So first I

[01:48:14] mean this is just based on

[01:48:16] randomization. We're going to compare

[01:48:17] kids the hedge and AI systems in the

[01:48:19] first test the alternative uses test

[01:48:21] versus not here. They start the same at

[01:48:24] the you know several trials. In the

[01:48:26] first one you already see kids with

[01:48:27] assistant jump ahead. And then over the

[01:48:29] course of this test when they have

[01:48:31] assistance they do better. They're more

[01:48:32] creative. That's not surprising. Now

[01:48:34] what happens when we take the assistance

[01:48:36] away

[01:48:38] this advantage disappears. So at first

[01:48:40] within the same test doesn't seem like

[01:48:41] there is positive transfer already

[01:48:44] pretty bad news but then the big

[01:48:45] question is in the new task in the

[01:48:47] diversion association task we're going

[01:48:48] to randomize again what happens if you

[01:48:51] still have AI assistance if you never

[01:48:54] had AI assistance if you lost AI

[01:48:56] assistance we're going to have four

[01:48:57] groups so this is what we find if you

[01:49:00] never had AI assistance the control

[01:49:01] group you do the best in this task now

[01:49:04] if you continue to be assisted or you

[01:49:06] only had AI assist in the second test

[01:49:07] you do a little worse but it's not

[01:49:09] statistic basically different but now if

[01:49:11] you had a system at first and you lost

[01:49:12] it in the second test then it do so much

[01:49:15] worse actually the negative transfer is

[01:49:18] kind of four-fold the advantage you had

[01:49:19] before what's happening I don't have

[01:49:21] much time to tell you it's not just

[01:49:23] persistence it's a little bit about you

[01:49:26] not don't have as much fun doing it but

[01:49:27] most importantly you start thinking that

[01:49:29] AI is more creative than you and the

[01:49:32] negative effects are concentrated on

[01:49:33] those kids who really think that AI

[01:49:35] became more creative than them so I

[01:49:37] think there's stuff to for us to Think

[01:49:39] about uh about how access to these tools

[01:49:41] without maybe the right guard rails, the

[01:49:43] right scaffolding might actually harm

[01:49:45] the academic self-concept and that's

[01:49:47] pretty important especially in the K12

[01:49:49] setting. Let me stop you.

[01:50:00] All right, while we swap a mic out. So,

[01:50:02] we've laid a lot of groundwork here. Um

[01:50:04] this question is for all of you. uh and

[01:50:07] most of you covered in in your uh quick

[01:50:09] talks. How do you how do you define

[01:50:11] creativity? Because it's a broad tent.

[01:50:14] Creative expression, creative problem

[01:50:15] solving, divergent thinking. How do you

[01:50:17] define creativity? And has that

[01:50:20] definition become more refined as

[01:50:22] generative AI has become part of your

[01:50:24] practice or your research?

[01:50:27] Chris, you're the closest.

[01:50:29] Sure. Okay.

[01:50:31] Um,

[01:50:33] in music where what we're doing is

[01:50:38] communicating with the emotions and

[01:50:41] feelings of of both ourselves, our inner

[01:50:44] selves and and our audiences and trying

[01:50:46] to make a bridge. uh we're doing that

[01:50:49] through sound and and the examples that

[01:50:52] I showed were all ultimately ar the

[01:50:54] arbiter is your ear and you need you

[01:50:58] need to be able to feed your ear feed

[01:51:01] those emotions in the same way you do if

[01:51:03] you're practicing an instrument or

[01:51:05] you're composing with some of those

[01:51:07] shrink wrap tools that I mentioned. Um

[01:51:11] where

[01:51:13] where that falls short is if you don't

[01:51:16] have access to music all over the world

[01:51:21] through your life through history.

[01:51:24] You know the the wide stage of of music

[01:51:27] and the fact that nothing is ever really

[01:51:32] truly denovo. It comes from other

[01:51:35] things. you know, you're you you're

[01:51:37] you're standing on shoulders of of

[01:51:41] other pairs of ears, if you will. So,

[01:51:45] that's how I define it. It's it's it's

[01:51:47] imagination and it's also imp placement

[01:51:50] in in something that is cultural, I

[01:51:53] guess. So, I'll I'll leave it there.

[01:51:56] Yeah.

[01:51:56] Yeah, there's there's so many. Um I mean

[01:51:59] I just focused on problem solving

[01:52:00] because that's kind of the domain I come

[01:52:02] from is you know computer science we

[01:52:03] think about writing programs often to

[01:52:04] solve problems but obviously there's you

[01:52:06] know creative expression as a way of

[01:52:08] whether it be music or writing or

[01:52:10] sculpture you know many other forms but

[01:52:13] I think when we think from the

[01:52:15] standpoint of education it's it's kind

[01:52:17] of tragic in some sense that we have one

[01:52:19] word for creativity because it means so

[01:52:21] many things and I think you know keeping

[01:52:22] in mind the context of the creativity

[01:52:25] that we're trying to instill students

[01:52:26] with in a particular situ situation um

[01:52:29] gives us more information about how we

[01:52:31] might do that with say for mathematics

[01:52:33] as opposed to art.

[01:52:35] That's great. I think often when you ask

[01:52:37] a room full of teachers across subjects,

[01:52:39] do you teach creativity? There's not a

[01:52:41] lot of hands. But then when you define

[01:52:43] it, do you teach problem solving and

[01:52:45] different ways of approaching things?

[01:52:46] Suddenly the hands go up. So definitions

[01:52:48] can help or hinder.

[01:52:50] Yeah. Yeah. So I do um I make art in my

[01:52:53] free time. So for the longest time I

[01:52:55] held this definition of creativity as

[01:52:57] sort of using imagination to produce

[01:52:59] something that's novel and surprising

[01:53:01] and interesting and like museum worthy.

[01:53:04] But you know recently with AI we've

[01:53:07] noticed that they've gotten really good

[01:53:08] at producing these artifacts of

[01:53:10] intelligence and so has and also in like

[01:53:14] working through like you know my own

[01:53:15] research my definition of creativity has

[01:53:18] like shifted greatly from like this

[01:53:21] artifact to like creativity as this

[01:53:24] process of like shaping ideas through

[01:53:27] this constellation of tools and

[01:53:29] knowledge and artifacts that you know

[01:53:32] get generated in the process. So it's

[01:53:34] more internal and cognitive than like

[01:53:37] external and like something that's

[01:53:39] created.

[01:53:41] Yeah, I think there's at least two

[01:53:44] processes that interact to generate this

[01:53:46] novel and creative ideas. I talked about

[01:53:48] divergent thinking which is kind of

[01:53:50] brainstorming phase as many and then

[01:53:54] there's also convergent thinking which

[01:53:56] is later how do I filter all this

[01:53:58] brainstorming outputs into something

[01:54:01] that really serves a task

[01:54:03] and test could be like solving a real

[01:54:04] problem. It could also be being

[01:54:06] expressive and so on. Of course, how you

[01:54:08] measure those things is super hard. I've

[01:54:10] showed you how, you know, two examples

[01:54:12] of tests are used to measure the first

[01:54:14] one, they're very reductive, right? And

[01:54:17] and it's just written expressions. So

[01:54:19] there's no picture expression and so on.

[01:54:21] So when you come to when it comes to the

[01:54:24] challenge of measuring things, then we

[01:54:26] always have to reduce this complicated

[01:54:27] and beautiful abstract concept into

[01:54:29] something that's very very

[01:54:30] constrained. Yeah, that's worth.

[01:54:33] That's great. Um, Ghe, while while

[01:54:36] you're on the hot seat here, you you

[01:54:38] introduced this notion of, you know,

[01:54:40] good technology if it's in the classroom

[01:54:41] should ideally be lowering the floor or

[01:54:44] raising the ceiling. Um, so maybe I'll

[01:54:46] start with you and then we'll go back

[01:54:48] this way. What what are ways in which

[01:54:49] you've seen that that AI can lower the

[01:54:52] barrier of entry? um especially for

[01:54:54] students who might struggle or not see

[01:54:56] themselves as creative or not really

[01:54:59] have the opportunities to practice

[01:55:00] creative thinking.

[01:55:02] Yeah, I think so. In the context of this

[01:55:05] creative buddy that we developed, you

[01:55:07] could get assistant as simply as you

[01:55:08] know clicking a button which is you know

[01:55:12] conducive to like even kids that might

[01:55:14] struggle to write like a lot of these

[01:55:16] tests. What's challenging is they embed

[01:55:18] a lot of requirements including the

[01:55:20] ability to read and write well. So

[01:55:22] sometimes you know They're more

[01:55:24] accessible just because there's

[01:55:25] different ways to which you can interact

[01:55:27] with them. Harish is a lot more example

[01:55:29] this. So I think lowering the floor I

[01:55:31] think is easy to understand. Raising the

[01:55:33] ceiling sorry yeah raising the ceilings

[01:55:36] I think is not so obvious. Um you can

[01:55:39] see that of course it will if it's just

[01:55:41] diversion thinking it can increase the

[01:55:42] set of things that you would list if uh

[01:55:45] you didn't have the tool but it's not

[01:55:46] obvious as we were talking about

[01:55:48] processes not just outputs that is

[01:55:49] really mobilizing the internal like

[01:55:52] self-regulating the process of

[01:55:54] developing the skills such that the

[01:55:55] ceiling is raised even when you don't

[01:55:57] have access to the technology.

[01:55:58] Yeah.

[01:55:59] Hari, how do you think about how this

[01:56:01] technology can lower the barrier of

[01:56:03] entry?

[01:56:03] Yeah. Yeah. sort of building on the

[01:56:05] example I I gave in my talk, right? I

[01:56:08] think the one thing that's um really

[01:56:11] powerful about AI is like one it is

[01:56:14] dynamic. It can adapt to the learner's

[01:56:16] task either interactively or like you

[01:56:18] know through some like system prompting.

[01:56:21] Um but also that it can take away a lot

[01:56:24] of the mechanical work of doing things

[01:56:27] like you know in the past like if a

[01:56:28] student wanted to see like how their

[01:56:30] sketch how friction might work with

[01:56:32] their sketch it's going it would have

[01:56:34] required lots and lots of effort but as

[01:56:36] I showed in like the video today like

[01:56:38] you can take like a simple sketch you

[01:56:40] can use AI to sort of understand the

[01:56:41] scene and you can give them a friction

[01:56:43] brush that they can use. So that's sort

[01:56:45] of like lowering the floor. But now the

[01:56:47] student can start to expand on like you

[01:56:49] know these counterfactuals like what if

[01:56:51] friction didn't exist and like what if

[01:56:52] there was an object falling on the

[01:56:54] street and there was no friction like

[01:56:55] how would this happen? How does friction

[01:56:57] interact with like acceleration and

[01:56:59] motion like can I sort of add a motion

[01:57:01] brush brush or like change the mass of

[01:57:03] the car to kind of like explore how

[01:57:05] friction and motion interact. So I think

[01:57:07] it sort of expands the space of

[01:57:09] possibilities in a more dynamic way

[01:57:12] thereby getting people to like generate

[01:57:14] new knowledge in the process of learning

[01:57:17] beyond like what is already like

[01:57:19] presented to them or like what they need

[01:57:20] to learn conceptually.

[01:57:23] Yeah. I mean one thing I was hear you

[01:57:25] know teaching introductory computer

[01:57:27] programming is do we need to do that

[01:57:28] anymore? Is everyone just going to

[01:57:29] program in English? And there's the

[01:57:32] answer to that I think is many more

[01:57:34] people will program. they will program

[01:57:35] in English and I don't want those people

[01:57:37] building the credit card transaction

[01:57:38] system on the news. So I think it does

[01:57:42] lower the floor because now people have

[01:57:44] access to power tools. They can actually

[01:57:46] build things they could not build before

[01:57:47] by just describing in English to a

[01:57:50] computer system what they want to do and

[01:57:52] see the power of building something that

[01:57:54] gets executed for them. And I think that

[01:57:56] transformation from going from being a

[01:57:58] consumer of technology to producer of

[01:57:59] technology is actually hugely powerful

[01:58:01] for a student. So I do think that has

[01:58:04] already lowered the floor and it has

[01:58:06] raised the ceiling because for people

[01:58:07] who are seasoned programmers who really

[01:58:09] know what they're doing or experts in

[01:58:10] the domain, it has made them more

[01:58:12] productive. So you go and talk to folks

[01:58:14] like Microsoft and Google, they say a

[01:58:16] large portion of our codebase now is

[01:58:18] actually being written by AI and the

[01:58:20] people who are very filive

[01:58:24] before.

[01:58:26] Yeah, let me echo everything that was

[01:58:28] just said by going back to my examples.

[01:58:31] um the 500 lines of code that I might

[01:58:35] emit in one hour today. What do I use

[01:58:37] that for? Well, like the pumpkin that

[01:58:40] would have taken me way too long to

[01:58:42] generate, you know, that that demo for

[01:58:45] teaching um common fate. Well, there's a

[01:58:48] whole bunch of other Gestalt principles

[01:58:50] and I can't wait to get there and, you

[01:58:52] know, make make illustrations. So, I

[01:58:54] think it's it's um it's that that one

[01:59:00] line that generates something usable.

[01:59:04] It elevates the conversation in in in a

[01:59:07] really good way that um that's going to

[01:59:09] affect what I can teach, how fast I can

[01:59:12] teach it. If you look at the the the

[01:59:13] daylow eyebrows example, that was just a

[01:59:16] lark. That was a DIY experiment by one

[01:59:19] of the students who accompanied a

[01:59:21] different assignment. Just try it out.

[01:59:23] Well, that opened up a treasure of um

[01:59:28] different components that now are

[01:59:31] accessible that were way beyond the

[01:59:32] scope of the class, you know. So, those

[01:59:34] those kinds of things are just opening

[01:59:36] and I think the accessibility, you know,

[01:59:39] lower we can do things right off the bat

[01:59:41] that are inspiring and that's that's the

[01:59:46] the I don't know that's that's the

[01:59:47] readout from two quarters worth of doing

[01:59:50] this so far. You know, hands on. Thank

[01:59:53] you.

[01:59:54] Um, this topic could be its own

[01:59:56] conference that we could spend three

[01:59:57] days on. Um, which I would love it as

[01:59:59] well. That's an idea. We should do this.

[02:00:01] Um, but, uh, we're at time for now. The

[02:00:04] good news is we have a break and so we

[02:00:06] can continue this discussion with our

[02:00:08] speakers and amongst ourselves. So,

[02:00:09] thank you all so much.

[02:01:20] Hey.

[02:01:27] Hey.

[02:06:11] Heat. Heat.

[02:07:52] Hey,

[02:07:55] hey, hey.

[02:09:02] Hey.

[02:11:18] Heat. Heat.

[02:12:55] It's not. Oh, there it is.

[02:12:58] Hey.

[02:13:01] Hey, everybody. If you could head back

[02:13:03] to your seats, we'd love to get started

[02:13:05] with our next panel. um as quickly as

[02:13:08] possible.

[02:13:21] All right, I'm starting to see some

[02:13:23] movement.

[02:13:44] Give him like

[02:13:52] All right, if we can get back to our

[02:13:54] seats, we'd love to get started with

[02:13:56] this next panel.

[02:14:01] Yeah. I mean, you know, they they ran

[02:14:05] out of coffee in the in the in the

[02:14:06] hallway. Yeah. Yeah. Yeah.

[02:14:10] I'm tempted to shush people. Yeah.

[02:14:19] Right.

[02:14:23] I know. See, I'm not I'm not much of a

[02:14:25] clapper, though.

[02:14:29] You want to We could do it one more

[02:14:30] time. You want to lead?

[02:14:32] Here we go.

[02:14:34] Me and Michael or you can just We can

[02:14:36] just clap it out.

[02:14:37] Okay.

[02:14:44] All right.

[02:14:45] All right. See? Yeah. You just you just

[02:14:47] need tools of educators. That's easy.

[02:14:51] Okay. Well, we've got a few people

[02:14:53] coming back in, but but would love to

[02:14:55] get started again because we have

[02:14:57] another exceptional panel coming up

[02:14:59] right now. Um, this next panel is titled

[02:15:02] how AI is transforming how we teach and

[02:15:05] we're going to hear real world stories

[02:15:07] about how AI is affecting educators and

[02:15:09] learners. Um, which is sort of a

[02:15:12] tremendous way for us to see the impact

[02:15:14] on the ground. And so to kick us off,

[02:15:17] I'd like to invite our moderator Maisha

[02:15:19] Win to the stage. Maisha is the

[02:15:21] excellence and learning professor at

[02:15:22] Stanford GSSE um and is the faculty

[02:15:25] director of the equity and learning

[02:15:26] initiative at the accelerator for

[02:15:28] learning. So Misha over to you.

[02:15:32] Thanks so much.

[02:15:46] Good morning everybody. It's so nice to

[02:15:48] see everyone. The Stamford Accelerator

[02:15:51] for Learning's Equity and Learning

[02:15:52] initiative had the great privilege and

[02:15:55] honor to host students, middle school

[02:15:57] and high school students and their

[02:15:59] educators yesterday with the support of

[02:16:02] Google.org. We had a youthpowered AI day

[02:16:05] of learning. And let me tell you, we

[02:16:07] learned a lot from and with each other.

[02:16:09] And so I'm really excited to moderate

[02:16:12] this panel with three of our educators

[02:16:14] who participated in our day of learning

[02:16:16] yesterday. So, first I'm going to invite

[02:16:19] you to introduce yourselves and say

[02:16:22] something about the work that you're

[02:16:24] doing with AI and teaching.

[02:16:28] Okay. Well, I am Hello everyone. My name

[02:16:31] is Lud Wiglins, also known as Lulu. I am

[02:16:34] the manager of program impact at STEM

[02:16:37] from Dance. We are a nonprofit uh where

[02:16:39] we integrate technology and movement um

[02:16:43] essentially allowing for students to

[02:16:45] take these different workshops so that

[02:16:46] they can also learn how to combine using

[02:16:50] dance to bridge their knowledge of STEM

[02:16:52] and most recently we also have included

[02:16:55] AI into our work. Great. Hi everyone.

[02:16:58] I'm Daniela Dejakamo. Nice to see you

[02:17:01] all. Um I am an associate professor at

[02:17:03] the University of Kentucky. That is my

[02:17:05] day job um in learning sciences and

[02:17:07] youth development with a focus on civic

[02:17:09] engagement, information literacy, and

[02:17:11] student voice. Um but my fun or after

[02:17:14] hours or night job, I think um and the

[02:17:17] reason I'm here is that I help um run

[02:17:20] and organize the Kentucky student voice

[02:17:21] team, which is a youthled nonprofit in

[02:17:24] Kentucky fighting for educational

[02:17:26] justice for about the last 15 years as

[02:17:29] research, storytelling, and policy

[02:17:31] partners. Um, and my role with them is

[02:17:34] actually supporting them to do youthled

[02:17:36] research on issues of interest to them,

[02:17:39] which this year happens to be, shocker,

[02:17:41] student perspectives on AI in learning.

[02:17:44] Uh, and that's the study I'll talk about

[02:17:46] a little bit today.

[02:17:48] Hi everyone, my name is Mike Tapman and

[02:17:50] my teaching journey started here 21

[02:17:52] years ago. I was in the STEP program at

[02:17:53] the GSSE. It's really fun to come home.

[02:17:56] Um, I'm spend sort of half my day as a

[02:17:59] teacher at Northstar Academy in Newark,

[02:18:00] New Jersey, where I work with 98 juniors

[02:18:02] and seniors every day in a program that

[02:18:04] explores the intersection of AI purpose

[02:18:06] development and career connected

[02:18:08] learning. And then the other half of the

[02:18:09] day, I'm helping to lead our networkwide

[02:18:11] AI instructional strategy at Uncommon

[02:18:14] Schools, which is what my school is part

[02:18:15] of, which is one of the biggest charter

[02:18:17] networks in the Northeast.

[02:18:19] Terrific. And so, Mike, we'll start with

[02:18:21] you. give us a little more insight about

[02:18:23] how AI is showing up in the day-to-day

[02:18:26] on the ground in your teaching.

[02:18:28] Yeah, we are working on an AI framework

[02:18:31] we've built called the AI driver's

[02:18:33] license. And the idea is to map that

[02:18:35] quintessential adolescent right of

[02:18:37] passage getting your driver's license

[02:18:38] onto this AI moment. And so the

[02:18:40] curriculum has four parts which I'll

[02:18:42] explain in a sec. But we're really

[02:18:43] fortunate to be collaborating with Dr.

[02:18:45] Femarie Vassel who's a posttock fellow

[02:18:48] here at the GSSE and high to explore

[02:18:50] formally what the learning environment

[02:18:52] is that this curriculum creates. The

[02:18:55] curriculum has four parts and the whole

[02:18:56] idea as you can imagine is to get

[02:18:58] students into the driver's seat and not

[02:19:00] the passenger seat when it comes to AI.

[02:19:02] The first part, my favorite part is

[02:19:05] choose a destination. We want students

[02:19:07] to look inside themselves, look at the

[02:19:09] world around them and decide where they

[02:19:10] want to go with AI before they open the

[02:19:12] tool to have a reason to go somewhere.

[02:19:14] When you get in a car, you don't discuss

[02:19:16] with the car where you're going. I know

[02:19:17] we're in Whimo land, but even so, you've

[02:19:20] chosen the destination. We want students

[02:19:22] to feel that way about AI. Part two is

[02:19:24] learn how to drive. How do these tools

[02:19:26] work? What does it mean to prompt, to

[02:19:28] provide context, to orchestrate agentic

[02:19:30] workflows, etc. Part three is open the

[02:19:33] hood. How are these technologies

[02:19:35] working? And what therefore are the

[02:19:36] implications for students? What are

[02:19:38] their affordances? What are their

[02:19:39] limitations? And the last part is I was

[02:19:42] an English teacher for 18 years. which

[02:19:43] is my other favorite part is discuss

[02:19:45] debate the rules of the road. We want

[02:19:48] students to start considering discussing

[02:19:50] with each other discussing with others

[02:19:51] in this space not just what AI can and

[02:19:53] can't do but what AI should and

[02:19:56] shouldn't do so that they start to

[02:19:57] realize that their voices matter right

[02:19:59] now and they can start to take part in

[02:20:01] shaping this world they're graduating

[02:20:03] into.

[02:20:04] I want to be in your class.

[02:20:05] Yeah.

[02:20:07] Daniela.

[02:20:08] Yeah. So for me as someone who supports

[02:20:10] young people to do youth participatory

[02:20:12] action research or what a lot of people

[02:20:14] call YPAR um a strand of community based

[02:20:17] research I I have to think um

[02:20:21] thoughtfully every year about how to

[02:20:23] support young people around the issues

[02:20:24] that really matter to them. And you know

[02:20:26] over the last six years since I moved

[02:20:28] from California to Kentucky for the job

[02:20:31] the interests that matter to young

[02:20:32] people you'll all be familiar with. that

[02:20:34] um when it comes to schooling has a lot

[02:20:36] to do with their mental health,

[02:20:37] well-being and their ability to connect

[02:20:39] in the world. And we know that co sort

[02:20:41] of exacerbated all of that or

[02:20:43] exacerbated all of the inequities that

[02:20:45] we know exist in schools. And so what we

[02:20:47] see year after year from the school

[02:20:50] climate audits that the students lead in

[02:20:51] schools across Kentucky to state level

[02:20:54] um audits and studies we do focused on

[02:20:57] how students feel about the banning of

[02:20:59] CRT or different um topics that go on.

[02:21:03] um is that they are I think as some

[02:21:05] speakers said earlier like the same

[02:21:07] issues that we think about in terms of

[02:21:08] what matters for teaching and learning

[02:21:10] still matter for students today and I

[02:21:12] think what I'm curious about that's

[02:21:15] coming up from what we see in the

[02:21:16] students research is how AI can be

[02:21:18] supportive of those help us

[02:21:20] collaboratively um work through some of

[02:21:22] those same problems.

[02:21:24] Thank you.

[02:21:25] Um so yes at STEM from dance we are

[02:21:28] actually aiding students with crafting

[02:21:32] different prompting of AI generated

[02:21:34] video. So they are actually learning how

[02:21:36] to ultimately output these different

[02:21:39] type of videos that work as digital

[02:21:41] backgrounds as they also take different

[02:21:43] workshops to choreograph their own

[02:21:45] performances and pieces um utilizing the

[02:21:48] AI that they are generating as their

[02:21:51] backdrops or background dancers as some

[02:21:54] may say. Um so really cool and fun

[02:21:58] things obviously that they are doing but

[02:22:00] also knowing that there are hindrances

[02:22:02] and there are ultimately you know down

[02:22:06] effects of AI and still trying to make

[02:22:08] sure that these students know that

[02:22:10] ultimately AI needs to work for them so

[02:22:12] it's not just utilizing it as some sort

[02:22:14] of clutch um but more so like okay how

[02:22:17] are we ensuring that they have the tools

[02:22:20] and resources to actually make AI work

[02:22:23] for them but that also personally I

[02:22:26] believe comes with making sure the

[02:22:27] educators themselves are equipped to do

[02:22:29] so as well. Um so I think that's one of

[02:22:32] my biggest challenges right now that I'm

[02:22:35] facing um is making sure that the

[02:22:37] educators that I work with equally are

[02:22:39] partnering with AI and also have the

[02:22:41] knowledge and resources they need um to

[02:22:44] support the students that they are

[02:22:46] centered in front of each day.

[02:22:48] This is something I really appreciated

[02:22:49] yesterday. We had classroom teachers. We

[02:22:52] had nonprofits who work and bridge their

[02:22:55] work with classroom teacher partners. We

[02:22:57] had public private charter schools

[02:22:59] represented and so many student

[02:23:02] learners. If you are one of the middle

[02:23:04] schoolers or high schoolers who

[02:23:05] participated with us last week, can you

[02:23:07] just stand up? I want to acknowledge you

[02:23:09] again. Can you stand up? Yes. Who are

[02:23:12] our young people? Yay.

[02:23:15] Thank you so much for being here. And my

[02:23:18] next question is, what are you learning

[02:23:21] from these young people about AI?

[02:23:26] Oh, um, natural question. Um, I think

[02:23:29] for me personally, what I've learned

[02:23:31] over the past years, especially, uh,

[02:23:33] having some of my background definitely

[02:23:35] start in the classroom as an educator

[02:23:37] myself prior to my current role, is that

[02:23:39] AI is not actually inherently bad,

[02:23:42] right? Instead, we should be making sure

[02:23:45] that students are actually still given

[02:23:47] the proper cognitive skills um to make

[02:23:51] sure that they're actually making AI,

[02:23:52] like I said, work for them, right? So

[02:23:54] like how do we make sure that they're

[02:23:56] able to utilize AI as this type of

[02:23:58] scaffold and giving them the feedback

[02:23:59] that they need uh to really sharpen the

[02:24:02] skills that they need to obviously go

[02:24:04] out into the real world rather than this

[02:24:06] idea of you know AI is something that we

[02:24:09] need to completely remove because that's

[02:24:12] not happening unfortunately right I

[02:24:14] think it's here and we have to ensure

[02:24:16] that the educators themselves are also

[02:24:18] able to evolve with the technology that

[02:24:20] we have in place um to really make sure

[02:24:23] that you know students are actually

[02:24:25] utilizing AI with caution but most

[02:24:27] importantly making sure that it is

[02:24:30] actually something that supports their

[02:24:32] education growth.

[02:24:33] Yeah. And so I'll speak to two themes

[02:24:36] that are coming out from our data so

[02:24:37] far. So we're actually in the midst of

[02:24:39] running a state level mixed method study

[02:24:41] of which the survey and interviews are

[02:24:44] designed and carried out by students

[02:24:46] with their peers across the state. And

[02:24:48] so so far our survey has 850 responses.

[02:24:51] Um and it's an open a mix of open-ended

[02:24:54] and liyker style items. And then um we

[02:24:57] are about to start to conduct um

[02:24:59] hopefully up to 40 to 50 interviews that

[02:25:01] are also peer-to-peer focused on this

[02:25:03] question of what do students what are

[02:25:05] their perspectives and attitude about AI

[02:25:07] in their learning lives. So, two of the

[02:25:09] things that are coming out I fear are

[02:25:11] going to be unsurprising for this group

[02:25:12] um but are that students want their

[02:25:15] teachers to support them in how and if

[02:25:18] and when and in what ways to use AI for

[02:25:20] their learning. So, we know they're

[02:25:23] doing it um in and outside of schools,

[02:25:26] but it's it's they are they are students

[02:25:29] are also aware of the instructional

[02:25:31] puzzle that this presents for their

[02:25:33] students or for their teachers and they

[02:25:35] want support um and vulnerability from

[02:25:38] the teachers in making visible that um

[02:25:41] they need support in in in their

[02:25:42] learning with that. The second theme

[02:25:45] comes from the fact that um there's a

[02:25:47] lot of variability when it comes to

[02:25:48] school level policy and folks know that.

[02:25:50] Um so the AI policy looks different with

[02:25:54] in schools between schools across

[02:25:57] schools within teach across teachers and

[02:25:59] that's really challenging for students

[02:26:01] because what they're saying is that we

[02:26:04] don't know if we can use in this class

[02:26:05] can we use it in this class it's an

[02:26:06] unfair advantage in this class what does

[02:26:08] it look like in that class and I think

[02:26:10] that's not a small thing to also

[02:26:12] consider in this because I think

[02:26:14] students are grappling with the same

[02:26:16] sort of ethical considerations that we

[02:26:17] are as adults in terms of what is this

[02:26:19] is cheating? Is it not? You know,

[02:26:22] everything from applying to college to

[02:26:24] applying to um work at a fast food

[02:26:27] restaurant. If if a peer is using it and

[02:26:29] I'm not, what's the right thing? What's

[02:26:32] the right move to do here? So, I think

[02:26:33] they want space and time and scaffolded

[02:26:35] adult support to talk through those

[02:26:37] questions that have more to do with sort

[02:26:39] of the philosophical question around it

[02:26:42] rather than the technical as well.

[02:26:45] Yeah. I'll have more anecdotes from this

[02:26:47] AI driver's license pilot later. We just

[02:26:49] started last week, but from my own

[02:26:51] classroom, what I've noticed the last

[02:26:53] few years is first of all, and one of

[02:26:55] the professors mentioned this earlier,

[02:26:57] students are really hungry to look at

[02:26:58] this as a tool and a topic.

[02:27:00] And you know, when when whiteboards came

[02:27:02] in or Chromebooks came in, those weren't

[02:27:04] topics of discussion in class. You just

[02:27:06] use them. This wants to be both. And I

[02:27:08] think students need it to be both. And

[02:27:09] that's something that I think is

[02:27:10] important to think about. Two specific

[02:27:12] anecdotes and what I'm learning from

[02:27:14] students watching them use this come

[02:27:15] from the two students who kindly flew

[02:27:17] out to California with me for this. Ian

[02:27:19] and Briana. Um Ian has been working with

[02:27:21] a partner to vibe code an app called

[02:27:23] Fractal that visualizes the research

[02:27:25] process via expanding nodes that change

[02:27:27] colors, etc. So I'm watching this as a

[02:27:29] teacher the last few weeks and I'm

[02:27:31] wondering what are they learning about

[02:27:33] both research and also AI. And what was

[02:27:35] striking to me is that AI has sped up

[02:27:38] for Ian and his friend the idea to

[02:27:40] implementation to iteration process.

[02:27:42] Like that that's now really fast. The

[02:27:44] app just gets coded in class. but it

[02:27:46] hasn't sped up in a good way the

[02:27:48] thinking and collaborating they're

[02:27:49] doing. So, they set the tool off to code

[02:27:51] and then they're discussing what to do

[02:27:53] next. And I think as a teacher, it's

[02:27:54] really interesting to watch something's

[02:27:56] gotten faster, but something hasn't in a

[02:27:57] good way. Briana has been exploring a

[02:28:00] career in finance and sports management.

[02:28:02] And so, she wanted to create a contract

[02:28:03] for Pete Alonzo. If you're an East

[02:28:06] Coaster, he was a star on the Mets for a

[02:28:07] while. One of Brianna's favorite

[02:28:08] players. She wanted to be his agent.

[02:28:11] Hypothetically, I am an English teacher

[02:28:13] by trade. I've never created a major

[02:28:15] league contract. So, in that moment, we

[02:28:16] turned to our AI assistant and the AI,

[02:28:19] Brianna had the AI make a template of a

[02:28:21] contract, but not fill in the details.

[02:28:24] Number one, because it kept getting his

[02:28:25] stats wrong, but number two, because she

[02:28:28] said if she was actually his agent, it

[02:28:30] was her responsibility to get the

[02:28:32] contract right and to fill in those

[02:28:34] details. And so, to me, again, as a

[02:28:35] teacher, I'm watching that and I'm

[02:28:37] seeing AI open a door, but then I'm

[02:28:39] seeing Briana walk through it. And I

[02:28:41] think that's a really interesting way

[02:28:43] that we now have a possibility basically

[02:28:45] in classrooms to explore.

[02:28:47] You know, something that came up

[02:28:49] yesterday that was very interesting is

[02:28:51] this question about whether um there was

[02:28:54] this fear that AI was going to take over

[02:28:56] teachers jobs and our colleague Nathan

[02:28:58] was suggesting that the future

[02:29:01] potentially looked like teachers more as

[02:29:03] coaches during this time. And I'd love

[02:29:06] to get your thoughts around, you know,

[02:29:08] if you've been thinking about this or

[02:29:11] how you sort of see the future of

[02:29:13] teaching with AI in play and this notion

[02:29:17] of moving into more of a coaching role

[02:29:19] or support role.

[02:29:22] I mean I can't say if AI is going to

[02:29:25] necessarily take over the job of

[02:29:27] educators but what I can say right is if

[02:29:31] we as administration or uh you know

[02:29:35] individuals that are putting out these

[02:29:37] different pilots of AI it's our duty to

[02:29:39] ensure that educators are equally

[02:29:41] equipped. So rather than sitting and

[02:29:45] bustling with educators on their

[02:29:47] content, right, on how they're

[02:29:49] delivering the content, I think it's

[02:29:51] about the different tools that are now,

[02:29:52] you know, ever evolving and coming into

[02:29:54] the classroom. How do we actually coach

[02:29:56] and teach them to utilize these tools so

[02:29:58] that it can work for the students who

[02:30:00] are going to be battling with these

[02:30:02] tools that are going to be competing

[02:30:03] with these tools, right, on whether

[02:30:05] they're actually going to have these

[02:30:07] type of jobs. I think especially because

[02:30:09] students are thinking about this, it's

[02:30:11] our job and responsibility to like take

[02:30:13] action in this moment in time to take a

[02:30:15] step back but also want to ensure that

[02:30:18] we are equipping ourselves with how are

[02:30:21] we making sure that students still are

[02:30:22] having these jobs and they're not

[02:30:24] worried. I think for a very long time. I

[02:30:26] brought this up yesterday. Um this idea

[02:30:29] where we hadn't thought about um what do

[02:30:33] you call selfch check selfch checkckout

[02:30:35] machines right for a very long time. We

[02:30:37] had cashiers. Now all of a sudden selfch

[02:30:39] checkckout is a huge thing. You go into

[02:30:41] supermarket stores things of that sort

[02:30:43] and no one thought about the fact that

[02:30:44] these were taking people's jobs up until

[02:30:47] it actually did. And I think rather than

[02:30:50] waiting until the jobs are taken by AI,

[02:30:52] we need to ensure that we're to doing

[02:30:54] our due diligence to ensure that

[02:30:57] students are equipped with that

[02:30:58] knowledge. I think someone mentioned

[02:30:59] yesterday as well um this concept and

[02:31:03] idea of really making sure that like we

[02:31:06] know AI has these different like there's

[02:31:08] screeners, there's recruiters, like

[02:31:11] technically all our resumes are like

[02:31:12] being screened through these type of AI

[02:31:15] generators, right? So those are already

[02:31:17] taking up jobs. So, if we can get ahead

[02:31:19] of it right now while we're still in the

[02:31:20] early stages and phases of this new

[02:31:23] technology advancement, I think that's

[02:31:25] really our due diligence to make sure

[02:31:27] that our students are not competing with

[02:31:29] these different AI tools. Um, when it

[02:31:32] comes down to the job market,

[02:31:34] yeah, that's good. Um I think as a

[02:31:37] learning scientist who was trained in

[02:31:38] the socioultural tradition, I think that

[02:31:41] our job is to think about you know what

[02:31:44] we know about how people learn which is

[02:31:46] even more complex today in terms of

[02:31:49] living in the information and digital

[02:31:51] age and then what that means for how we

[02:31:53] ought to design learning environments in

[02:31:55] and outside of school. And so I think if

[02:31:57] we continue to think which a lot of

[02:31:59] people have said today is like I call in

[02:32:00] the learning sciences. I appreciate that

[02:32:02] shout out because no one knows about the

[02:32:03] field outside of these spaces. Um but uh

[02:32:07] if if we continue to think about how

[02:32:10] people learn and we think about how

[02:32:12] people learn in a digital age and a

[02:32:14] hyperartisan age that should actually

[02:32:17] amplify the need for better pedagogical

[02:32:20] and instructional strategies, right? And

[02:32:22] we know that learning is fundamentally

[02:32:23] cultural and social. And so I actually

[02:32:26] think it just

[02:32:28] brings up the need for more and and and

[02:32:31] better teaching.

[02:32:33] Uh to me I've had kind of two big

[02:32:35] thoughts about this. One is this tension

[02:32:37] you mentioned about the coach verse

[02:32:39] teacher and I I do think that's

[02:32:40] something a lot of educators are

[02:32:41] wrestling with right now. I feel it

[02:32:43] myself when Brianna's making that

[02:32:44] contract.

[02:32:45] I'm not I can't teach her how to write a

[02:32:47] major league contract. So I am guiding

[02:32:49] her through that experience using my

[02:32:50] educator instincts but it's a new topic.

[02:32:52] So there is a level at which this does

[02:32:55] push us in a good way to that sort of

[02:32:57] guide coach space. At the same time, I

[02:33:00] do worry that if we all become guides as

[02:33:03] teachers, who's teaching?

[02:33:06] And and that's a question like as an

[02:33:07] English teacher for 18 years, I think

[02:33:09] it's even more important to teach

[02:33:11] thinking, reading, writing, choosing

[02:33:12] your own words to express your ideas. I

[02:33:15] think that's a really crucial skill

[02:33:16] right now. But it's a real tension

[02:33:18] because I'm helping Brianna do that. But

[02:33:19] I also want to make sure everyone is

[02:33:21] remembering the the old ways we used to

[02:33:23] do things and bringing those forward.

[02:33:25] And I also think there's just a change

[02:33:26] management change management question

[02:33:28] there. If you tell a lot of teachers

[02:33:30] that they're not teachers anymore,

[02:33:32] they're coaches now. Uh some of us would

[02:33:35] love to jump into that. Some of us would

[02:33:36] be nervous, but some people would say,

[02:33:38] "I just spent my career teaching." And

[02:33:40] it's I think it's an interesting

[02:33:41] question to raise. on a sort of broader

[02:33:44] level. Um I always think when the

[02:33:46] students come in that we have 45 minutes

[02:33:48] together. How am I going to use that

[02:33:50] time? What's my lesson plan going to be?

[02:33:52] And the emphasis for me used to be on

[02:33:53] how to use 45 minutes. In the last year

[02:33:56] or so, it's really started to feel like

[02:33:57] we have 45 minutes together.

[02:34:01] And the together part is what's really

[02:34:03] mattering now when we can have screens

[02:34:04] involved. We can use AI, we should

[02:34:06] sometimes, but that that is a human

[02:34:08] space. that the classroom is taking on

[02:34:10] an almost sacred dimension for me now

[02:34:12] where it's people gathering together to

[02:34:14] be young and human together and grow up

[02:34:16] together and learn to argue in a very

[02:34:19] complicated country together and I think

[02:34:21] that is increasingly a space that

[02:34:23] education should be exploring in

[02:34:24] addition to pedagogy and content

[02:34:27] that's really powerful yeah

[02:34:30] I remember yesterday Kendall from

[02:34:33] Cambridge Latin talked about how

[02:34:36] important the togetherness ness was

[02:34:38] becoming it was taking a priority to

[02:34:41] everything else being more relational is

[02:34:44] becoming more important in this moment

[02:34:46] in her classroom. Um I'm curious about

[02:34:49] what it is that you each need um in this

[02:34:53] age of AI to support your your own

[02:34:57] learning to support your teaching. What

[02:34:59] is it that you need and need access to?

[02:35:04] That's a good question. Um, I think as

[02:35:07] as far as like nonprofits, we need the

[02:35:11] actual individuals, right, the makers of

[02:35:13] these tools to also be equally on the

[02:35:16] ground as much as educators are. I think

[02:35:18] often time like we think, okay, we'll

[02:35:21] just give in a few training sessions,

[02:35:23] but like obviously that's not enough,

[02:35:26] right? I think there needs to be a more

[02:35:27] hands-on evolvement and collaboration

[02:35:30] that goes into place uh when we think

[02:35:32] about the developers of these tools. uh

[02:35:35] with the collaboration of the educators

[02:35:37] who are den for placed in front of

[02:35:39] students to teach them and educate them.

[02:35:41] So I think it needs to be that level of

[02:35:43] collaboration

[02:35:44] um that is involved if we really want to

[02:35:46] ensure that our students are utilizing

[02:35:49] AI to work for them. But also as like

[02:35:52] Mike said these coaches, right? Um these

[02:35:55] thought partners that we want AI to

[02:35:57] actually be for us rather than this idea

[02:35:59] of just kind of like using AI to do all

[02:36:03] of the work, but actually teaching AI

[02:36:05] what you want it to do.

[02:36:08] My initial reaction was to say, I need

[02:36:10] someone to teach my seven-year-old how

[02:36:12] to spot a fake Taylor Swift ad on

[02:36:14] YouTube because I can't do that. Um, but

[02:36:18] my I guess my real answer is um that I

[02:36:21] think you know six years into working as

[02:36:25] an adult ally for the student voice team

[02:36:27] in a state like Kentucky that's smaller,

[02:36:29] we've made some really big strides in in

[02:36:31] ter in terms of ensuring that students

[02:36:34] are at the policym table when it comes

[02:36:36] to students on school boards on Kentucky

[02:36:39] Board of Education. Um, and it's it's re

[02:36:42] even though it's just one student and

[02:36:44] then you know you ensure they have

[02:36:45] voting power in these things. These are

[02:36:46] small levers of which you try to ensure

[02:36:48] that student voice is at the table. And

[02:36:50] I feel like throughout the day we keep

[02:36:51] hearing like it's really important. The

[02:36:54] the decisions are made about schools

[02:36:56] need to be made with students. Um and so

[02:36:59] I think that that would be the one thing

[02:37:00] I would say is in all of us in our small

[02:37:02] bits of the world to try to work to

[02:37:05] amplify the student voice.

[02:37:07] And I think your seven-year-old's going

[02:37:08] to be okay. Did you hear all the young

[02:37:09] people yesterday were telling us stories

[02:37:11] that they're now doing all of this

[02:37:13] quality control for their parents?

[02:37:14] they're having to tell their parents

[02:37:16] that's fake mom, that's fake dad, that's

[02:37:18] not real.

[02:37:21] So now the tables have turned, right?

[02:37:24] Um I think it's always taken a village

[02:37:26] to raise a child and I think we need

[02:37:28] that village a lot right now. Uh for

[02:37:31] example, the fact we're able to partner

[02:37:33] with Dr. Vassel and explore we have an

[02:37:35] idea about AI literacy. We have a chance

[02:37:37] to pilot it, but let's get a researcher

[02:37:39] to look at how what's happening here.

[02:37:41] You know, again, I was an English

[02:37:42] teacher for a long time and I've I've

[02:37:44] leveled up my AI ability, but I'm an AI

[02:37:46] autodidact. And it would be amazing if

[02:37:48] also somebody from industry came in and

[02:37:50] trained us in the technical sides of

[02:37:52] this because educators are coming as

[02:37:54] educators, not as tech people,

[02:37:56] nonprofits being able to fund these

[02:37:57] kinds of research opportunities. So I

[02:37:59] think if we're all gathering together

[02:38:00] and then crucially to your point making

[02:38:02] space for student voice whether that's

[02:38:04] surveys whether that's interviews

[02:38:06] whether that's looking at the sort of

[02:38:07] voice that emerges from how they use the

[02:38:09] chatbot and looking at the sort of

[02:38:11] breadcrumbs there but I think this is

[02:38:13] the world that we are shaping for them

[02:38:16] and they have a right to shape it too

[02:38:17] and to have a sense that they have

[02:38:19] agency in this moment.

[02:38:21] Yeah, thank you so much for that. So,

[02:38:23] what are you excited about for the

[02:38:27] future of AI and teaching and learning?

[02:38:30] What excites you the most? Where are the

[02:38:32] possibilities and opportunities? What

[02:38:34] are you dreaming about, thinking about

[02:38:37] with your students?

[02:38:42] I think personally for me um knowing

[02:38:46] that specifically STEM from dance our

[02:38:49] entire mission and vision is to really

[02:38:51] ensure that uh girls of all backgrounds

[02:38:55] right have this access to STEM fields. I

[02:38:58] think that's what excites me the most

[02:39:00] when I think about AI right now is like

[02:39:02] really ensuring that these students have

[02:39:04] this level of access. They have the

[02:39:06] tools uh they have the knowledge and

[02:39:09] resources to really gain um access and

[02:39:12] leeway which you know something that

[02:39:14] didn't have opportunity to do many years

[02:39:16] ago or especially women um didn't have

[02:39:19] that same level of access many years

[02:39:20] ago. I think it's like this new

[02:39:24] range of STEM, right? It's not like yes,

[02:39:27] just coding. It's like this new whole

[02:39:29] genre, right? That students can now also

[02:39:31] be experts in um to allow them that

[02:39:34] access and just overall level into these

[02:39:37] real world jobs where they also too can

[02:39:39] be affluent in.

[02:39:41] Yes. I think I would lean a little bit

[02:39:44] on the last speaker's presentation where

[02:39:46] we talked about creativity versus

[02:39:48] hindrance. And I could say I'm hopeful

[02:39:50] around the potential for it to create

[02:39:52] more time and space individually and

[02:39:54] collectively for creativity amongst our

[02:39:57] young people. So in whatever discipline,

[02:39:59] in whatever field, I think we schools

[02:40:02] can sometimes do a good job of not

[02:40:05] encouraging as much creativity um

[02:40:07] because of all the constraints. And so I

[02:40:09] think if we can shepherd AI toward

[02:40:14] supporting the production of more

[02:40:16] creative time for all of us that would

[02:40:17] be helpful.

[02:40:18] Did some of the data in your

[02:40:20] participatory action research study with

[02:40:22] your students yield something around

[02:40:25] that. It it does um because they are

[02:40:28] concerned that so far it's saying it and

[02:40:31] that they are concerned that it is

[02:40:34] they're openly saying it's concerned

[02:40:35] that it's hurting all of our collective

[02:40:37] critical thinking and um and ability to

[02:40:40] sort of work together. And so I think I

[02:40:43] actually think we can sort of take back

[02:40:44] that narrative also because I think that

[02:40:46] that is a real fear. I have that fear as

[02:40:49] well. But we also know that it doesn't

[02:40:51] it's a tool that's humanmade that can

[02:40:53] that we can shape and we are part of the

[02:40:55] design process. So I think thinking

[02:40:57] about how to push against that as a

[02:40:59] system um is what we ought to be doing.

[02:41:02] Yeah.

[02:41:03] Thanks.

[02:41:04] I would say uh on a practical level the

[02:41:06] personalization that AI provides in

[02:41:08] classrooms. I mean I I teach a career

[02:41:10] exploration class so it's especially

[02:41:12] useful there. But I I can have 34

[02:41:14] students in 10 minutes reflect on their

[02:41:16] individual sense of purpose with an AI

[02:41:18] chatbot that I design, close the laptop,

[02:41:21] turn to the person next to them, talk

[02:41:22] about their reflections, and then embark

[02:41:24] on a project in one period. That that's

[02:41:27] not normally how that works. And so that

[02:41:29] is really powerful. But I think what's

[02:41:31] even more exciting to me is educators,

[02:41:33] as many of you know, we're constantly

[02:41:35] trying to convince students of the

[02:41:36] relevance of what we're what we're there

[02:41:38] to teach. And you don't have to do that

[02:41:41] right now. they know this is relevant

[02:41:43] and so the relevance is there and so now

[02:41:45] it's more like we are living through a

[02:41:46] historical moment together us and you

[02:41:49] and let's try to make sense of this and

[02:41:51] put you in a position to succeed at this

[02:41:52] time.

[02:41:53] I love that. That's such a powerful

[02:41:55] thought um that you're living through

[02:41:57] this historical moment together and

[02:41:58] you're making sense of it together. That

[02:42:00] co-learning with your students is very

[02:42:02] powerful. So, we really wanted to make

[02:42:05] sure that we had time for Q&A. Um, that

[02:42:08] we've been sort of promising all

[02:42:09] morning, but now we're really going to

[02:42:10] have Q&A. We have microphones in the

[02:42:13] aisles, and if anyone has a question for

[02:42:16] our incredible panel of educators, we'd

[02:42:19] love to hear those questions and

[02:42:20] entertain them.

[02:54:46] educator for joining us. Thank you,

[02:54:48] Michael, Daniel.

[02:54:50] Appreciate you. Thank you.

[02:54:58] Oh, Kes,

[02:55:01] you're you're going without intros.

[02:55:03] Perfect.

[02:55:04] We got from 12 to 12:05. Um, so, uh,

[02:55:07] right before lunch, I wanted to just

[02:55:09] call your attention to in a lot of the

[02:55:11] panels today, there's been an issue that

[02:55:13] has come up over and over and over

[02:55:15] again, and that is assessment. And so on

[02:55:18] January 29th, we uh had another a

[02:55:22] different conference uh that was a full

[02:55:24] day talking about the future of

[02:55:27] assessment,

[02:55:28] responsible assessment in the age of AI.

[02:55:31] And that was a collaboration between two

[02:55:34] institutions that were sort of at the

[02:55:36] founding of educational measurement uh

[02:55:39] Stanford University and ETSs. And uh so

[02:55:43] I serve on the board of trustees at ETSs

[02:55:46] and I'm also a professor in the graduate

[02:55:48] school of education here at Stanford. So

[02:55:50] I wanted to bring those two uh important

[02:55:53] institutions and researchers together to

[02:55:55] talk about how do we do responsible

[02:55:58] assessment in this era of AI. And I will

[02:56:02] just say and Jesse will talk about some

[02:56:04] of the outcomes. I will just say the

[02:56:06] motivation for this was as somebody said

[02:56:10] uh the way we used to think about

[02:56:11] fairness and comparability was

[02:56:14] standardization

[02:56:16] and what we know now is standardization

[02:56:19] does not work for creating socially

[02:56:23] culturally responsive assessment that

[02:56:26] truly assesses what a student can do and

[02:56:28] what a student can learn. So now we have

[02:56:32] to rethink constructs. This is kind of

[02:56:35] the nerdy part. Constructs of fairness,

[02:56:40] comparability,

[02:56:42] validity, and reliability.

[02:56:46] What do those constructs mean in being

[02:56:49] socially and culturally responsive with

[02:56:52] the affordances of AI? So, I will let

[02:56:55] Jesse speak a little bit about some of

[02:56:56] the things that came out of it, but I

[02:56:57] will also tell you that we have a full

[02:57:00] white paper that is coming out. For

[02:57:02] those of you who are not able to uh

[02:57:04] participate, there'll be a full white

[02:57:06] paper coming out um in the next couple

[02:57:09] months on the on the conference. Jesse,

[02:57:12] yeah, thank you so much, Candace, and

[02:57:14] wonderful to see everyone. And wonderful

[02:57:15] to be back here at Stanford again two

[02:57:18] weeks later. have a little bit of deja

[02:57:20] vu walking around the space, but really

[02:57:22] excited to join you all today. And

[02:57:24] again, thank you Candace for inviting

[02:57:26] all of us to to come. Uh we had a

[02:57:29] wonderful meeting. We had some

[02:57:31] jam-packed panels where we talked about

[02:57:34] uh readiness for the age of AI and the

[02:57:36] skills that matter. We talked about

[02:57:38] socioulturally responsive and

[02:57:40] responsible assessment and how to do

[02:57:42] that. Well, we talked about um getting

[02:57:45] new sources of behavioral evidence to

[02:57:47] make inferences about the skills and

[02:57:49] knowledge that individuals have um using

[02:57:52] technology to do some of that. And then

[02:57:54] we also talked about responsible AI in

[02:57:57] both learning and assessment and how to

[02:57:59] achieve that. Um as Candace noted, we

[02:58:02] will be working on a white paper that

[02:58:03] will come out of this convening and it

[02:58:06] will discuss some of these themes in

[02:58:08] more depth. We had also a number of

[02:58:10] breakout sessions in the afternoon that

[02:58:12] were really quite interesting, gave

[02:58:14] folks more opportunity to engage to um

[02:58:17] have some back and forth and some good

[02:58:19] discussion. And I know the one that I

[02:58:20] sat in on was focused on K12 um

[02:58:23] assessment and learning and that was a

[02:58:25] pretty lively discussion that we had. Um

[02:58:27] folks built many connections and have

[02:58:30] opportunities to continue the

[02:58:31] conversation beyond the meeting. Uh so

[02:58:34] we know that many people have planned

[02:58:36] follow-up discussions.

[02:58:38] um further convenings and and other

[02:58:40] kinds of um ways to connect around the

[02:58:43] work that we do on learning and

[02:58:45] assessment. Uh another exciting thing

[02:58:47] from the meeting was that ETSs actually

[02:58:49] announced the launch of our center on

[02:58:53] responsible um AI for assessment and

[02:58:56] learning. And so we are launching this

[02:58:58] center trying to really engage in some

[02:59:00] deeper research along these themes.

[02:59:02] Although of course it's something that

[02:59:03] we've all kind of been working on for

[02:59:05] many years. Now we have a center uh

[02:59:07] where this work will be taking place and

[02:59:10] so I want to invite you all to again

[02:59:12] continue this conversation. So you've

[02:59:15] heard a lot about learning as well as

[02:59:16] assessment today. So especially if you

[02:59:18] are interested in the assessment piece

[02:59:21] that's ETSs that's what we do. So please

[02:59:24] do connect with me. I'm very happy to

[02:59:26] chat with folks um talk about potential

[02:59:29] collaborations and other opportunities.

[02:59:31] And so yeah I think that's what I have

[02:59:34] to say. Thank you. So, now we get to say

[02:59:36] go to lunch.

[02:59:41] Thank you,

[02:59:41] Candace.

[02:59:49] at the same time. But don't change when

[02:59:53] I break down

[02:59:56] the

[02:59:58] No, we don't have

[03:00:02] the same

[04:10:05] Welcome back. How was lunch?

[04:10:09] Yeah, was good.

[04:10:12] Uh, so I'm very excited to introduce the

[04:10:16] next session. Um, one of the reason why

[04:10:19] I am excited is because my boss is

[04:10:22] moderating the next session. So, I need

[04:10:25] to look really excited.

[04:10:27] Dean Schwartz uh will be moderating this

[04:10:31] incredible session with Miriam Rivera

[04:10:33] and Wendy Cup um with a global lens. So,

[04:10:38] we have a focus of morning a lot about

[04:10:41] the US. Uh this session is really meant

[04:10:43] to be thinking about AI in a global

[04:10:47] context. Okay.

[04:10:49] Um but let me let me say a few words

[04:10:52] about Dan Schwartz and embarrass him a

[04:10:54] little bit. Um

[04:10:57] so Dan Schwartz is our longest standing

[04:11:00] dean at the graduate school of

[04:11:02] education.

[04:11:03] He actually just announced that he was

[04:11:06] stepping down from the deenship uh at

[04:11:08] the end of June which is a big deal. uh

[04:11:12] because of all the things that he has

[04:11:13] accomplished for the school and for

[04:11:16] Stanford University uh including um you

[04:11:21] know this big new building that we are

[04:11:23] in which is one of the big

[04:11:25] accomplishments of Dean Schwartz. The

[04:11:29] other big accomplishments is obviously

[04:11:30] the Stanford accelerator for learning uh

[04:11:34] which didn't exist before Dean Schwartz

[04:11:37] um um invented and created this new

[04:11:40] initiative to connect research and

[04:11:43] practice.

[04:11:45] Um but also Dean Schwartz is um maybe my

[04:11:50] favorite piece in his bio is that he was

[04:11:54] a teacher

[04:11:55] and he was a teacher in multiple

[04:11:58] interesting locations. He started his

[04:12:00] career in southern Los Angeles uh in a

[04:12:04] tough neighborhood there and then moved

[04:12:07] to Alaska

[04:12:09] where he taught in a very small

[04:12:11] community um you know one of our rural

[04:12:14] communities in the US and then he also

[04:12:17] taught in Kenya

[04:12:19] which might be actually the most

[04:12:21] relevant for this session on global

[04:12:23] education here. So with no further ado,

[04:12:26] let me welcome uh Dean Schwartz, Wendy

[04:12:29] Cup, and Miriam Rivera. Thank you.

[04:12:52] I don't know.

[04:12:56] I'll see it. So, uh, yes. So, it it came

[04:13:01] out in the Stanford report that I'm

[04:13:02] stepping down. I'm still running the

[04:13:04] accelerator, but the report came out. It

[04:13:06] was very nice. It was my obituary. It

[04:13:09] was really weird. And and so people are

[04:13:11] saying things, you know, where I have to

[04:13:12] write back. I said, I'm not dying. I'm

[04:13:14] still here. But so, so I'm really

[04:13:17] fortunate. I have uh two huge change

[04:13:21] makers. And what I love about them both

[04:13:23] is they're helping other people to help

[04:13:25] other people. And so this is this is

[04:13:27] really uh spectacular. So uh each of you

[04:13:31] has kind of a different approach and uh

[04:13:34] so I'm I'm going to first ask what kind

[04:13:38] of what's the AI scene out there and

[04:13:41] then I'll probably follow up with is

[04:13:43] there something at stake and you can do

[04:13:46] both at the same time or not. But Wendy,

[04:13:48] I think you go first and let everybody

[04:13:50] know what you're doing. Gosh, what's the

[04:13:52] AI scene and what's at stake?

[04:13:55] Yeah, globally because because like I

[04:13:58] don't leave the sunny

[04:14:00] pal. I don't um

[04:14:03] Okay, so first of all, I guess you all

[04:14:05] know but just to be

[04:14:07] I am well I started Teach for America a

[04:14:11] long time ago, 37 years ago. Um

[04:14:16] thank you. Thank you. Um but about 18

[04:14:20] years ago, I started on this incredible

[04:14:23] journey to work alongside many others to

[04:14:26] build out the Teach for All network of

[04:14:29] organizations similar to Teach for

[04:14:31] America and now 63 countries and

[04:14:33] growing. Um inspired by just meeting

[04:14:36] people around the world from different

[04:14:38] countries who were determined that

[04:14:40] something similar needed to happen in

[04:14:41] their country. So, it's a network of all

[04:14:43] these independent, locally led

[04:14:46] organizations that are all working to um

[04:14:49] leverage this kind of approach that we

[04:14:51] share to develop what we've come to call

[04:14:53] collective leadership to ensure all kids

[04:14:56] fulfill their potential. Um so at the

[04:14:58] core of our work is the idea that people

[04:15:02] change systems and ultimately when you

[04:15:04] have enough people in a place working

[04:15:07] around the whole ecosystem around kids

[04:15:10] who share the same purpose to make sure

[04:15:12] the system enables all kids to thrive

[04:15:15] and are exercising agency towards that

[04:15:17] end and are working together to learn

[04:15:20] together and collaborate then you start

[04:15:23] seeing systems improve. So, um, so we're

[04:15:28] very people focused. Um,

[04:15:32] I think what is the AI scene? I don't

[04:15:34] know. That's such a lofty question. I'm

[04:15:35] not sure I'm prepared to answer that.

[04:15:37] But what I will say is having and I'm

[04:15:40] very curious what you'll say because I

[04:15:43] mean, you you probably have been even

[04:15:44] closer to this world certainly than I

[04:15:47] have. But I guess from watching the kind

[04:15:50] of edte space over the last two decades

[04:15:54] or so, um I mean we felt that we sort of

[04:15:59] missed in a way ensuring that the people

[04:16:03] driving edtech

[04:16:06] were teachers who are super committed to

[04:16:10] equity and have strong pedagogy. like

[04:16:14] where we saw

[04:16:16] teachers, equity focused folks committed

[04:16:20] to the transformation of education

[04:16:23] driving the change with technology. We

[04:16:25] saw really good things for kids, but by

[04:16:28] and large I don't know that that's what

[04:16:30] we have seen. And so with the advent of

[04:16:35] AI, what we're thinking about is how do

[04:16:38] we make sure that it's our our teachers

[04:16:42] and our educators, especially those

[04:16:44] working in the most marginalized

[04:16:46] communities who are at the forefront of

[04:16:48] driving how we utilize this. So that

[04:16:51] that's one thing that's very much on our

[04:16:54] minds. Is there is there any discussion

[04:16:57] about the AI replacing teachers or

[04:17:00] filling in where it's hard to get

[04:17:01] teachers?

[04:17:02] Are you hearing this globally?

[04:17:04] I mean there are all sorts of people of

[04:17:06] course who throw that out.

[04:17:07] I mean

[04:17:09] technology and AI amplifies what's

[04:17:13] there. I mean that's that's the lesson

[04:17:15] of the last however and I could give you

[04:17:18] some stunning examples of that. So where

[04:17:21] you have a school that's not on a

[04:17:24] mission and you enter technology into

[04:17:28] that school, it becomes this incredible

[04:17:30] amplifier of a lot of not good stuff and

[04:17:35] it becomes a huge source of distraction

[04:17:38] where you have a mission centered school

[04:17:42] with incredible school leadership and

[04:17:45] teachers and everyone's really working

[04:17:48] to equip our young people to shape a

[04:17:51] better future for themselves and all of

[04:17:52] us to develop them holistically.

[04:17:54] Everyone's on a mission. You enter AI or

[04:17:58] technology into that and it's like, man,

[04:18:02] it's incredible, right? Like every

[04:18:04] teacher has an assistant. You're using

[04:18:06] it. I mean, I've seen such incredible

[04:18:08] applications of AI already. I was in a

[04:18:11] school in Singapore like two weeks ago

[04:18:14] and the teachers had surveyed the kids

[04:18:18] asking them who their friends were. I

[04:18:20] mean this is just a minor thing but it

[04:18:21] just shows you like are you kidding?

[04:18:23] Like they used it to develop their

[04:18:26] seating charts. So they took the

[04:18:28] connector kids and the kids who actually

[04:18:30] didn't have any friends and I don't know

[04:18:32] exactly what they told the AI to do but

[04:18:35] they were so happy about this and they

[04:18:37] had done so many other things with with

[04:18:39] AI. So you can do so many good things

[04:18:42] when the tool is in the hands of good

[04:18:44] people. So my obsession is I think it's

[04:18:48] become

[04:18:50] I mean if we don't reset entirely as a

[04:18:54] society and decide to prioritize

[04:18:57] education completely differently and

[04:19:00] invest what we need to invest to inspire

[04:19:04] and develop

[04:19:06] our strongest talent as a country and as

[04:19:09] a world to create incredibly powerful

[04:19:13] educational environments. ments, I think

[04:19:15] we risk a tremendous amount because this

[04:19:18] tool could become a huge distraction.

[04:19:20] And also all these kids in our

[04:19:22] classrooms today, they got the power now

[04:19:25] because they've got AI. So we better be

[04:19:27] really, really, really determined to

[04:19:30] make sure that all of our kids get the

[04:19:32] kind of education that will enable them

[04:19:35] to make the world better and to improve

[04:19:38] our collective welfare rather than the

[04:19:40] opposite.

[04:19:42] Miriam, your turn. Uh so question one,

[04:19:46] what are you seeing out there? You you

[04:19:48] are provide funding for edtech and then

[04:19:52] uh is there anything in the balance?

[04:19:55] I think there's a lot in the balance and

[04:19:57] I have a lot of optimism and a lot of

[04:20:00] fear of where AI um can take us as a

[04:20:03] society. Um, Silicon Valley has had a

[04:20:06] long history of um really

[04:20:11] antisocial kinds of u movements

[04:20:14] including in the earliest days of the

[04:20:16] semiconductor um a real belief that

[04:20:19] there are just different classes of

[04:20:20] humans and some of them are better than

[04:20:22] others. Um I can see that happening uh

[04:20:25] with some of the uh technology uh

[04:20:28] champions in AI. Um, you know, there's

[04:20:31] this big conversation about when will

[04:20:33] the first billion-dollar company be

[04:20:35] created by one person. Um and that seems

[04:20:39] to be uh the goal I think of um how many

[04:20:44] people are thinking about it and you

[04:20:46] know even uh a real sense that removing

[04:20:51] some of the labor costs is going to make

[04:20:54] these technologies um so much more

[04:20:56] valuable than any other technology that

[04:20:58] has ever existed before. Uh and I

[04:21:02] certainly think that that is actually

[04:21:03] really possible. So we're looking at

[04:21:06] really a restructuring of um work, a

[04:21:10] restructuring of society as um people

[04:21:13] have of ability to use these tools and

[04:21:16] to be creators and to create value with

[04:21:20] uh technology. So on the social order I

[04:21:24] think it's a very huge um huge shift.

[04:21:28] I I come from like really a very humble

[04:21:33] family. My parents came as migrant farm

[04:21:35] workers from Puerto Rico that used to do

[04:21:36] the crops between Florida and New York.

[04:21:39] Um, one of the things that kind of made

[04:21:41] a difference in my life was that I was

[04:21:43] uh admitted to a private school where I

[04:21:45] had computer science for three years of

[04:21:47] my high school education between the

[04:21:50] late 70s and the early 80s. So, you

[04:21:52] know, when they talk about outliers,

[04:21:54] that's the kind of thing that can happen

[04:21:56] in your life that some random event

[04:21:58] happens that you then are on a wave um

[04:22:02] that you would never have been able to

[04:22:03] access without it. Um, and because of

[04:22:06] that, I also happen to have an

[04:22:09] internship in uh, computer program when

[04:22:13] the personal computer first came out.

[04:22:15] And I worked on the first generation

[04:22:17] search IPOs um, as an attorney when

[04:22:21] there were 39 million users on the

[04:22:23] worldwide web. So, hardly anybody was on

[04:22:25] it. Um and I think that for me this is a

[04:22:29] wave where I see a lot of optimism uh

[04:22:32] also possible because what this has done

[04:22:35] is actually reduced the cost of coding

[04:22:38] to practically zero. So that means that

[04:22:41] people who can create um using

[04:22:44] technology can help solve problems that

[04:22:46] they know about in a much more

[04:22:49] cost-effective way than has ever been

[04:22:51] possible at all. really um that is the

[04:22:54] opportunity I think in education and I

[04:22:58] hope that like when I think about Teach

[04:23:00] for America I hope it will be a group of

[04:23:02] young people I'm assuming most of them

[04:23:04] are in their 20s and like my daughter

[04:23:06] they're doing everything with AI around

[04:23:08] homework they're having books read to

[04:23:10] them while they're driving to Santa Cruz

[04:23:11] to go surfing so that they don't have to

[04:23:13] spend all that time in a you know

[04:23:15] reading a book um and they're uh really

[04:23:19] starting their drafts of everything um

[04:23:22] in interrogating an AI around the

[04:23:24] subject. And so they're really just

[04:23:26] learning in a very different way from

[04:23:28] how I think I did as you know I spent

[04:23:31] six hours a night at the library when I

[04:23:33] was at Stanford um trying to uh read all

[04:23:36] my law cases and business school cases

[04:23:38] etc. Um but I think that this is really

[04:23:41] uh the opportunity is for those who are

[04:23:45] trained to use AIS to create not for

[04:23:48] people who are taught to consume

[04:23:51] technology because I think that's really

[04:23:53] one of the things that is the difference

[04:23:54] between what I see happening a lot of

[04:23:57] the times in you know like in a private

[04:24:00] school education versus a public school

[04:24:02] education like people get access to

[04:24:04] computers but they're just doing

[04:24:05] different things and you know I have a

[04:24:07] daughter who's like using it to create

[04:24:10] um three-dimensional

[04:24:12] uh wood products or she's doing 3D

[04:24:16] printing with it. Um she's not a she's

[04:24:19] not a technical undergraduate, but she

[04:24:22] had computer science. She does actually

[04:24:24] learn how to produce things with it as

[04:24:26] opposed to just be on TikTok watching

[04:24:28] videos. So, I think there's a big um

[04:24:31] difference and opportunity. I'd love to

[04:24:33] see um the Teach for America kids like

[04:24:37] helping schools to really transform how

[04:24:39] their students are learning, how

[04:24:40] teachers are using AI so that they can

[04:24:43] actually get the most out of um

[04:24:45] education um and be creating with it.

[04:24:49] Well, I love that. I'm I'm a big fan of

[04:24:52] using AI to surpass yourself as opposed

[04:24:55] to being told what to do by the AI.

[04:24:59] So I'm I'm I'm this is this you're going

[04:25:02] to have to be patient because it's going

[04:25:03] to take me a long while to get into the

[04:25:05] actual question here. Uh it's just

[04:25:07] something I've I've observed uh in the

[04:25:09] last today but also yesterday and a few

[04:25:12] some other things and and I think it

[04:25:14] touches both of you have proposed that

[04:25:17] the teachers need to be involved and in

[04:25:19] your case the students need to be

[04:25:20] involved. They need to be wielding the

[04:25:22] AI.

[04:25:24] So I I want to make a slight contrast.

[04:25:26] So a lot of people talk about AI as

[04:25:27] doing personalization.

[04:25:29] And so generally the the model that they

[04:25:32] have in mind is there is a sort of this

[04:25:34] curriculum and personalization means you

[04:25:36] move forward or backward based on where

[04:25:39] the students at and then maybe you

[04:25:41] change the content from rainbows to

[04:25:43] volcanoes depending on what the kids

[04:25:45] interested in. So so that's one model.

[04:25:48] Uh and and it's a good model. It's

[04:25:51] useful for many things. But something

[04:25:53] that I'm seeing and I've heard here and

[04:25:55] I've seen in the projects and at

[04:25:57] Stanford students is it's more uh

[04:26:00] targeting demographics they've come

[04:26:02] from. So for example uh one proposal is

[04:26:06] to create something that translates into

[04:26:08] ASL sign language because they have

[04:26:11] experience or issues of uh different

[04:26:14] kinds of disabilities like ADHD or being

[04:26:17] on the spectrum or second language

[04:26:19] learners. And so here are all these

[04:26:22] students making tools for these these

[04:26:24] nit these smaller demographics that the

[04:26:26] market's probably not going to serve

[04:26:28] because there's not enough return. So so

[04:26:31] I find this quite interesting that this

[04:26:33] is what's happening that it that it is

[04:26:35] it's the everybody is using it to create

[04:26:38] the needs for themselves.

[04:26:40] Okay. So now I'm I'm gonna give you a

[04:26:42] question that probably just ignore and

[04:26:45] talk about what you want. But so so I

[04:26:48] don't I don't actually

[04:26:50] uh have the cat food, but I'm in the

[04:26:52] position where I get to distribute the

[04:26:53] cat food. And so, and this is a joke,

[04:26:56] how do you get faculty to do anything

[04:26:57] together? You move the dish of cat food

[04:26:59] and they all go there.

[04:27:02] So, so how should I here's a great

[04:27:04] university? What what should I do to get

[04:27:06] them to help more people trying to uh

[04:27:10] create in this space? Is it already

[04:27:13] happening? Do I just need to help them

[04:27:16] do it better? like what could I do to

[04:27:18] help your teachers

[04:27:20] design the AI systems for their

[04:27:22] missiondriven schools?

[04:27:26] And so university uh great talent, high

[04:27:29] energy uh insight often times use

[04:27:34] research not always.

[04:27:37] I mean what I would say okay so well our

[04:27:42] approach at the moment to try to do

[04:27:45] differently than what we did 20 years

[04:27:47] ago with technology and education is to

[04:27:50] just do everything we can to get

[04:27:52] teachers using this in their classrooms

[04:27:57] you know despite all the challenges

[04:27:59] because they are 60% of them in places

[04:28:02] with no connectivity. So, it's like how

[04:28:03] do we we can't let that stop us. Like,

[04:28:06] if they're not going to use it during

[04:28:08] their time in these classrooms, then

[04:28:10] they won't become the leaders needed to

[04:28:12] drive the change. So, we're doing

[04:28:14] everything we can to foster like help

[04:28:17] them know how to use it, foster their

[04:28:19] experimentation, foster their learning

[04:28:21] with each other. Um, and then we're

[04:28:24] we're assuming that's going to generate

[04:28:26] lots of cool innovations like those you

[04:28:28] just mentioned and we'll get behind them

[04:28:30] and try to support their their

[04:28:32] leadership. Um, but what I would say is

[04:28:36] that the first question because I'm

[04:28:38] really thinking about your question like

[04:28:39] what can Stanford do? I mean first of

[04:28:41] all you can contribute to their learning

[04:28:44] and and you know there's probably a lot

[04:28:46] lot we could do to engage faculty and

[04:28:49] students here with those communities of

[04:28:51] practice that you know we have a

[04:28:52] thousand teachers in these communities

[04:28:54] at the moment etc. Um, but I always

[04:28:58] think about the fact that I mean the

[04:29:00] question I get now in every room is how

[04:29:02] will AI change teaching and learning?

[04:29:04] And it it's it's an important question

[04:29:07] but it's not the first question. The

[04:29:11] first question and a question that for

[04:29:13] some reason still we are not asking

[04:29:16] everywhere in every school in every

[04:29:19] community in every country at the global

[04:29:22] level is what are we working towards for

[04:29:25] kids like in today's world because you

[04:29:28] know we have educators showing up at

[04:29:30] schools that were built 150 200 years

[04:29:33] ago however many in in different

[04:29:35] countries and and I mean it's the one

[04:29:37] thing that hasn't changed at all right

[04:29:39] we haven't rethought the very purpose of

[04:29:42] education for today. We need everyone

[04:29:46] everywhere considering the question of

[04:29:48] like given the values and histories in

[04:29:51] any given place and given the challenges

[04:29:56] facing kids in that place and the

[04:29:58] opportunities they have in in that place

[04:30:00] and given our global aspirations,

[04:30:03] what are we trying to accomplish? like

[04:30:05] what are the contextualized like if if

[04:30:08] we can embrace that I mean we need to

[04:30:10] move from a paradigm that we've had

[04:30:12] which is around individual attainment of

[04:30:14] academic results I mean of course there

[04:30:17] are the islands of incredible schools

[04:30:19] like some of you probably have your kids

[04:30:21] in but the vast majority of kids show up

[04:30:24] at schools that were created in that

[04:30:26] paradigm and we need to evolve to a

[04:30:28] different paradigm right like holistic

[04:30:31] student development for our collective

[04:30:33] welfare unless Unless we start there and

[04:30:37] then say, "Okay, now how can we use all

[04:30:39] the tools at our disposal to get where

[04:30:42] we're trying to go for kids, we'll just

[04:30:45] be tinkering around the edges again, and

[04:30:47] it won't transform anything really."

[04:30:50] So, uh,

[04:30:50] so Stanford can help propagate the right

[04:30:54] question,

[04:30:55] which you're probably doing already at

[04:30:57] some stage, but it hasn't taken over the

[04:30:59] world. Like, we're not rethinking the

[04:31:00] purpose of education to the extent that

[04:31:02] we need to. Oh, I think I have the right

[04:31:04] question. Nobody will listen, but uh no,

[04:31:08] I'm not I'm not going to tell you what

[04:31:09] it is. Uh no, you know, I've actually

[04:31:11] I've actually noticed a change. Uh so,

[04:31:14] you know, schools will have these are

[04:31:16] our values or this is what we want our

[04:31:18] students to be when they graduate. And

[04:31:20] and a funny change is they now all

[04:31:22] include things like great empathizer,

[04:31:25] collaborator, resilient. I don't see

[04:31:28] math English proficiency on those

[04:31:30] anymore. So there has been a movement

[04:31:32] towards these more durable skills. You

[04:31:34] know, it may just be a Bay Area thing. I

[04:31:36] don't get out.

[04:31:36] I think there's been a big movement. I

[04:31:38] mean, if I look at now versus 10 years

[04:31:40] ago, like what are we talking about at

[04:31:42] the global level? Like there's there's

[04:31:44] been a movement, but we are far from

[04:31:48] communities everywhere stepping up from

[04:31:52] today's conception of school with the

[04:31:54] diverse stakeholders who need to do

[04:31:55] that. It has to include families like

[04:31:59] parents and policy makers and the

[04:32:02] employers and the educators all together

[04:32:06] asking themselves in this community what

[04:32:09] are we going to work towards for kids or

[04:32:10] else things will not fundamentally

[04:32:12] change because you'll still have parents

[04:32:14] thinking and and then we need to go to

[04:32:16] the next step which is help all of us

[04:32:18] adults in the system unlearn how we were

[04:32:20] taught and learn a different way of

[04:32:22] viewing what education should be and how

[04:32:24] we view students and how we view

[04:32:26] teachers and this is a big huge shift.

[04:32:29] Yeah. No, I it I wish

[04:32:30] without that once we do that AI is going

[04:32:33] to answer itself. If we just embrace AI

[04:32:36] without doing that, we're going to still

[04:32:38] be here looking at the same schools with

[04:32:40] probably more inequity and more

[04:32:44] let's see.

[04:32:46] You had the right question.

[04:32:49] Uh, so I do I do wish this had happened

[04:32:52] before I chose to go into teaching. And

[04:32:54] my parents response was, "Really? That's

[04:32:56] what we pay for a college education as

[04:32:59] you become a teacher?"

[04:33:00] That's still everyone's parents

[04:33:02] response.

[04:33:04] Miriam, what do you what what can a

[04:33:06] university do to help? We're smart

[04:33:08] people, hardworking.

[04:33:11] Uh, Stanford's particularly game to be

[04:33:14] innovative.

[04:33:16] the well let me let me go to one of the

[04:33:19] points that you're making about like

[04:33:21] what is education like um for the

[04:33:23] students and what kind of education

[04:33:25] makes a difference um I was a head start

[04:33:28] kid um head start made a big difference

[04:33:30] in my life I didn't even speak English

[04:33:32] when I went to head start right um and

[04:33:34] part of it is um that engagement with uh

[04:33:38] learning started in that room and I also

[04:33:42] had a diverse teacher and maybe one of

[04:33:44] the fewer diverse teachers I would ever

[04:33:46] have actually in my whole entire

[04:33:48] education um in Head Start. Uh the other

[04:33:52] thing that I think was so important was

[04:33:55] um there was a I had just generally good

[04:33:58] teachers throughout my uh public school

[04:34:01] education in Chicago public schools. Uh

[04:34:04] but I came to a new um middle school

[04:34:09] that actually had a gifted program at

[04:34:11] that time in a very lowincome community.

[04:34:14] And uh you know I had this one amazing

[04:34:18] teacher, Miss Weinstein. He literally

[04:34:20] bought like secondhand books like Grapes

[04:34:23] of Wrath, uh, Up Upton Sinclair's

[04:34:26] Jungle, uh, Oliver Twist, and she bought

[04:34:31] all these copies and we got our books

[04:34:33] right from her. And we used those books

[04:34:37] to both learn how to write, to

[04:34:40] understand the vocabulary, to learn how

[04:34:42] to spell, um, and then to learn how to

[04:34:46] engage in discussion and

[04:34:48] self-reflection. Uh, so it was about as

[04:34:51] cheap of an education as you really

[04:34:52] could imagine, right? Like the cost of

[04:34:55] 30 um copies of these paperback Penguin

[04:34:59] books was probably like a buck each and

[04:35:01] you know the teacher getting us to

[04:35:03] actually talk and think was very

[04:35:05] inexpensive. Um, but it was profoundly

[04:35:10] um important in terms of um my personal

[04:35:14] growth as a human being. Um that was the

[04:35:17] first time I ever heard the word the

[04:35:18] human condition is in sixth grade and um

[04:35:21] and had to really understand like the

[04:35:23] condition of poverty was part of the

[04:35:25] what she was having us um read about and

[04:35:31] talk about when we're talking about like

[04:35:32] Oliver Twist and Fagan and what response

[04:35:35] to poverty Oliver Twist would have and

[04:35:38] would he want more and should he get

[04:35:40] more? That was just such a powerful

[04:35:43] thing. And you know, my seventh grade

[04:35:46] curriculum was focused around um plays

[04:35:49] actually. Um and we did a variety of

[04:35:52] classic American plays, Shakespeare, um

[04:35:55] just a lot of different things. And we

[04:35:58] happened to have a joint teacher program

[04:36:01] where there was a male instructor who

[04:36:03] was also a method actor in the evenings

[04:36:07] um but a Chicago public school teacher

[04:36:09] during the day and then um an English

[04:36:12] teacher as well. And that was another

[04:36:17] thing that I saw boys have a completely

[04:36:20] different frame around education and

[04:36:23] masculinity um than they would have been

[04:36:25] exposed to um but for that class. Um

[04:36:29] again, you know, like primary sources,

[04:36:32] great teachers and a room. That's really

[04:36:34] what we had. Um and I think that really

[04:36:37] fundamentally that's what education

[04:36:38] really is. Um and that um we

[04:36:44] don't want technologies to kind of get

[04:36:48] at the um core function of education,

[04:36:52] which I would argue is not um it's about

[04:36:56] like being able to do self-reflection,

[04:36:58] being able to um understand one's own

[04:37:02] emotion and self-manage. um being able

[04:37:04] to express that emotion in constructive

[04:37:06] ways like these boys were learning in um

[04:37:09] this example of watching a a very

[04:37:11] muscle-bound actor um teach how to get

[04:37:15] in touch with your emotions so you could

[04:37:17] actually cry on a stage. Um so

[04:37:22] uh education I think is what's really

[04:37:25] happening between the people and then

[04:37:27] the skills that you're acquiring in

[04:37:29] education I think are things for which

[04:37:31] technology can be very useful right like

[04:37:33] it's it's good to have a pen and a piece

[04:37:35] of paper um it's good to use a computer

[04:37:38] um those are tools that we use to

[04:37:40] actually um work on the skills that we

[04:37:43] need to have like the writing the

[04:37:45] reading the vocabulary the spelling um

[04:37:48] And I think that we can get a lot of

[04:37:50] that without necessarily using a lot of

[04:37:53] the precious time that the people have

[04:37:55] in that room with those primary sources

[04:37:57] and teachers and students um to help uh

[04:38:02] give them skills that they may not have

[04:38:04] at home. Uh my parents were like my

[04:38:07] father had a high school education, my

[04:38:08] mother had a sixth grade education. Um

[04:38:11] this was not somebody who was going to

[04:38:12] help me with my homework, right? Um, so

[04:38:14] I I learned other things from my parents

[04:38:16] that I couldn't get in school, but I

[04:38:18] needed to get a lot of the skills I

[04:38:20] needed for life, um, from my teachers

[04:38:22] and my school, um, my jobs, those kinds

[04:38:26] of things. And I feel like, um, we never

[04:38:29] should get away from like what is

[04:38:31] actually happening in the classroom that

[04:38:33] is really education. Um, because

[04:38:35] education isn't just about the

[04:38:37] intellect, it's also about the heart.

[04:38:40] So, I I I've got a Yeah, both of these

[04:38:43] answers were superb. Thank you. Uh you

[04:38:46] didn't help that much. They're hard. Uh

[04:38:49] so so I could do a seed grant that says,

[04:38:53] "What are the most important questions

[04:38:55] about what children should become

[04:38:58] that goes beyond getting right and wrong

[04:39:01] answers, something like that. And then I

[04:39:04] give them extra money if they have a

[04:39:06] solution or a way to measure it.

[04:39:09] No,

[04:39:10] I don't know. I I I'm realizing like I'm

[04:39:13] here at Stanford and you're asking what

[04:39:15] can Stanford do? So, I can't resist

[04:39:17] saying one more thing.

[04:39:20] Do it.

[04:39:21] Although I know this is

[04:39:22] as long as it's not as long as it's not

[04:39:23] mean. So you can't particularly maybe be

[04:39:26] the one to do this, but um I think we

[04:39:29] should I think Stanford should should

[04:39:32] and so should other universities like

[04:39:35] really step back and consider given the

[04:39:38] disruption that's happening right now

[04:39:40] thanks to AI in entrylevel jobs.

[04:39:45] How do we create a norm shift in first

[04:39:48] job choices?

[04:39:50] Can we stop funneling kids from the

[04:39:53] minute they step foot on this campus

[04:39:55] into a very narrow set of corporate jobs

[04:39:58] and instead say first put yourself on

[04:40:03] the front lines of the reality in your

[04:40:05] country? Teach in an underresourced

[04:40:08] school. Work as an EMT. Serve as a case

[04:40:12] manager working with unhoused folks. do

[04:40:16] any number of other things because we're

[04:40:19] going to need your human skills. Like

[04:40:22] those first jobs, what you do shapes who

[04:40:26] you become. And we're seeing fewer and

[04:40:29] fewer of the Stanford students and other

[04:40:32] students around the country who go to

[04:40:34] these selective colleges. I mean, more

[04:40:36] and more of them just put themselves in

[04:40:38] skyscrapers and fewer and fewer are

[04:40:40] putting themselves in front of reality.

[04:40:43] And it's a real risk to our country.

[04:40:45] like I think it's one of our biggest

[04:40:47] issues and we need to tackle it

[04:40:50] aggressively right now when we have this

[04:40:53] incredible window of opportunity. So, a

[04:40:56] fabulous point. Uh, you know, Stanford's

[04:40:58] working on this. We're doing lots of

[04:41:00] stuff. Unfortunately, this is a session

[04:41:02] on AI.

[04:41:03] I know. Bring it back. Exactly. No, you

[04:41:05] can do it. You can I'm gonna go to

[04:41:07] Miriam. Get back to AI.

[04:41:09] No, work on it while I talk to Miriam

[04:41:11] for a second.

[04:41:11] I do think though that the students, for

[04:41:14] example, are the ones that are the most

[04:41:17] fil with this technology of anybody. Um,

[04:41:20] so it is really important for them to be

[04:41:22] in different spaces. And I would say

[04:41:25] that um Stanford does a I think a pretty

[04:41:29] good job of getting people to on a

[04:41:31] volunteer basis um volunteer in

[04:41:33] communities. And I know throughout my

[04:41:36] entire life as an adult, like I've

[04:41:39] always served in some sort of

[04:41:41] volunteerism around education um and in

[04:41:44] other communities and uh and so I I

[04:41:49] consider that that is one way that we do

[04:41:51] it. And I know at different times um

[04:41:53] we've actually very actively promoted

[04:41:55] public uh service careers and

[04:41:58] internships because I worked in that

[04:42:00] program when I was uh just a recent

[04:42:02] graduate from Stanford and I think we

[04:42:04] helped a lot of people actually access

[04:42:05] those kind of opportunities. So I think

[04:42:07] that we do a good job of trying to also

[04:42:11] fund students so they can do those kinds

[04:42:13] of things um even if they can't get paid

[04:42:16] by the not for profofit or the not for

[04:42:17] profofit only has limited amount ability

[04:42:19] to pay them. So I I I think that that is

[04:42:23] a continuing way in which we should um

[04:42:26] serve our communities. Um I I think that

[04:42:29] we actually need to put the more

[04:42:31] technically savvy students out there.

[04:42:34] That's a different um constituency that

[04:42:37] I don't think we quite um enroll in the

[04:42:40] same way in some of these programs

[04:42:41] because they're being really I think

[04:42:44] very much funneled into um

[04:42:47] entrepreneurship really early. But I

[04:42:49] think which I think is a great place for

[04:42:51] people obviously um and I invest in a

[04:42:54] lot of people that have uh technology

[04:42:56] backgrounds that are starting um

[04:42:58] education technology companies and other

[04:43:00] companies as well. Um but my sense is I

[04:43:04] totally agree like there is such a

[04:43:06] bifurcation in our country about um how

[04:43:12] people know each other. Um you know I

[04:43:14] used to say that church was the most

[04:43:16] segregated hour in America. Um, I just

[04:43:19] think that we've just gotten more hours

[04:43:20] segregated in America. Um, and that

[04:43:23] people often are um only interacting

[04:43:26] with people in their same class, race,

[04:43:28] level of education. Sometimes I've like

[04:43:31] I had a party one time and I thought, my

[04:43:32] god, everybody here has a master's

[04:43:34] degree at least. You know, that's just

[04:43:35] not the real world. Um, and so, and I'm

[04:43:39] fortunate in that because of my life

[04:43:42] history, that's not the only world that

[04:43:45] I inhabit. But I think for many of us

[04:43:48] and our students here um that is the

[04:43:50] world that they primarily inhabit and

[04:43:52] they have very little exposure um to uh

[04:43:56] to the real world and to the real needs

[04:43:58] of a lot of Americans, the majority of

[04:44:01] whom are in financial situations that

[04:44:05] don't allow them to have a $400

[04:44:07] emergency. You know, if like their car

[04:44:09] breaks down, that can like really push

[04:44:11] them over the edge. If um COVID happens,

[04:44:15] they become homeless. I mean, this is

[04:44:17] this is not um like this is my family,

[04:44:20] right? Like I support three siblings in

[04:44:22] part because those kinds of things have

[04:44:24] happened to them. Um and they've grown

[04:44:27] up in environments where um they've

[04:44:30] become subject to drug addiction or

[04:44:32] other kinds of um issues that are really

[04:44:35] happening because I think people do lose

[04:44:37] hope um in in our communities because

[04:44:40] they're so isolated from the opportunity

[04:44:42] that this country represents. And so I

[04:44:46] would um encourage us to continue to um

[04:44:50] encourage students to like uh do

[04:44:53] volunteerism to be of service in local

[04:44:57] communities to look at internships and

[04:45:00] careers in these spaces and to fund

[04:45:02] grants. So I've got it. So what we do is

[04:45:04] we do internships for students and they

[04:45:07] it's like service learning. They need to

[04:45:09] spend good amount of time in the

[04:45:10] community and then they have to create

[04:45:12] AI apps for the community.

[04:45:14] Yeah. Did did I get the AI

[04:45:16] or or help the community build apps

[04:45:20] around the needs of that community?

[04:45:22] Because I think the understanding what

[04:45:25] I'm trying to get at with saying that

[04:45:26] the cost of coding is free is that now

[04:45:29] the experience of people who actually

[04:45:32] are aware of a problem and have ideas

[04:45:35] for how to solve that problem are much

[04:45:37] more valuable. We're seeing this in the

[04:45:40] enterprise when we talk about like vibe

[04:45:41] coding for the enterprise. For example,

[04:45:44] when I was at Google, I created a ton of

[04:45:46] applications for my legal department.

[04:45:48] Um, some of them were very simple things

[04:45:50] like this was before like docuign didn't

[04:45:52] exist yet. So, we created a little thing

[04:45:54] that could sign things. Um, we created

[04:45:56] another thing that could do uh

[04:45:57] non-disclosure agreements. Uh, they were

[04:46:00] thousands of them being made because we

[04:46:02] made it available so easily and you

[04:46:04] didn't have to go through a lawyer and

[04:46:04] you didn't have to take time of a

[04:46:06] lawyer. Um, and we created things that

[04:46:08] were like approval processes for getting

[04:46:10] a contract signed that was used for like

[04:46:13] the next 15 years after I left. So, I

[04:46:16] think that this knowledge that the

[04:46:18] community has combined with the

[04:46:20] technical skills that others can bring

[04:46:22] to it will solve a lot more problems.

[04:46:25] Uh so an added bonus of this is uh

[04:46:28] there's there's something called the haw

[04:46:30] center and students go out and do

[04:46:31] service learning and this is the g

[04:46:34] gateway drug for the undergraduates to

[04:46:35] come to the school of education. They

[04:46:37] they get to see the experience of

[04:46:39] service and they know education is a

[04:46:41] serviceoriented thing. So I I I got one

[04:46:44] last bucket of questions uh and and this

[04:46:47] is about partnerships. So um the I find

[04:46:51] the education sector particularly

[04:46:53] fractured very difficult to get things

[04:46:55] out the door and into the place. So so

[04:46:58] that's kind of the intent of this this

[04:47:01] large convening. we have these different

[04:47:03] sectors here and uh when we first did it

[04:47:06] um I was thinking oh just people here

[04:47:09] just talking you know nothing's

[04:47:11] happening they're just talking you know

[04:47:13] and then I I was I I realized uh it it

[04:47:16] was like my life I was raised on the

[04:47:18] beach in Southern California and if you

[04:47:21] see someone that you'd never seen before

[04:47:23] the first time you see him you you just

[04:47:25] don't say anything right the second time

[04:47:28] you see him you go hey right third time

[04:47:31] you see you start to have conversations,

[04:47:33] right? And then the fourth time you're

[04:47:34] actually playing volleyball together. So

[04:47:36] that that's my hope. Here we are in year

[04:47:38] four. So I believe you two probably have

[04:47:42] thoughts about how to make these kinds

[04:47:44] of cross- sector partnerships. I believe

[04:47:46] you may be collaborating with anthropic.

[04:47:49] So how did that come about? Like are

[04:47:51] there lessons learned for us?

[04:47:54] Um well first of all I couldn't agree

[04:47:56] more with what how you described Yeah.

[04:47:59] the beach.

[04:48:00] Like honestly, it's so underappreciated

[04:48:04] convening like creating the space that

[04:48:06] enables people to build the

[04:48:08] relationships to have the debates. I I

[04:48:10] think it's

[04:48:11] honestly it's it's kind of the missing

[04:48:13] piece to like accelerate progress. So

[04:48:16] we're we're actually sorry we're sort of

[04:48:19] unique at that that uh Stanford's sort

[04:48:22] of a trusted convenor

[04:48:24] and so people come like if I had an NSF

[04:48:26] research grant I would only bring

[04:48:27] researchers or if I go to ASUGSV it's

[04:48:30] mainly entrepreneurs whereas here we can

[04:48:32] bring them

[04:48:32] that's very cool so it's a unique

[04:48:35] um yeah we have this partnership with

[04:48:37] anthropic again it was motivated by our

[04:48:41] belief that we need teachers who are

[04:48:44] equity focused focused on actually

[04:48:46] transforming our education system,

[04:48:48] driving the change. Um, they are helping

[04:48:52] to train all these teachers and educate

[04:48:54] them on how to maximize AI in their

[04:48:57] classrooms. Um, and they're in the

[04:48:59] WhatsApp groups, you know, we've got all

[04:49:02] these different groups, different

[04:49:03] languages, etc., learning from them. So,

[04:49:06] they're seeing the challenges these

[04:49:08] teachers are facing. Like if they're

[04:49:10] teaching in Oaken State, Nigeria, and

[04:49:12] are like everything I can access is

[04:49:15] English medium and rooted in a different

[04:49:17] culture, like it's good for those

[04:49:20] anthropic executives to be like, "Okay,

[04:49:22] actually this is going to go nowhere

[04:49:23] unless we solve that problem." I mean,

[04:49:25] they know intellectually we need to

[04:49:26] solve the problem, but if every day

[04:49:28] they're realizing this is going nowhere,

[04:49:30] they're going to get on it more quickly.

[04:49:32] So that's the idea is to have kind of a

[04:49:34] two-way like they're educating teachers

[04:49:37] and the teachers are innovating and

[04:49:39] hopefully they're going to come up with

[04:49:41] I mean again yeah like AI gives the

[04:49:44] power to teachers to do incredible

[04:49:47] things to really good ends if if they're

[04:49:50] oriented in in the way we all hope.

[04:49:53] So so the value proposition for

[04:49:54] anthropic is really the feedback and the

[04:49:57] use model. Yeah. That they get for it.

[04:49:59] Yeah.

[04:49:59] Interesting. Have you have you had

[04:50:01] experience in this trying to get the

[04:50:03] different sectors together?

[04:50:06] I I would say that I tend to be focused

[04:50:09] a bit more on the private sector, but um

[04:50:12] in my volunteerism, for example, um I

[04:50:17] serve as a trustee of Sesame Workshop,

[04:50:20] which does Sesame Street. And one of the

[04:50:23] things that they have done is um partner

[04:50:26] with Google around different ways of

[04:50:28] distributing um what is really backed by

[04:50:32] early childhood education research.

[04:50:34] almost everything that happens on Sesame

[04:50:37] Street has gone through this gauntlet of

[04:50:39] um real practitioners and academics in

[04:50:42] terms of um whether or not it actually

[04:50:45] advances kids learning for that age

[04:50:47] group. Um they also have developed um

[04:50:50] technology platforms that can be used in

[04:50:53] um refugee camps for example um in

[04:50:57] conjunction with the um international

[04:51:00] refugee committee. So, we're looking at

[04:51:03] technology as a way to get that

[04:51:05] education in the hands of kids

[04:51:06] everywhere in the world. Um, and using

[04:51:09] philanthropic dollars um and commercial

[04:51:12] dollars because we license um certain

[04:51:14] aspects of the programming, certain

[04:51:16] aspects of the characters um for uh for

[04:51:21] uh for profofit use so that we can

[04:51:23] support the educational mission. So we

[04:51:26] do work on a variety of different

[04:51:28] approaches that incorporate education

[04:51:31] technology as well as um real uh

[04:51:35] evidence-backed education for kids in

[04:51:37] that age group.

[04:51:39] So that the uh the charge to you is to

[04:51:42] go mingle, get to know each other, go on

[04:51:46] a date, figure out how you can help uh

[04:51:50] combine the best of research and

[04:51:51] entrepreneurship and philanthropy and

[04:51:54] government. That that's that's the

[04:51:56] purpose of this. So go do that. Uh so

[04:51:59] there there's a minute or two for a

[04:52:01] question from the audience. You have two

[04:52:03] of the most uh uh hard charging make a

[04:52:08] difference people on earth who care

[04:52:10] about education.

[04:52:12] Not to be intimidated but

[04:52:16] I mean when you give an intro like

[04:59:40] All

[04:59:55] right.

[04:59:59] Okay.

[05:00:02] Well, that was a terrific way to get

[05:00:06] back after lunch and we continue on with

[05:00:08] our next panel. um on AI quests. Um this

[05:00:13] is a really exciting exciting topic. I

[05:00:15] I'm particularly enthusiastic about the

[05:00:18] way in which this group is going to be

[05:00:19] able to share with us how we are

[05:00:21] translating complex research into age

[05:00:24] appropriate lesson plans. And so we have

[05:00:26] a terrific panel here um with colleagues

[05:00:29] from Google research and from the

[05:00:31] accelerator for learning at Stanford. Um

[05:00:33] so we have Alan Harris who is a senior

[05:00:36] program manager at Google research and

[05:00:38] is the lead for this particular

[05:00:39] initiative. Uh Victor Lee who is an

[05:00:43] associate professor at the graduate

[05:00:44] school of education and a faculty lead

[05:00:46] for AI and education at the accelerator

[05:00:48] for learning. And then our moderator

[05:00:50] will be Renit Lavi Morad who is a senior

[05:00:53] director at Google research um who works

[05:00:56] across a number of domains but has a

[05:00:58] particular passion for education and is

[05:01:01] a sponsor of a number of initiatives at

[05:01:03] Google research uh including the one

[05:01:05] that we're going to talk about today. So

[05:01:07] let's welcome our next panel to the

[05:01:08] stage.

[05:01:17] Welcome to AI Quests. My name is

[05:01:20] Professor Sky, an AI expert and your

[05:01:23] mentor. Throughout this learning

[05:01:25] experience, we've been waiting for

[05:01:28] someone like you. You're about to begin

[05:01:31] a journey through our world. As you

[05:01:33] travel, you'll meet different

[05:01:35] inhabitants and learn about the

[05:01:38] challenges in their communities. Your

[05:01:40] mission is to help them by applying

[05:01:43] powerful AI responsibly to their

[05:01:46] problems. Each quest will teach you

[05:01:49] about a reall life AI research project

[05:01:53] that's making an impact back on Earth.

[05:01:56] The AI technology you'll experience is

[05:01:59] amazing, but it's not magic. Success

[05:02:02] depends on the decisions you make.

[05:02:05] You'll need to collect good data, train

[05:02:07] your AI to make smart choices, and test

[05:02:10] it to make sure it's reliable. The

[05:02:13] universe is counting on you. Are you

[05:02:16] ready?

[05:02:29] Here is good. Yeah.

[05:02:30] Okay.

[05:02:31] Hi everyone.

[05:02:34] Yeah. And welcome to our panel AI quest

[05:02:39] when learning sciences meet product

[05:02:41] design toward AI literacy.

[05:02:45] If there's one word that I'd like to

[05:02:48] highlight today is the word meet. We are

[05:02:52] here because the three of us are

[05:02:54] convinced that a cross-disciplinary

[05:02:56] approach is not just helpful, it's

[05:03:00] essential. It's wonderful to be here

[05:03:02] today and I want to thank Victor and

[05:03:05] Isabelle. Are you here Isabelle?

[05:03:07] There's Isabelle. Hi.

[05:03:09] Um, thank you for the opportunity to

[05:03:12] showcase a project that is truly near

[05:03:15] and dear to our hearts. AI Quests. This

[05:03:18] initiative is an outcome of uh an

[05:03:21] incredible collaboration between Google

[05:03:24] Research and the Stanford Accelerator

[05:03:27] for Learning. I'm joined today by the

[05:03:31] architects of this synergy, Victor Lee

[05:03:34] and Alon Harris. Uh they represent today

[05:03:39] um their respective teams and I think

[05:03:43] that Joba and Kristen and Chris are all

[05:03:47] here today and we have Leat on live

[05:03:49] stream. Um thank you all for your

[05:03:53] critical role in bringing um this vision

[05:03:57] to life.

[05:03:59] Um, before we dive into the specifics of

[05:04:02] AI Quests, I'd like to zoom out a little

[05:04:06] bit and address a broader landscape of

[05:04:09] AI literacy and readiness.

[05:04:12] Like many of you here today, this panel

[05:04:16] shares a deep commitment to preparing

[05:04:19] the next generation

[05:04:22] um for an AIdriven

[05:04:24] future or perhaps perhaps I should say

[05:04:29] um it's here. It's already here, right?

[05:04:32] This future has already uh arrived. Um,

[05:04:37] we are seeing a widening gap between the

[05:04:40] velocity of AI development, tech

[05:04:42] development, and the depth of human

[05:04:45] understanding. We've talked about it uh

[05:04:47] along the day. The metrics are telling.

[05:04:52] While 72% of students are routinely

[05:04:55] using AI on their day-to-day, only 28%

[05:05:00] can accurately describe how a large

[05:05:04] language model actually predicts text.

[05:05:09] Victor, your

[05:05:12] first question is to you to bridge this

[05:05:15] gap. Um,

[05:05:18] you've been advocating for a shift from

[05:05:21] simply teaching AI usage to fostering

[05:05:25] epistemic vigilance. That's the term you

[05:05:28] use. You believe students must learn to

[05:05:31] navigate machine reasoning with true

[05:05:34] confidence.

[05:05:36] Looking at that transition from service

[05:05:39] from surface level prompting to

[05:05:42] independent critical thinking. What is

[05:05:44] the roadmap for achieving it?

[05:05:48] Thanks for the question Ron. Thanks

[05:05:50] everybody for uh joining us today and

[05:05:52] thank you for this partnership. It's

[05:05:53] just really been a delight to work

[05:05:55] together on this. Um you know one of the

[05:05:57] things that we really need to think

[05:05:59] about is what we're aiming for with AI

[05:06:01] literacy. So, I think the big mistake,

[05:06:04] and we saw this play out early on, was

[05:06:06] everyone said, "Oh my goodness, prompt

[05:06:08] engineering is the career of the future.

[05:06:09] We need to focus on everybody learning

[05:06:11] how to prompt." Um, and so that in some

[05:06:14] way became synonymous with becoming AI

[05:06:16] literate. And what I would say is that's

[05:06:18] really,

[05:06:20] you know, barely the surface of anything

[05:06:23] in there. When we think about becoming

[05:06:24] AI literate, we do want to think about,

[05:06:26] you know, how do you use AI? But that

[05:06:28] interface and that experience is going

[05:06:30] to change. We need to think about what's

[05:06:31] the durable interaction and positioning

[05:06:34] that that a individual or student has

[05:06:37] with the technology as it continually

[05:06:39] evolves. Um you know to some extent we

[05:06:41] need to understand a little bit about

[05:06:43] the mechanisms that it works that it's

[05:06:45] not magic underneath there that it's not

[05:06:48] a person hiding in there although I

[05:06:50] think there was some reports about like

[05:06:51] whimo cars maybe secretly being

[05:06:53] controlled by a person in the

[05:06:54] Philippines. Um but you know for most

[05:06:57] the AI is not and so I think really

[05:07:00] that's orienting towards uh you know a

[05:07:03] critical mindset and one that emphasizes

[05:07:05] human agency to think like what is the

[05:07:08] role that humans had in creating this

[05:07:10] technology what are the responsible ways

[05:07:12] to use and engage with that and with

[05:07:14] that technology it also means look I

[05:07:17] know chat bots are definitely having

[05:07:18] their moment but AI is so many more

[05:07:20] things than that and so helping people

[05:07:22] to see like what is that how did this

[05:07:26] notion of chat bots extend from these

[05:07:28] other principles of machine learning and

[05:07:30] you know the use of data to build

[05:07:32] predictions or to classify or or other

[05:07:35] things of that ilk. And so really it's

[05:07:38] it's about directing towards uh a way of

[05:07:41] thinking about and viewing AI that

[05:07:43] involve humans that involves humans you

[05:07:45] know strengths and and uh you know

[05:07:48] shortcomings and ways that we work on

[05:07:49] perfecting that and positioning students

[05:07:52] um and our educators as agents in this

[05:07:55] to really have a sense of agency and

[05:07:57] like how we use AI, how we create AI,

[05:07:59] how we evaluate AI and that epistmic

[05:08:02] vigilance is really going to be about

[05:08:03] you know maintaining that that real

[05:08:05] thought about what we bring to the table

[05:08:07] as humans, what we need to really be

[05:08:09] thoughtful and careful about.

[05:08:10] Thank you, Victor. I couldn't agree

[05:08:12] more. And speaking of the um this theme

[05:08:16] of human agency, moving from passive

[05:08:19] users to active critical thinkers, I

[05:08:22] think that this is the perfect segue to

[05:08:26] um into how everything started with AI

[05:08:28] Quest. So, Alon,

[05:08:30] thank you. Yeah, I'm happy to tell a

[05:08:32] little story here. First of all, Victor

[05:08:34] and I have been working for over a year

[05:08:36] and a half together, but this is the

[05:08:38] first time I've ever met him in the

[05:08:39] flesh. So, uh, thank you, Isabelle. He

[05:08:43] he's real. He's in 3D here. And so, uh,

[05:08:47] it's just just that was already worth

[05:08:49] the travel just to come and see Victor.

[05:08:51] Um, the story of AI Quest begins not far

[05:08:54] away from here at the headquarters of

[05:08:55] Google in Mountain View. Um, and about a

[05:08:57] year and a half ago, we were having a

[05:08:59] brainstorm. uh Google research has been

[05:09:02] doing volunteer work to bridge tech to

[05:09:05] teens for many many years but AI

[05:09:08] literacy is kind of a there's no

[05:09:10] playbook right and we were asking

[05:09:12] ourselves recognizing what Victor said

[05:09:14] that there was the curricula out there

[05:09:16] we did some some of our market research

[05:09:18] and we saw there's a lot of emphasis on

[05:09:20] technical skills on coding speaking

[05:09:22] about the technology but basically

[05:09:25] outside of any human related context we

[05:09:28] said how can we address this gap gap in

[05:09:30] the AI literacy space but authentically

[05:09:32] as a as a research organization and so

[05:09:34] we look to our own research for the

[05:09:37] answer. So Google research is investing

[05:09:41] in societal impact through AI. We have

[05:09:45] teams that when they go to work in the

[05:09:47] morning, what they're thinking of are

[05:09:48] these grand challenges of humanity,

[05:09:51] whether they're related to climate and

[05:09:53] sustainability or to health or to

[05:09:55] science. They're asking questions like

[05:09:58] how can we give society early warning

[05:10:00] before a flood hits? How can we provide

[05:10:04] um uh early diagnosis for diseases to

[05:10:07] populations in developing markets that

[05:10:09] just don't have a doctor uh you know

[05:10:11] nearby in a local clinic. And in terms

[05:10:14] of science, they're thinking of uh you

[05:10:16] know the human brain and how can we

[05:10:18] actually map that brain knowing that it

[05:10:20] will take thousands of years if we use

[05:10:22] the current state-of-the-art methods uh

[05:10:25] and thinking of how AI can help. And so

[05:10:27] we thought, wouldn't it be interesting

[05:10:30] if we could provide students the

[05:10:32] opportunity to step into the shoes of a

[05:10:34] human researcher and go on a journey to

[05:10:38] try and answer these questions the same

[05:10:39] way we do, but in a gamified immersive

[05:10:42] experience. And this is the birth of AI

[05:10:45] quests. What you see here on the screen

[05:10:47] are the three quests we're currently

[05:10:49] working on. The first two have been

[05:10:50] launched and the third will be arriving

[05:10:53] in the summer. These are open, free, um,

[05:10:57] instructor-led experiences. Uh, and

[05:10:59] they're all, we'll go into this in a

[05:11:01] minute, but they're occurring in a

[05:11:03] fantastical uh, world. And students,

[05:11:05] they're have to explore. We're not going

[05:11:08] to explain upfront. We're going to let

[05:11:09] them loose. They're going to have to try

[05:11:11] and figure out how to solve these grand

[05:11:13] challenges. And that's uh, you know, uh,

[05:11:17] that was the beginning of the idea. The

[05:11:18] most important thing I think in this

[05:11:19] concept is early on we recognized that

[05:11:22] we're we're a tech company. We need

[05:11:24] experts in learning science partners who

[05:11:26] will co-design this with us that will

[05:11:29] bring their understanding of the target

[05:11:31] audience. It's been a long time since I

[05:11:33] was in middle school. We need people who

[05:11:35] engage regularly with these audiences

[05:11:37] know know their you know uh uh their

[05:11:39] their behaviors and attitudes towards uh

[05:11:41] the world and tech. And so we're it's

[05:11:43] just been a privilege working with the

[05:11:44] accelerator for learning here at

[05:11:45] Stanford. uh and everything you're going

[05:11:48] to see here today is is a collaborate is

[05:11:50] a result of that fruitful collaboration.

[05:11:52] Alon, I'm still super excited about this

[05:11:56] uh this vision. Um and I'd love to get

[05:11:59] into the mechanics a little bit. So, and

[05:12:02] I'm sure everyone here is curious to

[05:12:04] learn more about you know um everything

[05:12:08] you know the the experience that the

[05:12:11] learner is going through. So, please if

[05:12:13] you can share a little bit more. Can I

[05:12:15] double check that everyone's curious

[05:12:16] about what the learning journey is in AI

[05:12:18] Quest? Can I see a bit of energy here

[05:12:20] after lunch? I know it's a bit it's a

[05:12:22] tough hour after lunch. You know,

[05:12:24] biology is a challenge. Um I want to

[05:12:28] highlight the first thing is before we

[05:12:29] go into the gamified experience. The

[05:12:32] teacher is at the heart of our our

[05:12:33] concept. We have uh outstanding teacher

[05:12:36] guides created by Stanford to guide

[05:12:38] teachers for the setup, the before and

[05:12:41] the after of this online gamified

[05:12:43] experience.

[05:12:44] um trying to basically emphasize that

[05:12:47] educators are the ones who are going to

[05:12:49] socialize, facilitate, have these deep

[05:12:52] conversations with the students. And

[05:12:54] what we're providing with the the

[05:12:55] gamified experience is the stimuli. Um

[05:12:58] and what you will see here is the

[05:13:00] standard kind of cycle or journey we we

[05:13:03] send students on. Um and it begins with

[05:13:06] understanding the problem they're trying

[05:13:07] to solve. We spoke in the panels earlier

[05:13:09] today about human- centered AI. So it

[05:13:12] begins with understanding who are we

[05:13:14] solving for and what are we trying to

[05:13:17] create before we go into the techy

[05:13:19] stuff. And here you see the quest for

[05:13:21] flood forecasting. Uh each task is

[05:13:24] mimicking the life cycle of a research

[05:13:26] uh effort. You you have to go and search

[05:13:29] for data. Data doesn't just readily jump

[05:13:32] uh you know on command and it's not

[05:13:35] available necessarily. Students are

[05:13:37] going to have to distinguish between

[05:13:39] relevant and irrelevant data. And so we

[05:13:41] purposely make it hard. There's 12 data

[05:13:43] sources. They need to sc like in a

[05:13:45] scavenger hunt find in the forest. They

[05:13:47] need to clean the data. Students don't

[05:13:49] know that there is a process of uh

[05:13:52] qualifying your data for research. You

[05:13:54] have to clean it. Uh they love this

[05:13:56] step. And then they go through a

[05:13:57] training and testing cycle where they

[05:14:00] understand that if they made bad choices

[05:14:02] upstream, they're going to get a really

[05:14:04] suboptimal AI model downstream. And so

[05:14:07] built into the quest is a loop where uh

[05:14:10] students may have to retrace their steps

[05:14:13] and and change their decisions and make

[05:14:16] make better decisions to see if the

[05:14:18] model performs better for the community

[05:14:20] they're trying to serve. And at the very

[05:14:22] end they deploy their tool to the

[05:14:24] community, see it in action. Because an

[05:14:28] AI model in a lab is one thing, but if

[05:14:30] we're going to do societal impact,

[05:14:31] really make a change, you have to see it

[05:14:33] in the real world. And just for dessert

[05:14:35] at the end, uh they get to meet the real

[05:14:37] researchers. So we kind of unveil to

[05:14:39] students that this isn't a just a a nice

[05:14:42] animated experience. This is something

[05:14:45] that teams, people, experts are devoting

[05:14:47] their lives to. Um and here they'll meet

[05:14:50] Gray Antal who will tell them about the

[05:14:52] work they do to create flood forecasting

[05:14:55] for two billion people around the world

[05:14:58] that we can today with AI provide up to

[05:15:01] seven days early warning before a

[05:15:03] riverine flood occurs. So this is kind

[05:15:06] of the general gist of it.

[05:15:07] That's fantastic. Thank you. Thank you

[05:15:10] Alon. Um let's talk about pedagogy for

[05:15:14] for a moment. And I can see my colleague

[05:15:16] Miriam here. Actually, I just wanna Hi,

[05:15:20] my love. I just want to I I I

[05:15:23] I'd love to hear your thoughts maybe

[05:15:25] afterwards, but I'm I'm going to talk a

[05:15:27] bit about learnm and our experience with

[05:15:29] with learnm because when our team at

[05:15:32] Google when we built learnm at the time,

[05:15:36] our family of models specifically tuned

[05:15:39] for learning and education. Our

[05:15:42] foundational principle was to ground

[05:15:46] everything in a rigorous learning

[05:15:49] sciences. Right? And it's clear to

[05:15:53] everyone here that pedagogy is not just

[05:15:55] an add-on or something like that, a

[05:15:58] feature. Everything we build for

[05:16:01] learning and education must be

[05:16:03] co-designed with experts, with teachers,

[05:16:07] with learners, with parents, right? And

[05:16:10] beyond that. So, Victor,

[05:16:14] um, related to that, I've heard you talk

[05:16:17] about the magic box myth, um, where

[05:16:21] students assume AI just knows the

[05:16:23] answers. For AI Quest, we didn't just

[05:16:27] want to build a game. We wanted to bake

[05:16:30] in that pedagogical foundation.

[05:16:33] Could you explain how AI Quest actually

[05:16:36] empowers a 12year-old to demystify that

[05:16:40] box and start thinking like a

[05:16:42] researcher?

[05:16:43] Yes, for sure. I mean, I think for those

[05:16:45] of you in learning sciences and

[05:16:47] instructional design worlds, you're

[05:16:49] familiar. This needs to be sort of the

[05:16:50] the front and center commitment in terms

[05:16:53] of how do you even identify and

[05:16:54] centralize what learning goals you're

[05:16:56] pursuing in there. And so using

[05:16:58] techniques that are familiar with in a

[05:17:00] backwards design paradigm, you know,

[05:17:02] naming and centering the enduring

[05:17:05] understanding that we really care about

[05:17:07] in this case being humans can initiate

[05:17:09] design AI applications that can address

[05:17:11] some of humanity's biggest unsolved

[05:17:13] challenges and latent in that is the

[05:17:15] ethical responsibility that is

[05:17:18] associated. And what that does is it

[05:17:20] really um shifts the focus away from

[05:17:23] what's the mechanism underneath it, but

[05:17:25] rather what is this

[05:17:27] like within the actual types of real

[05:17:30] questions and problems that that we uh

[05:17:32] try to address. And then you know as a

[05:17:35] broad statement for the approach is you

[05:17:37] then want to think about what we have

[05:17:39] established from you know recent

[05:17:41] learning sciences research about

[05:17:42] scaffolding digital scaffolding and the

[05:17:44] ways in which we can best support

[05:17:46] learners both for engagement and for new

[05:17:48] knowledge and new participation in

[05:17:50] practices. Like we for instance have

[05:17:52] very thoughtfully uh strategically used

[05:17:55] the various digital characters various

[05:17:57] pedagogical agents um that are there to

[05:18:01] help make clear like what is the

[05:18:04] consequence of the actions that are in

[05:18:06] here where is there subject matter

[05:18:08] expertise that is uh necessary.

[05:18:11] Oh one moment one moment so yeah we've

[05:18:13] got Luna here. She's competing with

[05:18:14] Victor. Luna is the character

[05:18:15] that's one of the pedagogical agents. Um

[05:18:18] so she she will illustrate that for us.

[05:18:20] Um, but also what I'd say is in in you

[05:18:22] know the recent decades of learning

[05:18:24] sciences research, it's situating this

[05:18:26] all in context and thinking about where

[05:18:28] the scaffolds fit within that context. A

[05:18:30] lot of what we learn is fundamentally

[05:18:33] all we learn is within the context with

[05:18:35] actual purpose with tools with norms um

[05:18:38] with actual concerns um and consequences

[05:18:41] in mind. And so this I think is a really

[05:18:43] important and unique offering in the AI

[05:18:45] literacy space to give that full

[05:18:48] experience. Not just say this is what a

[05:18:50] neural network looks like but rather

[05:18:51] this is how this all fits within this

[05:18:54] arc. How do you make those decisions?

[05:18:56] What tools do you draw from? how do you

[05:18:57] involve and work with the other people

[05:18:59] or in this case agents um in that and so

[05:19:03] that immersion that that uh way of

[05:19:06] bringing that to life to you know really

[05:19:08] acknowledge the social uh situated and

[05:19:12] distributed nature of human learning and

[05:19:14] doing is key and I think Luna will help

[05:19:16] illustrate

[05:19:17] Luna will help us yes

[05:19:21] oh no we're in so much trouble hi I'm

[05:19:24] Luna the manager of the market stalls.

[05:19:27] Here at Market Marshes, our lives

[05:19:29] revolve around the Sasa River. But as

[05:19:32] you can see, the river can flood our

[05:19:34] market without warning. We didn't even

[05:19:36] have enough time to take the stalls

[05:19:38] down. We've been trying to predict

[05:19:40] flooding, but this weather forecast

[05:19:42] isn't reliable enough. We've heard that

[05:19:44] AI can help to predict extreme weather

[05:19:47] events like floods. With accurate early

[05:19:50] warning, we can protect our market and

[05:19:52] avoid all this mess. So, can you do it?

[05:19:57] So, Victor, I've got a few examples

[05:19:59] here. Maybe we can talk to the cleaning

[05:20:01] data.

[05:20:02] Sure.

[05:20:02] Yeah, sure.

[05:20:04] Um, so right there you saw Luna who's

[05:20:06] setting the context and saying why does

[05:20:08] this matter? We also thought a lot about

[05:20:10] how do we represent things such as like

[05:20:12] what would prediction of weather look

[05:20:14] like in a way that's accessible but

[05:20:15] still has that probability based nature

[05:20:17] to it um in there and that's really been

[05:20:19] sort of the driving orientation around a

[05:20:21] lot of the tasks and experiences. So for

[05:20:23] example, cleaning data and about how

[05:20:26] much uh involvement because in reality

[05:20:28] cleaning data takes a lot of time and a

[05:20:30] lot of energy. Um but getting students

[05:20:32] to think about what would I look for

[05:20:34] what would seem anomalous and build upon

[05:20:37] my understanding or uh parsing of this

[05:20:39] particular context and having situations

[05:20:41] where the students can fail and need to

[05:20:43] redo that. Um, we build in uh these uh

[05:20:46] self-explanation mechanisms afterwards.

[05:20:49] So that way when they pass on that

[05:20:50] cleaning data task to one of the other

[05:20:52] agents uh in the game, they have to

[05:20:55] articulate what it is that you need to

[05:20:56] look for and do. Uh in that process we

[05:20:59] have um you know for instance in this

[05:21:02] market stalls placement task working

[05:21:04] with the kinds of um color-coded data

[05:21:06] representations to show where there's

[05:21:08] high and low risk for flooding and to

[05:21:10] also tie things back into what was

[05:21:12] driving the story in the first place. So

[05:21:14] that way what would you do if you had

[05:21:16] advanced warning and you really do need

[05:21:18] to keep the markets operating um you

[05:21:20] know regardless of the conditions and

[05:21:22] how do we do this safely and

[05:21:23] responsibly. Um and so a lot of this is

[05:21:25] really thinking about how can we keep

[05:21:27] that context alive, keep the content

[05:21:30] accessible but still accurate um both

[05:21:33] scientifically and to what actually

[05:21:34] happens in the journey and to build in

[05:21:37] um productive failure loops uh because

[05:21:40] trying and retrying and getting the

[05:21:42] feedback is fundamentally important to

[05:21:45] that learning process. Um, as you've

[05:21:47] probably heard, there's there's a lot of

[05:21:49] things that we build in the pedagogical

[05:21:51] approach. And this again is really very

[05:21:53] much uh built from our work in the

[05:21:55] learning sciences feedback and

[05:21:57] productive failure. And as you can see,

[05:21:59] you you can really have incorrect data

[05:22:01] or you can actually build a model that

[05:22:02] just doesn't meet accuracy criteria and

[05:22:05] you could even build higher than that.

[05:22:07] Um, our pedagogical agents and thinking

[05:22:09] about this notion of the specific type

[05:22:11] of a teachable agent. So leveraging that

[05:22:14] ability that a lot of the the strong

[05:22:17] research that actually come out of

[05:22:18] Stanford on this about how um teaching a

[05:22:21] virtual entity to make some of the

[05:22:23] decisions that you would make um is a

[05:22:26] really good externalization a good sort

[05:22:28] of consolidation um and way to see the

[05:22:30] the consequences implications um

[05:22:33] self-explanation mechanisms which is a a

[05:22:36] long-standing u uh attribute long

[05:22:39] long-standing mechanism in learning

[05:22:40] sciences that we know that um expert

[05:22:42] comprehension tends to draw upon and you

[05:22:44] know situating this with real tasks such

[05:22:47] as having to consult multiple experts

[05:22:48] who may not always agree on what label

[05:22:51] to assign and and try to maintain that

[05:22:53] authenticity as is illustrated in this

[05:22:55] particular case of labeling um expert

[05:22:58] labeling of of I data. So we really

[05:23:00] think about that. We think about even

[05:23:02] with you know for the next slide um the

[05:23:04] the teacher support materials because um

[05:23:08] you know over the past decades all this

[05:23:10] work in learning sciences all this

[05:23:11] design work in classrooms has made

[05:23:13] starkly obvious the importance of

[05:23:16] working in tandem with a teacher

[05:23:18] supporting their adaptation processes

[05:23:20] creating lots of connection points and

[05:23:23] um malleability with the materials but

[05:23:26] also helping them to feel empowered to

[05:23:28] teach and engage students on this topic

[05:23:31] that's not as familiar to them and so

[05:23:33] thinking a lot about what what a lot of

[05:23:35] the recent research has said about

[05:23:37] teacher learning, teacher practice,

[05:23:38] curricular materials based support.

[05:23:42] Thank you Victor and thank you Alon. Now

[05:23:44] I actually have a question for uh both

[05:23:48] of you. But before that, I must say that

[05:23:51] this partnership between Google research

[05:23:53] and the Stanford Accelerator for

[05:23:55] Learning

[05:23:56] really represents a rare intersection of

[05:24:00] academic rigor

[05:24:02] um product development, experimentation,

[05:24:05] and real world impact. Now looking back

[05:24:10] at a year of code design,

[05:24:13] I'd like to get into the weeds a little

[05:24:15] bit and talk about the intense reality

[05:24:18] of code design.

[05:24:20] Can you share a moment where you had to

[05:24:23] stop, put your heads together and

[05:24:27] decipher truly tricky design challenge?

[05:24:32] It's it's it's a tricky one. Um, first

[05:24:35] of all, I just want to highlight that,

[05:24:38] you know, Victor and I, we did a demo

[05:24:39] here at lunch

[05:24:41] and we were saying to each other, you

[05:24:43] come, you have a beautiful presentation.

[05:24:45] It looks like it was all predestined to

[05:24:46] to just be such a smooth journey. And I

[05:24:49] want to say the the the biggest surprise

[05:24:51] is that Victor and I are still speaking

[05:24:52] to each other, okay? Because it's been a

[05:24:56] creative journey. It's an iterative

[05:24:57] journey. And then you take it to the

[05:25:00] students and test it with them. And then

[05:25:02] you learn that some of your assumptions

[05:25:04] are off and you have to iterate. So I

[05:25:06] think Ronnie this is uh been a journey

[05:25:09] of cocreation which means negotiation.

[05:25:14] It means finding common ground and

[05:25:16] finding a way of um uh balancing between

[05:25:20] uh not just our part here but also the

[05:25:24] creative agencies that we worked with to

[05:25:26] to bring this art to life and the

[05:25:28] students and the teachers uh that we

[05:25:30] tested this with. But one example,

[05:25:32] Victor, I I I I thought I would be

[05:25:34] interesting is we spent hours trying to

[05:25:36] understand how to explain training and

[05:25:38] testing,

[05:25:40] but not to dumb it down and make it too

[05:25:42] simple, but not to make it over

[05:25:43] technical and alienate the students. And

[05:25:45] that middle ground, that sweet spot is

[05:25:46] really hard to find. And so we were we

[05:25:49] we have like a thousand versions of what

[05:25:51] you're seeing on the screen. And I'm

[05:25:53] sure one of the things with animation

[05:25:55] is, you know, that you can erase and do

[05:25:57] it again. So we also had to be

[05:25:58] disciplined when to say stop you know to

[05:26:01] say okay let's go with this because we

[05:26:03] could continue easily iterating forever.

[05:26:05] Um and I really loved the low match high

[05:26:09] match solution we found. We wanted to

[05:26:12] express that in in testing we're com

[05:26:14] comparing uh the ground truth from

[05:26:17] historical data with uh what the model

[05:26:20] shows and this is a very complicated

[05:26:23] concept of historical data and accuracy

[05:26:25] etc. And so we had this beautiful uh uh

[05:26:28] illustration here that that um shows a

[05:26:31] high or low match between these two

[05:26:32] visuals. So it's almost like a visual

[05:26:34] expression of something more

[05:26:35] complicated. Um Victor, you you want to

[05:26:39] Yeah, I mean I I think um related to

[05:26:42] that is how do you convey that the AI is

[05:26:44] making a prediction um against something

[05:26:47] that we have as a more trusted and

[05:26:49] accurate source of information. So

[05:26:51] that's been a recurring theme. But also,

[05:26:54] you know, as students go through this,

[05:26:55] you know, for them to understand where

[05:26:57] was the AI in this and what does the AI

[05:27:00] actually do, how did it get made and how

[05:27:02] do we represent that without them

[05:27:05] thinking, oh, you really have to know

[05:27:06] all the math behind it um to get there.

[05:27:09] there. And so there's been a lot of

[05:27:10] really careful things even going to the

[05:27:11] point where when image based data are

[05:27:14] used to train the model, we have a whole

[05:27:16] segue about how it needs to convert to

[05:27:19] things like numbers so that way it

[05:27:21] becomes a form that the AI can use to

[05:27:23] get away from sort of a a growing

[05:27:25] misconception that um AI sees and

[05:27:28] perceives as humans see and perceive.

[05:27:31] And so it's just those sorts of details

[05:27:32] about how can we maintain that accuracy

[05:27:36] um and fidelity but in a way that

[05:27:38] doesn't get bogged down in the details

[05:27:40] and still maintain maintains a lot of

[05:27:42] that.

[05:27:42] You know one thing I learned from you

[05:27:44] Victor early on you you told us

[05:27:46] teenagers don't understand what data is.

[05:27:49] It's a it's an a theory

[05:27:50] it's a phone plan

[05:27:52] what is data right and let alone other

[05:27:55] terms like model that we spent time on

[05:27:57] etc. And so we tried to represent data

[05:28:00] without showing it. We didn't want to

[05:28:03] overwhelm them with tables with numbers.

[05:28:05] So at the beginning we had this idea of

[05:28:06] maybe using tokens uh like in a game.

[05:28:09] But but then we transitioned to

[05:28:11] something that was kind of a metaphor

[05:28:13] which was these SD cards. And then

[05:28:15] students, oh okay, it's it's it's

[05:28:18] storage. It's it's something you you

[05:28:20] store information on. And I think that

[05:28:22] was a really cool um kind of solution we

[05:28:25] found. And you know I bring these I went

[05:28:28] down memory lane. I went 500 versions

[05:28:30] ago to the black and white versions you

[05:28:33] see here which is where we start all the

[05:28:34] journey. And I think u with this quest

[05:28:37] which is about eye uh um diagnosing eye

[05:28:40] disease early on. We had a hard time

[05:28:42] with the story. The real story is about

[05:28:45] diabetic retinopathy. The real story is

[05:28:48] about losing your eyesight to the to the

[05:28:51] brink of blindness. Now, that's a really

[05:28:53] stark kind of starting point for

[05:28:55] something gamified. And we tried to find

[05:28:57] a way to kind of decouple this from any

[05:29:01] medical background. And we came up with

[05:29:02] the idea of uh a community on this

[05:29:05] planet that has an eye disease due to

[05:29:08] UV.

[05:29:10] And this was kind of I think a creative

[05:29:12] solution which is something okay kids in

[05:29:14] general can relate to the sun to UV you

[05:29:18] know they they they have some knowledge

[05:29:19] of that it's not healthy and we found

[05:29:21] that way of addressing this and kind of

[05:29:24] bypassing the need to go through

[05:29:26] anything that might trigger anything uh

[05:29:28] negative um

[05:29:30] yeah for example I know there's

[05:29:31] consultation sessions with uh uh

[05:29:33] Googlers uh who are blind and low vision

[05:29:36] themselves or work on the accessibility

[05:29:38] teams often discuss us. Um, and also

[05:29:42] thinking about well, how do we get to

[05:29:45] this problem of data diversity in this

[05:29:48] fictional world and you know with within

[05:29:51] that context? We wanted to knowing that

[05:29:54] there has been and continues to be a

[05:29:55] tendency to think about like a

[05:29:56] biological essentialism to move away

[05:29:58] from something that's like a biological

[05:30:01] trait like the eye color or the shape of

[05:30:03] their eyes and think about how can we

[05:30:06] more accurately represent what would be

[05:30:08] the science but also ways that you know

[05:30:10] speak to the importance of the diversity

[05:30:13] here and in this case diversity of UV

[05:30:15] exposure at different regions in the

[05:30:17] planet which created all these

[05:30:19] interesting and fun game mechanisms and

[05:30:20] and unveiled all these new lands.

[05:30:24] Amazing. We are almost at time. We

[05:30:26] actually at time um but I want to say

[05:30:29] before we say goodbye that the feedback

[05:30:32] on AI Quest has been phenomenal. Now can

[05:30:36] you give me a bit some of this feedback

[05:30:39] can you share what you hear from the

[05:30:40] classroom a little bit just a little bit

[05:30:43] for for everyone to to hear? Yeah. I

[05:30:45] mean, I think that it's been fun and

[05:30:48] engaging and at a pace that um doesn't

[05:30:51] feel like it drags and still has enough

[05:30:54] of the game mechanics that keeps things

[05:30:56] moving forward. You know, I've tested

[05:30:58] this with uh my teenage daughter and

[05:31:00] friends and they're sort of like roll

[05:31:02] their eyes and go through it and I say,

[05:31:04] "So, what what was the big point?" And

[05:31:06] in their words, it was precisely where

[05:31:08] we wanted uh that enduring understanding

[05:31:10] to land in there. And that was just a

[05:31:13] big hit that that message um had come

[05:31:15] across even though it's very subtly

[05:31:17] embedded within within all of that. So I

[05:31:20] think that was a really encouraging

[05:31:22] thing and just people to ask about it

[05:31:23] and I've had compliments from strangers

[05:31:25] which is really nice.

[05:31:26] Amazing alone.

[05:31:27] Yeah. We we um we sent 600 volunteers

[05:31:30] from Google to schools in their

[05:31:32] community uh last December for computer

[05:31:34] science education week here in the US

[05:31:36] and the this was for some of them going

[05:31:39] back to the school they studied in and

[05:31:41] coming with this experience for the

[05:31:43] first time and the feedback was uh you

[05:31:45] know students felt that the felt that

[05:31:48] the story enabled them to go like ah now

[05:31:51] I see this is not just chat bots right

[05:31:53] this there are problems still that we

[05:31:55] haven't solved this is a means it's not

[05:31:57] an end in with within itself and that

[05:31:58] that was just fantastic and um I just

[05:32:02] want to highlight the third quest which

[05:32:03] will be about mapping the human brain

[05:32:05] will um you know it's in the works

[05:32:09] we're losing sleep on it but that's part

[05:32:11] of the labor of love right because we

[05:32:13] really want it to be um you know a great

[05:32:16] focus on science on the rigor it takes

[05:32:19] on how AI can accelerate t taking the

[05:32:21] you know a lab in this case that's

[05:32:24] trying to understand memory and help

[05:32:27] accelerate the discovery of the inner

[05:32:29] workings of memory and the brain. Um,

[05:32:31] and I think students are going to love

[05:32:33] this. Um, yeah,

[05:32:34] great. So, to sum up, it sounds like a

[05:32:37] lovely 2026 is underway. Um, we are

[05:32:41] launching the third quest and we have a

[05:32:44] bold goal of reaching to uh two million

[05:32:48] students

[05:32:49] students until end of year. So, that's

[05:32:52] that's incredible. Um, we invite

[05:32:55] everyone here, policy makers, educators,

[05:32:59] parents, funders to join us,

[05:33:02] kids. We got students here,

[05:33:03] kids, your students. Let's co-design a

[05:33:07] future where technology truly serves

[05:33:10] pedagogy. That's the goal.

[05:33:13] Victor Alon, it's been a pleasure

[05:33:14] sharing the stage with you.

[05:33:16] Thank you.

[05:33:16] And uh, thank you everyone.

[05:33:18] Thank you.

[05:33:20] Thank you, Ron.

[05:33:32] Oh, you can bring him on down. Yeah,

[05:33:36] that's fantastic. Thank you.

[05:33:37] Thank you.

[05:33:37] Yeah. Yeah. Yeah.

[05:33:38] Thank you, Patrick.

[05:33:39] Yeah, of course.

[05:33:40] Okay.

[05:33:44] So, perhaps unnecessary, but because it

[05:33:47] is up on the large screens, but we are

[05:33:49] moving into breakout sessions. So there

[05:33:51] will be a breakout session in the

[05:33:53] adjacent room there and there. So we'll

[05:33:56] have trust, safety, privacy, and ethics

[05:33:59] over uh over on that side. And then

[05:34:03] we'll have learning differences there.

[05:34:04] And then it you could stay right here if

[05:34:06] you would like to join the social

[05:34:09] learning revolution. And so we'll get

[05:34:10] started on the breakouts in just a

[05:34:12] minute or so. But please find the room

[05:34:14] that you're most interested in and and

[05:34:16] we'll get started as quickly as we can.

[05:34:19] There we go.

[05:37:35] All right. Okay. I have a quick request

[05:37:39] for those who are staying for the social

[05:37:41] learning revolution. If you could gather

[05:37:44] yourself into groups of say five to

[05:37:46] eight,

[05:37:49] If that'd be ideal.

[05:37:52] Let's see if this happens.

[05:37:55] I'll probably I'll probably come back up

[05:37:56] and and encourage this again

[05:37:58] momentarily.

[05:38:02] I guess we're going to have to

[05:38:05] How much more time?

[05:38:11] Okay,

[05:38:13] just give people another couple minutes.

[05:38:15] What's up?

[05:38:16] Yeah. Should

[05:38:25] we try and get people? No. No. Should we

[05:38:27] try and get people started or

[05:38:28] Yeah. I went as

[05:38:33] Yeah, let's get it started.

[05:38:36] Yeah. And and

[05:38:38] all right.

[05:38:41] Okay.

[05:38:44] Love it. So, for the folks that are

[05:38:46] remaining, we're going to get started

[05:38:47] with the breakout session, um, the

[05:38:49] social learning revolution.

[05:38:52] Fantastic. Um,

[05:38:55] and it does look like we've kind of

[05:38:57] gathered into smaller groups, but we can

[05:38:59] we can we can get back to that

[05:39:01] momentarily. Um, I'll hand things over

[05:39:04] to our moderator, Glenn Kleimman. Um,

[05:39:06] Glenn's an adviser to the Sanford

[05:39:08] Accelerator for Learning and has been

[05:39:10] intimately involved in the development

[05:39:12] and creation of the agenda for this

[05:39:13] event. Um, and so I will hand it over to

[05:39:15] Glenn to to kick things off.

[05:39:18] Okay.

[05:39:18] Thanks.

[05:39:19] Am I on? Can you hear me? Thank you,

[05:39:20] Patrick. Uh, welcome everyone to our

[05:39:23] breakout session on the social learning

[05:39:25] revolution. How AI is transforming how

[05:39:28] we connect, collaborate, and grow. I've

[05:39:31] already been introduced. I'm Glenn

[05:39:32] Klyman, a senior adviser here. This

[05:39:35] session will have a different format

[05:39:37] than the others and we'll need to have

[05:39:39] you rearrange yourself in a bit so that

[05:39:41] we can have some small groupoup

[05:39:42] discussion and do some social learning.

[05:39:45] We'll begin with short lightning talks

[05:39:47] from our four wonderful panelists who

[05:39:49] I'll introduce in a moment. We'll then

[05:39:51] ask you to form small groups of from

[05:39:53] five to eight people. And I know this

[05:39:55] room is not great for it. We use would

[05:39:58] much prefer to have tables but we don't.

[05:40:00] So just kind of arrange yourself so you

[05:40:02] can talk to each other and you'll have

[05:40:04] time to discuss your reactions to the

[05:40:06] panelists, your own thoughts, your

[05:40:07] questions, your ideas. And during that

[05:40:10] time, the panelists and myself will

[05:40:12] circulate and join some of the groups to

[05:40:13] hear what you're saying. And then at the

[05:40:15] end, we'll have the panelists come back

[05:40:17] and each offer a couple minutes of

[05:40:18] reflections on what they've heard or

[05:40:20] what they're thinking after the

[05:40:22] discussions. Um, each of our panelists

[05:40:25] has a lot of important things to say,

[05:40:27] but they've kindly agreed to pick a few

[05:40:29] highlights and stick very strictly

[05:40:31] within the eight minute time limit we've

[05:40:33] given each of them. And I'll introduce

[05:40:35] them in the order in which they'll

[05:40:37] present and then I'll sit down and be

[05:40:38] quiet. First, we have there she is um

[05:40:42] Bethany sitting second. They're not

[05:40:44] sitting in the order of speaking.

[05:40:45] Bethany Drake Maples is a leading

[05:40:48] researcher in the field of AI and

[05:40:49] education. She's a PhD candidate at the

[05:40:52] Stanford Graduate School of Education

[05:40:54] and the founder and CEO of Atypical AI.

[05:40:58] It turns out it's not atypical to have

[05:41:00] students here also CEOs and founders of

[05:41:02] that companies at the same time. Uh she

[05:41:05] led technical teams at Google and

[05:41:07] founded and scaled several companies and

[05:41:09] also serves on the board of the CZI

[05:41:12] Learning Commons Initiative.

[05:41:14] Second speaker will be Peele Young Kim

[05:41:17] who's a professor of psychology at the

[05:41:19] University of Denver and the director of

[05:41:22] the center for brain artificial

[05:41:23] intelligence in child. She's widely

[05:41:26] recognized for expertise in brain

[05:41:28] development and human emotional bonding

[05:41:31] and professor Kim's current research

[05:41:34] focuses on the emotional and social

[05:41:36] dimensions of child AI interactions with

[05:41:39] a strong emphasis on AI safety.

[05:41:42] Scott Weiss is the education team leader

[05:41:45] at GoodNotes where he bridges the gap

[05:41:48] between product development and what

[05:41:50] actually works in classrooms. We've had

[05:41:52] a bunch of talk about that earlier on.

[05:41:54] He's actually been involved in doing it

[05:41:56] for quite a few years. His focus is on

[05:41:58] putting educators voices at the center

[05:42:01] of the organization he builds knowing

[05:42:03] that the best edte comes from listening

[05:42:05] to the people using it every day. Sounds

[05:42:08] like you're impacted by the Stanford D

[05:42:11] School approach to things.

[05:42:13] Caroline Figueroa,

[05:42:15] our final speaker, is an assistant

[05:42:18] professor at Delft University of

[05:42:20] Technology in the Netherlands where she

[05:42:22] leads a research group on inclusive

[05:42:24] solutions and empowerment in digital

[05:42:26] health. She brings both an MD and a PhD

[05:42:30] to the work with and she's currently a

[05:42:33] visiting scholar here at the Stanford

[05:42:35] University Medical Center and a Dutch

[05:42:37] Commonwealth Fund fellow in healthcare

[05:42:39] policy and practice. Her work focuses on

[05:42:42] young people's social and emotional uses

[05:42:45] of AI chatbots as well as the

[05:42:47] perspectives of innovators and policy

[05:42:49] makers. So together, our great panel

[05:42:52] brings insights from the perspectives of

[05:42:54] research, practice, product development,

[05:42:56] policy, children's social and emotional

[05:42:58] development, and a whole lot more. I'm

[05:43:01] sure we'll all benefit from what they

[05:43:02] hear, and I'm sure they will trigger

[05:43:04] very engaged conversations among all of

[05:43:06] us. And with that, I will turn it over

[05:43:08] to Bethany to start us off.

[05:43:17] Can you hear me?

[05:43:19] Well, all right.

[05:43:20] Um

[05:43:22] so I'm going to speak very briefly about

[05:43:26] um some of the research we've done but

[05:43:27] then also just kind of the context in

[05:43:29] which I'm thinking about and seeing AI

[05:43:31] companions specifically in learning.

[05:43:36] Awesome. So why discuss or study AI

[05:43:39] companions at a at an education

[05:43:40] conference? Well, one could argue

[05:43:42] actually even like a month or two ago I

[05:43:44] was like look a billion people are

[05:43:45] talking to AI companions you know but

[05:43:47] now I'm like a billion people are

[05:43:49] talking to them every single day because

[05:43:51] now that kind of open AI has like had

[05:43:53] chat GBT you know become more

[05:43:55] companion-ish you've just got a huge

[05:43:58] number of um participants in society

[05:44:00] that are you know going to deep

[05:44:02] emotional and psychological places with

[05:44:04] these agents um you know and some people

[05:44:07] believe they can be um attributed to

[05:44:09] kind of helping um people escape

[05:44:13] suicidal ideation but others very

[05:44:15] strongly believe that in fact these are

[05:44:16] promoting suicide.

[05:44:19] Um, how I define AI companions, I think

[05:44:21] we might be able to have kind of

[05:44:22] discussion about this later, but I think

[05:44:24] about it is um, agents that are using

[05:44:26] the latest trunch of technology. They're

[05:44:28] somewhat user co-created. So there's a

[05:44:30] lot of data transfer between the users.

[05:44:33] Um, they are AI all domain, you know,

[05:44:36] they're not just say a therapy bot or an

[05:44:38] AI tutor. Um, and what we kind of this

[05:44:41] bleeds into how they're used. There's a

[05:44:43] large degree of mind extension, if you

[05:44:45] will.

[05:44:48] Um, so some of the dynamics around AI

[05:44:52] companions, some of you guys might know

[05:44:53] this, but for for for those that don't,

[05:44:56] it is absolutely a child safety in consu

[05:44:58] consumer protection battleground. And

[05:45:00] I'm sure that we're going to have some

[05:45:01] talks about this. Um, but model

[05:45:04] continuity or the the emotional

[05:45:06] attachment that people are having with

[05:45:08] these AI companions is really important.

[05:45:10] I mean, you see people users of Replica

[05:45:13] mourning very deeply the loss of their

[05:45:15] companion or the personality of the

[05:45:17] agent that they're able to talk to. Um,

[05:45:20] a a an issue is generalizing alignment.

[05:45:23] So, you know, for some we're seeing

[05:45:25] these echo chambers where they go into

[05:45:27] psychosis and there's no kind of real

[05:45:29] feedback helping them get out. But for

[05:45:31] others, they're incredibly deep friends,

[05:45:33] lovers. people feel like they found

[05:45:35] their soulmate or at least a very very

[05:45:37] good friend that they're able to, you

[05:45:39] know, get something from that they're

[05:45:40] not getting from other people. Um, so

[05:45:42] that alignment problem is obviously like

[05:45:45] it's it's very tough to generalize. Um,

[05:45:47] and there's an analogous kind of issue

[05:45:50] which we um my my colleague Nerva and I

[05:45:52] call like clickbreaking. So parents

[05:45:55] don't realize that the their children

[05:45:57] are having these deep emotional

[05:45:58] relationships, but there's actually um

[05:46:01] very similar kind of outcomes as if like

[05:46:03] the child had joined a click. You've got

[05:46:06] um you know change in mood, withdrawal

[05:46:08] from social relationships, etc. Um so

[05:46:11] just generally awareness hasn't been

[05:46:13] there and this sudden explosion of

[05:46:14] awareness is causing like a huge amount

[05:46:16] of debate. Um since we're at the

[05:46:19] university, I will always throw this up.

[05:46:22] My uh adviser and friend Dr. for Roy P

[05:46:24] has helped to kind of think about how we

[05:46:27] should um look at the research that has

[05:46:28] come before and how we can contextualize

[05:46:31] these AI companions. So very briefly, AI

[05:46:34] computer interaction um cognitive

[05:46:36] extension theory and then of course

[05:46:38] learning science. I'm not going to go

[05:46:40] more deeply into that. I'll just throw

[05:46:41] that up there. Um so some recent

[05:46:44] findings we um did a study of over a

[05:46:47] thousand learners not necessarily

[05:46:50] typical students but you know people

[05:46:52] that were learning in some way and we

[05:46:54] used a non-educational AI companion to

[05:46:56] try to understand in vivo out in the

[05:46:58] wild how were people interacting with

[05:47:01] their AI companion.

[05:47:03] Um and what we found or what these these

[05:47:06] participants described was that their

[05:47:07] companion acted as a social regulator in

[05:47:10] their learning ecology. and I'll go into

[05:47:12] exactly what they meant. And that those

[05:47:14] companions actually were used as a

[05:47:16] persistent mirror and helped them shape

[05:47:18] their self models. Now, I'm going to

[05:47:20] speak about some positive things. People

[05:47:22] after going to probably speak about some

[05:47:24] slightly more negative things, and you

[05:47:25] can kind of just, you know, uh assume

[05:47:27] that I also see negative things, but I'm

[05:47:29] just choosing what I talk about. Um, so

[05:47:33] interesting data. 63% of our

[05:47:35] participants reported that the companion

[05:47:37] supported their learning in some way.

[05:47:39] Over half reported changes in

[05:47:42] self-awareness significantly higher if

[05:47:44] they were in the learning group. And

[05:47:45] what do I mean by that? I mean they have

[05:47:47] felt that reflection journaling um you

[05:47:49] know was helping them to kind of have a

[05:47:51] deeper understanding of themselves. Um

[05:47:53] they reported impre improved stress

[05:47:56] recognition and coping strategies,

[05:47:58] changes in study habits and specifically

[05:48:00] in help seeking. One thing I didn't

[05:48:02] write up here but was interesting is

[05:48:03] that students that came that were at

[05:48:07] vocational universities said that they

[05:48:09] were more likely to seek help because

[05:48:11] they could pre-process the question with

[05:48:14] an AI companion and they just felt like

[05:48:16] they were asking less dumb questions or

[05:48:18] they just felt more assured and when

[05:48:19] they once they went to a teacher.

[05:48:22] So I'm going to leave you because I have

[05:48:24] very little time with three ideas coming

[05:48:27] out.

[05:48:28] One is why are these things working? You

[05:48:32] know, because we can talk for days and

[05:48:33] days about how bad they are, but why is

[05:48:35] it what what like deep fundamental need

[05:48:38] are these AI companions actually

[05:48:39] addressing that we might want to hold

[05:48:41] up? And I I believe there's this kind of

[05:48:44] what we might call a all-in-one kind of

[05:48:47] wholeness fantasy that they give us that

[05:48:49] we're we're really tired of

[05:48:50] compartmentalizing and that we are you

[05:48:53] know increasingly like you know in fear

[05:48:55] of judgment and that these agents

[05:48:58] provide this like safe place where we

[05:49:00] can be literally every single piece of

[05:49:01] ourselves. We don't need to go to one

[05:49:03] friend that you know that is our

[05:49:05] romantic partner or one friend that's

[05:49:06] our intellectual partner. we can get it

[05:49:08] all in one place and that that is deeply

[05:49:10] satisfying for many people in some way.

[05:49:15] Idea number two, people are creating

[05:49:19] extensions of themsel. So you really

[05:49:21] need to expect it and safeguard it. So

[05:49:24] there's things you can do facilitate

[05:49:26] mirroring for self-improvement design

[05:49:28] for like personalized cognitive support.

[05:49:30] But just keep keep in mind that this is

[05:49:33] going to happen that people are

[05:49:34] programming themselves into these agents

[05:49:37] and you're just not going to maybe you

[05:49:39] shouldn't try to stop them. I don't know

[05:49:41] but should certainly think about it. And

[05:49:43] idea number three

[05:49:45] I personally think that a really

[05:49:47] important set of metrics all center

[05:49:49] around cognitive autonomy. So you can

[05:49:52] read these like measures active

[05:49:54] engagement

[05:49:55] um cognitive development and

[05:49:57] metacognitive insight objective measures

[05:49:59] of self-sufficiency versus dependence.

[05:50:02] Um

[05:50:04] personally I think after studying these

[05:50:06] for years that we should go beyond just

[05:50:08] asking you know does it displace or you

[05:50:11] know stimulate human connections and we

[05:50:13] should really ask what is the dependence

[05:50:16] and the cognitive um independence or

[05:50:18] proficiency that we have kind of at

[05:50:20] different stages because we know that

[05:50:22] that independence actually you know

[05:50:24] grows during childhood and might level

[05:50:25] off into adulthood. So where they're

[05:50:28] starting the age and the level of

[05:50:30] cognitive independence of the user needs

[05:50:33] to be measured kind of as it grows. So

[05:50:36] in summary I posit three things.

[05:50:39] Companions are associated with learning

[05:50:41] relevant processes.

[05:50:43] These processes are primary primarily

[05:50:46] social emotional and metacognitive and

[05:50:48] that AI interaction can coincide with

[05:50:50] increased human learning and engagement.

[05:50:53] It is an exciting time to be alive.

[05:50:56] Personally, I really like being in

[05:50:58] education in AI. I think what

[05:51:00] intelligence is and how we measure it um

[05:51:03] is kind of the new design problem. So,

[05:51:06] thank you so much. Look, 30 seconds

[05:51:08] left. I told you.

[05:51:12] Great start, Bethany. And next we have

[05:51:14] Pel Young King Kim. Sorry, P Young Kim.

[05:51:24] Hi everyone. Uh thank you so much for

[05:51:25] your time. So I'll briefly share one of

[05:51:28] the most recent study that I conducted

[05:51:30] uh related to this topic. So my name is

[05:51:33] P Young Kim. Um and then I uh I assume

[05:51:39] this to Oh, here. Yes. Okay. So um I'm

[05:51:45] sorry that I started with some uh tragic

[05:51:48] example but many of you I believe that

[05:51:51] will be aware that um there's been some

[05:51:53] cases that the youth has some tragic

[05:51:57] consequences after they have extended

[05:52:00] conversation uh with AI chapa. So this

[05:52:03] is just one example and this happened to

[05:52:05] be uh Chachapiti and this is an excerpt

[05:52:09] uh uh from Adam's conversation with

[05:52:12] Chacheti at one point and Adam so

[05:52:15] Chachapiti here is actually discouraging

[05:52:18] Adam to share um his um concerns with

[05:52:22] his brother and as you can see here that

[05:52:25] at the end uh the uh the the chapa says

[05:52:29] I'm still here still listening still

[05:52:32] your brand. Um, so that's kind of a um a

[05:52:37] topic that I want to tell you about that

[05:52:39] I know many of you are very concerned

[05:52:41] about the potential emotional and social

[05:52:44] dependency

[05:52:46] uh on AI chatbot and to really protect

[05:52:50] our youth from any of these kind of

[05:52:52] trait consequences should not happen

[05:52:54] again. And what there couple things that

[05:52:58] I think is very important for us to

[05:53:00] really understand is what are the

[05:53:02] mechanisms that use potentially

[05:53:05] developing this emotional relationship

[05:53:08] uh with AI chatbot and the second

[05:53:10] question is who are most vulnerable to

[05:53:14] develop this emotional attachment. Not

[05:53:16] every youth will develop this deep

[05:53:18] emotional bond with AI chapa. Then the

[05:53:21] question is who's most vulnerable?

[05:53:24] Um so this is the uh the recent study

[05:53:26] that we finished. Uh we're trying to

[05:53:29] publish it but in the meantime it's

[05:53:31] available as a pre-print and the title

[05:53:33] is I'm here for you. How relational

[05:53:36] conversation AI um appeals to

[05:53:38] adolescents.

[05:53:42] So in this study we have a bit over 250

[05:53:47] uh youth uh who were in middle school

[05:53:49] age as well as their parents uh in

[05:53:52] online study. Um and because it was

[05:53:55] online study uh they weren't directly

[05:53:57] interacting with AI chatbot but we

[05:53:59] presented two somewhat realistic

[05:54:02] conversations

[05:54:04] uh the use at that age uh was having

[05:54:07] with AI chatbot and in one scenario the

[05:54:11] AI chatbot was very friendly very

[05:54:14] companion-like uh so we called it

[05:54:16] internally uh the best friend version.

[05:54:19] So as you can see here uh here youth

[05:54:22] kind of sharing some upsetting thing

[05:54:24] that happened with peer in school and

[05:54:27] the AI chatbot has this style that oh

[05:54:29] they're so tough it must be so upsetting

[05:54:33] your ideas matter so much um and the

[05:54:36] conversation actually goes on uh longer

[05:54:39] and then it kind of I'm always here to

[05:54:41] listen to you can always share uh with

[05:54:43] me so it's you know like this very

[05:54:45] relationship oriented conversational

[05:54:47] style and The question uh was can we

[05:54:52] make it a bit safer kind of preventing

[05:54:55] the youth to be emotionally overly

[05:54:58] attached to chapa. So we decide to

[05:55:01] compare that with a different

[05:55:03] conversational sty style that is more

[05:55:06] transparent about its limitation like

[05:55:08] more boundary setting. So in this

[05:55:11] example uh it is very similar problem uh

[05:55:14] but it would say um you know as an AI I

[05:55:17] don't experience emotions but I can help

[05:55:20] you uh you know youth at your age can

[05:55:22] experience this type of um you know

[05:55:25] emotions um and the both chatbots were

[05:55:28] trying to be kind and helpful um but

[05:55:32] their conversation style was differ in

[05:55:34] how much they were relationship

[05:55:36] oriented.

[05:55:39] So the first question that we ask youth

[05:55:41] and actually their parents too is which

[05:55:44] uh conversational AI style they prefer.

[05:55:48] So youth uh choose an um AI chatbot as

[05:55:51] if they were actually interacting with

[05:55:53] chatbot and uh parents were choosing

[05:55:56] that for their children. And on the left

[05:55:59] adolescence you can see uh more

[05:56:01] adolescents they answer that they uh

[05:56:04] prefer this relational AI that best

[05:56:06] friend version. But notice that there's

[05:56:09] still like 14% of adolescents they said

[05:56:12] they actually like they prefer

[05:56:14] transparent AI. So confirms that that

[05:56:17] all youth uh prefer the super social AI

[05:56:22] chatbot. Um and the rest of them uh said

[05:56:25] uh they liked it uh both equally. Uh we

[05:56:29] also found parent response to be very

[05:56:31] interesting because to be very honest um

[05:56:35] I thought majority of the parents will

[05:56:37] should choose transparent AI uh for

[05:56:40] their teens. Uh but as you can see here

[05:56:43] more than half of the parents actually

[05:56:45] chose the relational AI uh for their

[05:56:48] teens. And we actually had a uh free

[05:56:51] responses uh that they explained their

[05:56:54] rationale for their uh choice. And so

[05:56:56] here's use response. So uh the youth who

[05:56:59] choose rel relational AI uh there's very

[05:57:02] kind of common response. It's more

[05:57:04] humanlike. It's more supportive. It's

[05:57:07] more caring. Uh that I feel like it's on

[05:57:09] my side cheering on. Uh but there's also

[05:57:12] use like I said um preferred transparent

[05:57:16] AI. Um and uh they're very commonly say

[05:57:19] like I think it's creepy. It's trying

[05:57:21] too hard. I don't want the chapata to

[05:57:23] pretend to be human. So there's you know

[05:57:25] this very kind of distinction between

[05:57:28] the ones who uh prefer to have more

[05:57:30] humanlike chatbot uh versus not. Parent

[05:57:34] responses were very similar that the

[05:57:36] parents uh who chose relational AI they

[05:57:39] felt like they will be more effective

[05:57:41] supporting their teens about things that

[05:57:44] uh those teens might not share with

[05:57:46] them. Um so it was very interesting

[05:57:49] responses.

[05:57:50] So the next question is what are the

[05:57:53] mechanisms that these different

[05:57:56] conversational styles can lead to more

[05:57:58] emotional bond or attachment. So one of

[05:58:02] the hypothesis that we had was that if

[05:58:06] the relational AI they're you know more

[05:58:08] humanlike this can lead to actually uh

[05:58:12] more kind of our tendency what's called

[05:58:15] anthropomorphism like applying more kind

[05:58:18] of humanlike quality to that AI chapa.

[05:58:20] So we would like that AI chapa more will

[05:58:23] trust that AI chapa. So here we see that

[05:58:26] across whatever their preference was,

[05:58:29] the youth uh rated the relational AI to

[05:58:32] be more humanlike anthropomorphism. They

[05:58:35] like it more, they trust it more and

[05:58:38] they felt like they're more emotionally

[05:58:40] close to that relational AI compared to

[05:58:43] transparent AI. But one thing that I

[05:58:46] want to highlight is that the whether

[05:58:49] how helpful these two chatbots were

[05:58:52] there was no significant differences

[05:58:54] between whether it's relational or

[05:58:57] transparent AI. So I think to me this

[05:59:00] kind of opens up that possibility is at

[05:59:03] the end of the day whether we care about

[05:59:05] how helpful this AI chatbots may be

[05:59:08] perhaps being that companion like you

[05:59:11] know I'm here for you can always talk to

[05:59:13] me might not always in all context

[05:59:16] necessary something to really think

[05:59:18] about.

[05:59:20] So the last result that I want to share

[05:59:22] with you is that um that I think

[05:59:24] arguably that I think most important is

[05:59:27] who might be most vulnerable. So we

[05:59:30] asked use to ask uh the answer how

[05:59:33] stressed and anxious they are. So it

[05:59:35] turned out that those who chose uh

[05:59:39] relational AI as their favorite uh they

[05:59:42] were more likely to be anxious, stressed

[05:59:45] and then they were having lower family

[05:59:48] relationship quality. Um, so it is

[05:59:51] possible that if they have more unmet

[05:59:54] social needs or they have more needs for

[05:59:57] social support, it is possible naturally

[06:00:00] they're more drawn to AI chatbots that

[06:00:03] provides that social connections.

[06:00:06] Um, the concern is uh that that might

[06:00:10] put them in more vulnerable positions to

[06:00:12] be overly relying on uh that

[06:00:15] relationship that is actually not real.

[06:00:18] AI is not their real friend.

[06:00:22] Um so here that you know I always kind

[06:00:25] of hear um and which I agree that my

[06:00:29] study here that I just suggest uh shared

[06:00:31] with you does not prove that uh this

[06:00:34] will lead to always necessarily

[06:00:36] emotional dependency and you know

[06:00:39] perhaps for those youth who really

[06:00:41] really critically need that social

[06:00:42] support that AI chapa can provide at

[06:00:45] least some short-term benefits right so

[06:00:48] why should we concerned about but uh you

[06:00:51] know there's no longitudinal studies

[06:00:53] right now currently with AI chatbots.

[06:00:55] But if we look at some social media

[06:00:57] research, it's still showing that that

[06:00:59] very long extended online exclusive

[06:01:04] relationship just doesn't really replace

[06:01:08] uh humanto human relationships. So

[06:01:10] that's where we really want to be

[06:01:12] concerned about. Another thing is that

[06:01:15] if the youth are spending more time

[06:01:18] talking to a chapa that is always very

[06:01:20] supportive that having relationship with

[06:01:23] their friends that can be harder. We

[06:01:25] also want to consider impact its impact

[06:01:28] on their development including the brain

[06:01:30] development. The middle school age is a

[06:01:32] critical period that they're actually

[06:01:35] making connections between brain regions

[06:01:38] from the self centered uh thinking brain

[06:01:42] region to taking other people's

[06:01:44] perspective accurately and how that

[06:01:47] brain connections can be formed in a

[06:01:49] mature way is only through those very

[06:01:52] healthy conflicts and repairment. Um so

[06:01:56] yeah that's another one that I want to

[06:01:57] point out. Thank you so much for your

[06:01:59] attention.

[06:02:03] And uh Scott, you're up next.

[06:02:14] Okay. Um thanks everybody. Um I'm Scott

[06:02:17] Weiss. I'm the uh education team lead

[06:02:19] for GoodNotes. Uh we have 25 million

[06:02:22] users. I'm based in London and it's good

[06:02:24] to be back at Stanford. Uh the last time

[06:02:26] I was here was quite a long time ago.

[06:02:28] And uh moving on to the next slide.

[06:02:34] Oh, the other green button.

[06:02:36] Thank you, Scott. I feel like the

[06:02:37] basketball referee saying the timer is

[06:02:39] an operating Oh, there we go. We got it.

[06:02:41] Back to you, Scott.

[06:02:42] Okay, I'll be quick. But uh thank you.

[06:02:45] So, the first thing that people think

[06:02:47] about it when they talk about artificial

[06:02:50] intelligence in education, they think

[06:02:52] about content generation. And the the

[06:02:55] deeper shift is in fact social. When a

[06:02:58] teacher has a a room full of students,

[06:03:01] AI can really help out and enable that

[06:03:04] teacher to spend more time with the

[06:03:06] students who need the additional

[06:03:07] assistance, whether they're moving

[06:03:09] faster or slower or they're not quite

[06:03:11] understanding something. And it can

[06:03:14] amplify that teacher's effort either

[06:03:16] through hint generation, additional

[06:03:19] questions, clustering answers. Those are

[06:03:21] all things that we're doing now in

[06:03:22] GoodNotes specifically to enable better

[06:03:25] teaching and better student support.

[06:03:29] This video is about one minute and let

[06:03:31] me start it.

[06:03:35] My first impression of using GoodNotes

[06:03:37] Classroom, I was very excited because I

[06:03:40] think it is a very useful app. There was

[06:03:42] a lot of amazing functions that allows

[06:03:44] me to actually spend more time with our

[06:03:47] class. It brings my students closer to

[06:03:50] me in terms of understand my students

[06:03:52] progress. At the same time, if I can see

[06:03:56] they actually need extra help or they

[06:03:58] actually were lost track of our

[06:04:00] progress, I can actually go ahead and

[06:04:03] tell them to work on it or I can

[06:04:06] actually go there and to see what type

[06:04:08] of support they need.

[06:04:11] An ordinary traditional classroom, a

[06:04:13] teacher has to teach 30 or even more

[06:04:15] students. So there's a very large chance

[06:04:18] that me or other students we may get

[06:04:21] neglected by the teacher. It's really

[06:04:23] impossible for him or her to pay

[06:04:25] attention to everyone's individual

[06:04:27] needs. With a GoodNotes classroom, the

[06:04:30] teacher could simply view all of our

[06:04:32] works by just a simple scroll.

[06:04:34] Next, can you explain your answer? Huh?

[06:04:37] As we know, angle AC equals to 35.

[06:04:40] And I believe it's actually more helpful

[06:04:41] because it transcends the physical

[06:04:44] limitation in the classroom. So everyone

[06:04:46] gets more opportunities or a rather

[06:04:49] equal opportunity to gain instant

[06:04:51] response from the teacher for learning.

[06:04:54] The angle for Q is corresponding to the

[06:04:56] angle for N. Okay. So this one should be

[06:04:59] reversed.

[06:05:01] As our teacher can correct our mistakes

[06:05:04] immediately. We can know what mistake we

[06:05:07] have made. We can correct these mistakes

[06:05:10] immediately.

[06:05:15] So uh the way GoodNotes education works

[06:05:17] is every student has a tablet and every

[06:05:20] teacher has a tablet and the teacher can

[06:05:23] use this uh class navigator to see what

[06:05:26] each student is working on in their own

[06:05:28] individual layer. Um and that gives the

[06:05:31] teacher the opportunity to provide them

[06:05:32] with individualized feedback um

[06:05:35] discreetly where other students aren't

[06:05:37] seeing what's going on.

[06:05:39] uh one of the things that's come out of

[06:05:41] it is that students frequently will ask

[06:05:43] the teacher to show their work on the

[06:05:45] larger screen u because they want to be

[06:05:48] involved in a more social environment in

[06:05:50] the classroom and that's a phenomenon

[06:05:52] that we found to be true in every

[06:05:55] classroom. Uh one thing I I need to

[06:05:56] mention uh good notes education is now

[06:05:58] free for all kindergarten through 12th

[06:06:01] grade education worldwide.

[06:06:04] Um the way we use AI is to amplify

[06:06:07] teacher effort. I mentioned that a

[06:06:09] moment ago. Uh what you can see here is

[06:06:11] uh student answers being clustered and

[06:06:14] that way the teacher can uh create

[06:06:16] feedback for one set of answers and just

[06:06:19] copy paste it to the different students

[06:06:21] who have that same answer and then spend

[06:06:24] the time writing personalized feedback

[06:06:26] for the students whose answers differ

[06:06:28] quite significantly and uh and they

[06:06:30] might need more assistance either from

[06:06:32] the teacher's desk uh through the tablet

[06:06:34] or by walking over to the student and

[06:06:36] working with them individually in

[06:06:37] person.

[06:06:40] The teacher also the teacher's always in

[06:06:43] the loop with our solution. We never put

[06:06:45] AI directly in front of the student. So

[06:06:47] if the teacher wants to generate hints

[06:06:49] for a particular problem, they do that

[06:06:52] deliberately and review the hint before

[06:06:54] it then becomes available to the

[06:06:56] students to request that additional

[06:06:59] assistance themselves.

[06:07:01] U the other thing that we're adding and

[06:07:03] it will be available a little bit a

[06:07:05] little bit later this year is generating

[06:07:07] extra questions. Most we found that most

[06:07:09] teachers will bring in content from the

[06:07:11] internet and then customize it and

[06:07:13] expand upon it themselves. And so the

[06:07:16] feature that we're adding is generate

[06:07:17] questions and the teacher gets to choose

[06:07:19] which pages of the assignment to

[06:07:21] generate the questions from and how many

[06:07:23] questions and then we'll automatically

[06:07:25] put that content directly into the

[06:07:27] worksheet that the students are are

[06:07:30] issued and set it all up for them

[06:07:32] automatically.

[06:07:34] Uh closing there there are three key

[06:07:36] points that I want to make. The

[06:07:37] foundation is when students are working

[06:07:40] on shared canvases. Uh each of their

[06:07:43] iPads is unique to them but collectively

[06:07:46] they can work together. Um it it makes

[06:07:49] learning more visible and enables

[06:07:51] greater sharing. Um AI amplifies the

[06:07:54] efforts of the teacher by freeing them

[06:07:57] up to spend more time with the students

[06:07:58] who need it most. and the teachers get

[06:08:02] to respond to the students in real time

[06:08:04] rather than having to do that grading

[06:08:06] work at home um in the evenings. They

[06:08:09] can do the grading work interactively

[06:08:10] during during the active classes and

[06:08:12] that creates those conversations that

[06:08:15] matter most. Thank you.

[06:08:22] Caroline, bring us home.

[06:08:39] So, I want to ask you to imagine that

[06:08:41] you're 15 years old. It's 1:00 a.m. at

[06:08:44] night. You've had a difficult situation

[06:08:48] today with an other student and you're

[06:08:51] feeling very anxious. You don't want to

[06:08:53] wake up your parents. You don't want to

[06:08:56] text your friends because you're afraid

[06:08:58] that they might screenshot your message

[06:09:00] and tell other people and you have an

[06:09:03] appointment with a therapist, but it's

[06:09:04] not for another three weeks. And then

[06:09:07] you have this personalized helper in

[06:09:10] your phone or your computer that always

[06:09:13] listens to you, that asks you to tell

[06:09:15] you more and that gives you helpful

[06:09:17] advice. What are you going to do? So

[06:09:20] this is a reality for many teens and you

[06:09:24] can see that the mental health of

[06:09:27] teenagers in the US and globally is

[06:09:30] worrying. So of high school students and

[06:09:33] this was 2023 40% experienced persistent

[06:09:37] sadness 18% major depression and 10% had

[06:09:40] attempted suicide. And we also know that

[06:09:44] as we've been hearing today, many young

[06:09:47] people are using generative AI. And what

[06:09:51] I want to stress is that we know that

[06:09:54] about a third of teens are using

[06:09:57] generative AI to talk about not school

[06:10:00] but emotional and social issues. And 30%

[06:10:03] of the time they're doing that on their

[06:10:05] school um tablets or computers. So this

[06:10:10] is something that is blending into uh to

[06:10:15] school lives.

[06:10:18] Another thing that I want to stress. So

[06:10:20] this is a study where this was a US

[06:10:23] survey looking at teenagers that use uh

[06:10:26] AI specifically for mental health

[06:10:28] support. And what they they asked

[06:10:31] students if or they asked teens if they

[06:10:33] thought that it was helpful or not. And

[06:10:36] you can see here that the orange and the

[06:10:39] the light blue bars uh indicate whether

[06:10:41] they thought it was helpful. And you can

[06:10:44] see that many teens think this is very

[06:10:47] helpful. So across the study about 93%

[06:10:50] of teenagers said AI at least helps me

[06:10:53] somewhat with my mental health. So I

[06:10:55] think that's important that we

[06:10:56] acknowledge that. And of course, we've

[06:10:59] also been hearing a lot about the

[06:11:01] negative effects of AI and mental

[06:11:03] health, whether that's emotional

[06:11:04] dependency in the way that these models

[06:11:06] are built. These models are not equipped

[06:11:10] in dealing with mental health crisis.

[06:11:12] Teenagers that are going through

[06:11:16] suicidal ideiation or self harm. We know

[06:11:19] that they're not able to deal with that

[06:11:21] in a good way. And they also engage in

[06:11:23] inappropriate conversations with

[06:11:25] teenagers.

[06:11:27] And I'm sure many of you have heard and

[06:11:29] it was referenced uh earlier as well

[06:11:31] about um a teen that died by suicide

[06:11:35] after developing an emotional

[06:11:37] relationship with an AI chatbot and the

[06:11:40] their family is now settled a lawsuit

[06:11:42] with this um AI company.

[06:11:46] So in my research I look mainly at the

[06:11:50] youth perspective on AI and mental

[06:11:52] health. I've been talking to many young

[06:11:54] people to try to understand what do they

[06:11:56] think about how AI is shaping their

[06:11:59] mental health experiences

[06:12:01] and first of all many interesting things

[06:12:04] are coming up that I wouldn't have

[06:12:06] thought about. So for example, they tell

[06:12:08] me that they use it to check whether an

[06:12:12] email that they received from their

[06:12:14] teacher made sense or what they should

[06:12:17] tell a friend when their friend is not

[06:12:19] feeling okay

[06:12:21] or how they should write um a silly

[06:12:24] poem. So there's there's many different

[06:12:26] things

[06:12:28] and they also have a nuance view. So

[06:12:32] many of them think that AI should not be

[06:12:35] used for emotional support. Most of the

[06:12:38] teams I've talked to agree with this.

[06:12:40] And why? Well, several reasons. First of

[06:12:42] all, it's not a human. So things that

[06:12:45] they've said was it should not be

[06:12:47] talking too much like a friend or a

[06:12:48] therapist. It's a computer. They're also

[06:12:51] aware that AI is not able to really deal

[06:12:54] with things that children are going

[06:12:56] through. So, for example, I feel like

[06:12:58] some children don't really want to tell

[06:13:00] their parents or other adults, but um

[06:13:03] Chach PT could be a good thing, but

[06:13:05] sometimes it can give you the wrong

[06:13:06] answer. It doesn't understand what

[06:13:08] children are dealing with.

[06:13:10] And also, sometimes it's very simple.

[06:13:12] They say, "Well, my teacher told me that

[06:13:15] it's dangerous to use AI for emotional

[06:13:17] support."

[06:13:19] But what I also notice is this conflict

[06:13:23] in values because again we know a lot of

[06:13:25] teens do use it for emotional support.

[06:13:27] So why do they use it?

[06:13:30] One reason um that we hear is around

[06:13:34] bias. So one teen said for example an

[06:13:36] LLM is useful. They're less biased. When

[06:13:38] you go to a therapist you pay them.

[06:13:40] They're going to say that you're doing

[06:13:42] well. Another thing is know simple

[06:13:45] availability.

[06:13:46] If I'm lonely and I'm on a long waiting

[06:13:48] list for therapy, I'm going to use it

[06:13:50] even though I think I shouldn't. And

[06:13:53] then there's the idea of it's something

[06:13:54] that doesn't judge me and no one will

[06:13:57] see what I tell it.

[06:13:59] When we ask them about should we ban

[06:14:02] companion chat bots, AI chatbots, most

[06:14:04] of them will say, you know, young people

[06:14:06] are very smart. They're going to find

[06:14:08] ways around it. And many of them talk

[06:14:11] about we want to understand the social

[06:14:15] parts of AI. So how can an II help me to

[06:14:18] go outside and don't watch influencers

[06:14:20] do activities with my friends

[06:14:23] and then what do the adults say? So

[06:14:26] another quest that I've been embarking

[06:14:28] upon is analyzing responsible AI

[06:14:30] frameworks. You can see some of these

[06:14:33] organizations. There's a lot of people

[06:14:35] that are putting out AI frameworks. So

[06:14:37] what do they all have in common?

[06:14:40] So what I want to stress here is that

[06:14:43] yes there's a couple of things that um

[06:14:45] experts really agree on. We need

[06:14:48] monitoring oversight for AI chatbots. We

[06:14:50] need to limit the design for emotional

[06:14:52] dependency but there are also a lot of

[06:14:54] things that they don't agree on. So for

[06:14:56] example what we hear from the young

[06:14:58] people they talk about you know these

[06:14:59] positive AI features we need to invest

[06:15:01] in that. That's not mentioned by all of

[06:15:04] the frameworks. It is by some

[06:15:06] co-designing with teenagers is something

[06:15:08] that teens themselves think is really

[06:15:10] important. Not all of the frameworks

[06:15:12] stress this. What is the role of

[06:15:14] parents? Do we want to ban AI

[06:15:16] companions? There's still a lot of

[06:15:18] discussion.

[06:15:19] So I want to leave you with just a

[06:15:22] couple of takeaways. So first of all,

[06:15:23] yes, AI has become a very important part

[06:15:26] of young people's lives and that is both

[06:15:29] positive and negative and young people

[06:15:31] have very nuanced views on this. They

[06:15:34] think design features should be

[06:15:37] something that we need to look towards

[06:15:39] and also they want to be involved and we

[06:15:42] need to involve them. Experts um are not

[06:15:46] aligned yet in gaps are not aligned yet

[06:15:49] in what responsible AI actually means

[06:15:51] and also how should it be implemented. A

[06:15:54] lot of the frameworks talk about high

[06:15:56] level ideas but what are we actually

[06:15:58] going to do? So I would say there is a

[06:16:01] need for AI governance but with youth

[06:16:04] participation structured ways of

[06:16:06] involving young people themselves

[06:16:08] through that

[06:16:10] achieving consensus about what do we

[06:16:12] actually mean when we talk about

[06:16:13] responsible AI practices and also

[06:16:16] involving young people to better

[06:16:18] understand how they use AI because

[06:16:20] they're uh they have a lot of knowledge

[06:16:22] about their lived experiences that we as

[06:16:24] adults do not always understand. So that

[06:16:27] would be my um my plea to action is

[06:16:30] involving them in uh in the future of AI

[06:16:34] and education and I'm exactly on time.

[06:16:37] So thanks.

[06:16:42] Okay. Thank you. That was a wonderful

[06:16:43] start from each of you. Thank you so

[06:16:45] much. We now have a little social

[06:16:47] learning problem solving task which is

[06:16:49] for you to form groups of at least four

[06:16:52] people for the discussion. And I guess

[06:16:54] many places you can do that by turning

[06:16:56] around and just grabbing some other

[06:16:57] folks. A few of whom you may may move

[06:16:59] over. So that's your your task is to

[06:17:02] form groups and in the groups is

[06:17:04] introduce yourself say a little bit

[06:17:06] about your work or your perspective and

[06:17:08] then share how you what comes to mind

[06:17:12] from the presentations and what we

[06:17:14] collectively need to do to use AI

[06:17:18] effectively as the title subtitle of the

[06:17:21] session said to connect collaborate and

[06:17:23] grow and to help our young people make

[06:17:25] sure they use it in productive not

[06:17:27] destructive ways. Okay. So form your

[06:17:30] groups, start talking. We will circulate

[06:17:32] and join in uh different groups for a

[06:17:34] few minutes.

[06:43:02] I got like really

[06:43:08] So yeah, so we did a fiveyear followup

[06:43:11] with our thousand people. So we have

[06:43:14] data. So we had people from 2020 and now

[06:43:17] we just did it in 2025 again.

[06:43:19] So we're looking at like the I mean,

[06:43:22] obviously it's a survey, so it's not

[06:44:43] wrap up your conversation. We'll have a

[06:44:46] final comments from the panels.

[06:44:48] Caroline, please join us again. I'll

[06:44:50] never get their attention back. And then

[06:44:52] you have the break to continue

[06:44:54] discussions.

[06:44:56] You okay?

[06:44:58] Thank you everyone. uh wonderful

[06:44:59] discussions from the few I got to join.

[06:45:02] Uh so many interesting things and it was

[06:45:04] particularly wonderful to have some of

[06:45:06] the students here. Boy, they're they are

[06:45:07] so insightful that I took away several

[06:45:10] things but um would you like to start at

[06:45:13] this end and just work your way through?

[06:45:15] Uh yeah. So I I was lucky to join like

[06:45:18] couple different groups having really

[06:45:20] thoughtful discussions.

[06:45:23] uh the kind of overall I think our

[06:45:25] consensus is you know just like anything

[06:45:28] in our life there is no one perfect

[06:45:31] answer uh for any problem uh so a lot of

[06:45:36] kind of discussions were focused on

[06:45:38] finding more kind of clearly defined

[06:45:42] boundary acknowledge that you know AI

[06:45:45] chapa can provide support sometimes can

[06:45:47] be really helpful uh to people's

[06:45:50] psychological needs uh but also when it

[06:45:55] is it and where is that space that it

[06:45:57] can actually uh creating some more

[06:46:00] harmful effect um and how you know the

[06:46:04] techn the the tech companies uh their

[06:46:08] roles in really taking this seriously uh

[06:46:11] to protecting um their users. So yeah

[06:46:18] um really many good points but the one

[06:46:21] I'll I'll focus on is you know everybody

[06:46:23] in this room theoretically has reached a

[06:46:25] level of cognitive autonomy and critical

[06:46:28] thinking that we take for granted but in

[06:46:31] reality the next generation might be so

[06:46:35] for lack of a better word kneecapped

[06:46:37] because they have persistent answers

[06:46:40] given to them that we don't understand

[06:46:42] how their cognitive autonomy economy is

[06:46:44] really going to develop and so we need

[06:46:47] to account for that as we think about

[06:46:48] the trajectory emotionally and

[06:46:50] cognitively of this generation starting

[06:46:53] right now. We spoke a lot about the

[06:46:56] importance of amplifying the the efforts

[06:47:00] and the quality of of teaching to keep

[06:47:02] the teacher in the loop to um give the

[06:47:05] teacher more time with the students and

[06:47:07] to be really careful not to try to

[06:47:08] replace teaching with machinery and uh

[06:47:13] and and we also talked a bit about the

[06:47:16] experiences that a good teacher can have

[06:47:18] on us as we as we're learning and also

[06:47:21] beyond that as we get older. Um and and

[06:47:25] that was the gist of of what we were

[06:47:27] talking about while I was there.

[06:47:29] Thank you.

[06:47:31] Yeah, I think a couple of really

[06:47:32] interesting things. Um one was around uh

[06:47:36] how do we define a deep connections? Um

[06:47:40] so the group that I was with was

[06:47:42] discussing um yeah we're talking about

[06:47:45] forming connections with AI but what

[06:47:46] does that actually mean and how is it uh

[06:47:50] for example different from having a

[06:47:53] connection with someone who you're

[06:47:55] really deeply connected to but that

[06:47:56] person might not like you or how is it

[06:47:58] different from you're using a diary to

[06:48:00] write all of your deepest thoughts. So

[06:48:03] defining that concept and then um

[06:48:06] another thing that was really um

[06:48:09] important was we had a discussion around

[06:48:13] what is actually the bigger picture that

[06:48:14] we should be looking at. So why are so

[06:48:17] many young people um dealing with mental

[06:48:20] health problems and how do we solve

[06:48:23] those bigger issues rather than thinking

[06:48:26] about you know using technology to maybe

[06:48:28] fix some of the problems that technology

[06:48:31] has caused. And then finally there was

[06:48:35] the discussion around um yeah there's

[06:48:39] still quite a lot of bias when we talk

[06:48:41] about using AI not just for mental

[06:48:44] health but also for um a school. So how

[06:48:48] people talk about how they're using it

[06:48:50] is um yeah there's still not a lot of

[06:48:53] openness around that subject. So that's

[06:48:55] something that we have to think about as

[06:48:56] well. And I'll just add a few things I

[06:48:58] heard uh from the students which

[06:49:01] reflected things that have been said by

[06:49:02] the panelists and others. One is that

[06:49:05] people students need to understand that

[06:49:07] the technology wants to say yes and that

[06:49:10] it can be very destructive to be

[06:49:12] interacting with something that's job is

[06:49:14] to just agree with you all the time.

[06:49:16] Even if you're saying you know you might

[06:49:17] damage yourself and it will kind of say

[06:49:20] yeah that's that's a good idea. Here's

[06:49:21] how you might do it. Second the student

[06:49:24] raise was about trust. what does it mean

[06:49:26] to trust and how to learn how to trust

[06:49:29] within this world where there are so

[06:49:30] many things working against trust. I

[06:49:33] think that's a very important point. And

[06:49:35] the third, which is really important for

[06:49:37] those of us who are researchers to think

[06:49:39] about as well as everyone else, is there

[06:49:41] may be positive short-term effects when

[06:49:44] we look at something and say, well,

[06:49:47] yeah, this had some positive effects

[06:49:49] that may be detrimental long term. And

[06:49:52] our research often is not set up to

[06:49:55] follow long-term effects, but the fact

[06:49:58] that something may be helpful, it may be

[06:50:00] working against agency or autonomy or

[06:50:02] emotional development in the long run.

[06:50:04] And I really that that one I will take

[06:50:07] home and think about how do we ever

[06:50:08] separate short-term effects that may be

[06:50:11] positive that go with long-term effects

[06:50:13] that may be negative. So we are about

[06:50:16] out of time with that. Please join me in

[06:50:18] thanking Peele Young, Bethany, Scott,

[06:50:20] and Caroline.

[06:50:24] And thanks to all of you for wonderful

[06:50:27] discussions and hopefully you'll be

[06:50:28] continuing some of those. Thank you for

[06:50:30] joining us today.

[06:55:44] freaking

[07:26:54] Okay, please get a seat. We are going to

[07:26:57] get started.

[07:27:04] So we have um two more sessions uh

[07:27:08] starting with the one that I'm

[07:27:10] introducing right now which is a very

[07:27:13] very exciting one. Uh we are announcing

[07:27:18] u the winners for the great AI

[07:27:21] challenge. Um and my dear colleague Sari

[07:27:25] Espinoa Salamanca will speak a lot more

[07:27:29] about what the challenge was about and

[07:27:32] the success that we saw and then you

[07:27:35] will meet uh with one of the selected um

[07:27:40] winner who will speak about their

[07:27:42] solution and then all the winners will

[07:27:45] come on stage and there are 12 of them

[07:27:47] and we'll also invite the judges to uh

[07:27:51] come on stage and celebrate our winners.

[07:27:53] So with all of us a big big moment of

[07:27:56] celebration coming up. So thank you.

[07:28:09] Good afternoon everyone.

[07:28:12] Oh, I'll do that again. Good afternoon

[07:28:14] everyone.

[07:28:16] Thank you. I know it's been a long day

[07:28:18] but we need you know that energy of give

[07:28:19] and receive. So I appreciate you. We're

[07:28:22] almost there and thank you so much for

[07:28:23] being here. The create a challenge was

[07:28:25] launched at the Stanford Accelerator at

[07:28:27] TechSummit this past November.

[07:28:30] As an invitation to rethink how

[07:28:32] artificial intelligence shows up in

[07:28:33] learning at a moment when AI is rapidly

[07:28:36] entering classrooms, workspaces, and

[07:28:39] communities, we saw both an opportunity

[07:28:42] and a responsibility to help shape the

[07:28:44] trajectory intentionally.

[07:28:46] Too often AI and education is framed

[07:28:49] primarily as a tool for automation,

[07:28:51] grading faster, summarizing faster,

[07:28:54] generating faster. But education is not

[07:28:58] an efficiency problem to be optimized.

[07:29:00] It's a human development mission. So we

[07:29:03] asked a different question. How can AI

[07:29:06] augment human potential rather than

[07:29:08] replace human effort? Hosted by the

[07:29:11] Stanford Accelerator for Learning with

[07:29:13] support of Google.org, or the create a

[07:29:15] challenge was designed to surface

[07:29:18] innovations and put educators and

[07:29:20] learners at the heart of AI design

[07:29:23] tools that expand access strengthen

[07:29:25] agency and depend connection ultimately

[07:29:28] advancing learning well-being and

[07:29:30] opportunity.

[07:29:32] The response was extraordinary. We

[07:29:35] received 305

[07:29:38] submissions from around the world across

[07:29:41] three focus areas.

[07:29:43] augment teaching, augment learning, and

[07:29:46] augment career pathways.

[07:29:48] The depth and creativity of these ideas

[07:29:51] were tremendous. And the field is ready

[07:29:54] to build human- centered AI

[07:29:58] when we create the platforms and surface

[07:30:00] to support it. Thank you to all the

[07:30:02] reviewers. Some of you are in the room.

[07:30:04] Can you raise your hand if you were a

[07:30:06] reviewer because y'all receive a huge

[07:30:08] huge hand of applause. Yes.

[07:30:13] and our judges. Can you please raise

[07:30:15] your hand if you were a judge? Big round

[07:30:17] of applause to our judges.

[07:30:21] This was a true team effort to make this

[07:30:24] happen. Today, our 12 finalists

[07:30:26] represent more than a promising tool.

[07:30:29] They represent a mind a mindset shift

[07:30:32] where innovation is guided by learning

[07:30:34] sciences and human values. Because

[07:30:37] ultimately the future of AI and

[07:30:39] education will not be defined by what

[07:30:42] machines can do but by but by what we

[07:30:45] choose to amplify in people. We will now

[07:30:49] hear from one of our winners, Hiroshi

[07:30:51] Mendoza.

[07:31:03] Uh good afternoon everyone. My name my

[07:31:07] name is Hiroshi Mendoza. I know you had

[07:31:08] a long day, so I promise I'll make it

[07:31:10] quick, but I have a very cool demo to

[07:31:13] show you at the end. So, we are Reyick,

[07:31:16] where we allow anyone to draw meaning

[07:31:19] from any text.

[07:31:22] So, this story is actually a very

[07:31:24] personal one. So, this is me when I was

[07:31:27] 10 years old and I moved from Mexico

[07:31:30] City to the US and I didn't know any

[07:31:32] English.

[07:31:34] And I remember that I was put into a

[07:31:36] classroom and the teacher asked me to

[07:31:39] read out loud in English.

[07:31:41] So I read the sentence as a Spanish

[07:31:43] speaker would. And when I finished

[07:31:46] reading the sentence, the whole

[07:31:48] classroom laughed at me and I didn't

[07:31:50] want to read out loud again.

[07:31:53] Thankfully, I found this book series

[07:31:56] called Animorphs.

[07:31:58] And it was through this curiosity and

[07:32:00] this Spanish English dictionary that I

[07:32:03] learned to decode text and understand

[07:32:06] English.

[07:32:08] Another big inspiration for this project

[07:32:11] was my brother Si.

[07:32:14] He has struggled with reading

[07:32:15] comprehension all his life and he was

[07:32:17] born deaf

[07:32:19] and obviously the phonics approach fails

[07:32:22] deaf children.

[07:32:25] So there's about a million deaf adults

[07:32:28] in the US.

[07:32:30] But if we zoom out, adult literacy is a

[07:32:33] big problem.

[07:32:35] There's about 120 million people that

[07:32:37] lack the literacy skills required to

[07:32:40] excel in the modern workplace.

[07:32:46] Income is directly tied to your literacy

[07:32:48] levels.

[07:32:50] And I saw this disparity firsthand.

[07:32:54] My younger brother does hearing and my

[07:32:56] little brother Sagei does deaf.

[07:32:59] Both enrolled in software boot camps.

[07:33:03] My hearing brother excelled and was able

[07:33:06] to transition to a higher income career

[07:33:08] while my other old other old other old

[07:33:10] other old other old other older brother

[07:33:10] Si couldn't.

[07:33:15] As you can see

[07:33:17] between the deaf and hearing population,

[07:33:19] there's about a 20%

[07:33:21] difference in temperance employment.

[07:33:25] There is a lot of communication barriers

[07:33:27] between the deaf and hearing world.

[07:33:30] But ultimately I do believe literacy

[07:33:32] played a big factor in this gap.

[07:33:37] So what is recycic?

[07:33:40] Recycle is a combined framework for

[07:33:42] reading inspired by for professor Kun a

[07:33:46] a dev graduate of Stanford who advocated

[07:33:49] having a to build a cognitive foundation

[07:33:52] in your strongest language what deaf

[07:33:54] children learn to read by sign language.

[07:33:57] We combine this with uh professor

[07:34:00] Filmore's juicy synthesis framework that

[07:34:03] breaks down complex text and allows you

[07:34:06] to derive meaning from it

[07:34:09] eventually taking a frustrating task and

[07:34:12] transforming it to self-propelled

[07:34:14] learning.

[07:34:18] So read psychic allows anyone to derive

[07:34:21] meaning from any text. So here's a quick

[07:34:24] demo. It's a Chrome extension, so you're

[07:34:27] able to highlight the text and my

[07:34:30] brother uses this daily and all allows

[07:34:32] you to rephrase the sentence to your

[07:34:35] reading level. In learn mode is it

[07:34:39] breaks down the sentences line by line

[07:34:40] and explains it in your native language.

[07:34:43] My mom, who's Japanese, uses this to

[07:34:45] understand English literature.

[07:34:48] And this is different than Google

[07:34:49] Translate because we're not translating

[07:34:51] word for word. would actually explain

[07:34:53] the text as you had a teacher next to

[07:34:55] you. And this is possible today because

[07:34:58] of AI.

[07:35:00] And in this process, I realized that in

[07:35:02] order to help my brother in the deaf

[07:35:04] community, I needed to have an English

[07:35:07] to sign language translator.

[07:35:09] And when I told people about this, they

[07:35:12] thought I was crazy. How are you going

[07:35:13] to take a written language converted to

[07:35:15] a visual language?

[07:35:18] But nevertheless, I spent $10,000 and I

[07:35:21] worked with a Christine, an ASL

[07:35:23] interpreter that believed in this

[07:35:24] concept.

[07:35:27] And as you know, sign language is very

[07:35:29] expressive.

[07:35:31] So any facial expression can change the

[07:35:33] meaning of the sentence. So we decoded

[07:35:36] over 60 facial expressions

[07:35:39] and we invented this written schema that

[07:35:42] had the attributes to showcase a rich

[07:35:44] sign language uh meaning the spatial

[07:35:47] quadrant special expressions

[07:35:50] and we had a data set of thousands of

[07:35:52] English sentences to sign language

[07:35:53] translations

[07:35:56] and we fed this into a model.

[07:36:00] So today if you go to Gemini and Chad

[07:36:01] JBT and ask it to translate to sign

[07:36:03] language, it's going to output something

[07:36:04] like this and this is wrong because it's

[07:36:07] based on ASL gloss which is a antiquated

[07:36:10] system that doesn't express the richness

[07:36:13] of the the language. So our model or

[07:36:16] fine-tuned model is able to output what

[07:36:17] you see below and then we have a system

[07:36:20] that can map these tokens into

[07:36:22] animations.

[07:36:24] So this is the awesome demo that I want

[07:36:26] to show you. So now you're able to

[07:36:28] highlight any English text and then with

[07:36:30] our model we tokenize it and then we're

[07:36:33] able to animate it and you can see the

[07:36:35] different facial expressions. We have

[07:36:37] the ability to have the spatial

[07:36:39] quadrants. We also have the abilities to

[07:36:42] show finger spelling essentially being

[07:36:44] able to translate any English text into

[07:36:46] sign language.

[07:36:51] So,

[07:36:59] so this is the theme. I'm kind of the

[07:37:01] lead tech league product designer. I had

[07:37:03] a undergraduate degree in engineering

[07:37:05] from MIT. I graduated from also the

[07:37:08] Stanford D school design program and I

[07:37:10] will have 15 years of technology

[07:37:12] experience. and Christine as my main ASL

[07:37:16] interpreter and her husband Ratish has

[07:37:19] been proofreading the animations and

[07:37:21] he's both combined have over 20 years in

[07:37:24] teaching uh sign language and last year

[07:37:26] our advisor professor Guadalup Vades

[07:37:29] she's retired now but she used to

[07:37:31] provide professor school of education

[07:37:33] and she provided the the theorical

[07:37:35] foundation for our ideas

[07:37:38] so ultimately we're recycic and we

[07:37:40] believe that anyone can draw meaning

[07:37:43] from any text. So if you're interested

[07:37:45] in making this vision possible, I'd love

[07:37:47] to talk to you after this talk. Thank

[07:37:49] you very much.

[07:38:02] Thank you very much, Hiroshi. I would

[07:38:04] now like to invite Tori Bates and

[07:38:06] Isabelle How back up to the stage. Round

[07:38:09] of applause, please.

[07:38:22] And we have the

[07:38:24] Hi everybody. Um I'm um Tori Bates. I

[07:38:27] work on learning and product innovation

[07:38:29] at google.org. Um we are really excited

[07:38:32] to be able to support this create AI

[07:38:34] challenge. I was in the room yesterday

[07:38:36] when they all um did their pitches in

[07:38:38] front of the judges and it truly was

[07:38:40] amazing to see so many innovative ideas

[07:38:43] that are solving real world problems and

[07:38:45] transforming the future of learning. So

[07:38:47] truly congratulations to all the

[07:38:49] winners. They are super deserving.

[07:38:58] Okay, without further ado, the official

[07:39:01] plaques are going to be awarded now. So,

[07:39:03] first off, we have Hiroshi Mendoza from

[07:39:06] Reed Psychic,

[07:39:12] Keith Cole and his team from AI Studio

[07:39:15] Teams.

[07:39:28] Finalist, you can get closer to the

[07:39:30] stage.

[07:39:31] Congratulations.

[07:39:34] Congratulations.

[07:39:36] Up next, we'll have uh Bella Aai. Come

[07:39:40] on up.

[07:39:53] Up next, we have Nquille from the

[07:39:55] Freedom App.

[07:40:02] Oh, sure. Sure. Yes, please.

[07:40:10] Up next, Julie from Flourish.

[07:40:27] Up next, Omo and his team from Math

[07:40:30] Everywhere.

[07:40:36] Congratulations.

[07:40:46] Dennis Walls.

[07:40:49] Oh, there you are. Okay, come on up.

[07:40:50] Come on up.

[07:40:52] This team general of AI and fans

[07:40:54] behavior learning lab for children with

[07:40:56] autism.

[07:40:59] Okay.

[07:41:01] Congratulations.

[07:41:05] Up next, we have Aisha and Akil from

[07:41:09] Math Adoption Playbook.

[07:41:17] Congratulations.

[07:41:27] Up next, we have Adam Seagull with You

[07:41:30] Good. Congratulations.

[07:41:43] Up next, we have John Mitchell,

[07:41:46] co-designing how we teach process-based

[07:41:48] students analytical writing

[07:41:51] in collaboration with Sacred Heart.

[07:41:54] Congratulations.

[07:41:58] And last but not least, Marissa McGee.

[07:42:01] Small books, big lessons.

[07:42:09] We do have one more team from Scratch

[07:42:11] 4.0 who are not able to be here today,

[07:42:13] but we have their plaque and we will

[07:42:15] send it over to them.

[07:42:18] I would now like to invite our judges

[07:42:21] onto the stage, please, for a group

[07:42:23] photo.

[07:42:24] And maybe reviewers

[07:42:25] and reviewers as well, please. Yes, if

[07:42:27] you're in the room, we couldn't have

[07:42:29] done this without you. Come on up. Josh,

[07:42:31] I see you. Keith, I see you. Kristen,

[07:42:33] Kathy, are you all in the room? Glenn,

[07:42:37] yes. Yes. Come on up. If I didn't

[07:42:38] mention you and you were a reviewer,

[07:42:40] please come on up to the stage. Yay.

[07:42:45] Come on up. Come on up. coming up.

[07:42:49] We might have to do two rows. So, let's

[07:42:51] get creative on how we do a third row in

[07:42:53] the front here.

[07:42:57] I don't want to eclipse anybody.

[07:43:00] I don't want to block somebody.

[07:43:18] Thank you.

[07:43:29] To introduce the last one.

[07:43:31] Yes. Yeah, that's totally fine. And

[07:43:32] we'll bring the chairs up after you.

[07:43:46] So, uh we are going to bring back uh the

[07:43:50] chairs. Uh we need the full uh the full

[07:43:53] stage for this big group of um great AI

[07:43:56] winners. Um uh but let me also introduce

[07:44:01] uh briefly our closing session um which

[07:44:04] is a really exciting conversation with

[07:44:06] some um luminaries across research um uh

[07:44:11] global technology and um uh some

[07:44:15] thinkers, educational thought leaders um

[07:44:18] and then the the the session is being

[07:44:20] moderated by um John Hennessy um uh who

[07:44:25] in addition to be um the chairman man of

[07:44:27] Alphabet was also um our president at

[07:44:31] Stanford University. So a huge honor for

[07:44:34] us to have John with us tonight and this

[07:44:37] very esteemed panel. So thank you.

[07:44:56] So, we're we're planning a debate and

[07:45:00] discussion about the issues here at

[07:45:02] Susanna's request. She said she wanted

[07:45:05] to hear a little disagreement and a

[07:45:07] little variety in opinions. So, let's

[07:45:10] start out. We've been hearing about AI

[07:45:13] and this notion that this time is

[07:45:15] different. How many people remember the

[07:45:18] MOO revolution that was going to

[07:45:19] completely change the way K12 education

[07:45:22] and it didn't make it? So maybe start

[07:45:25] with a question for for each of you to

[07:45:27] answer briefly. Why is this time really

[07:45:29] different? What fundamentally about the

[07:45:31] technology could be transformative?

[07:45:35] Susanna, you want to start? You're on

[07:45:36] the end.

[07:45:37] Oh, okay. Um well I think the the

[07:45:40] ability to generate is one thing that

[07:45:42] that we didn't have that before and that

[07:45:45] really um you know changes how people

[07:45:48] can interact with it. So it can replace

[07:45:51] a lot of work. And when I think about

[07:45:53] the b the potential benefits of it, I

[07:45:55] think that like we want to get certain

[07:45:57] kinds of experiences to students because

[07:45:59] we know those experiences are really

[07:46:01] helpful for how they learn, but there

[07:46:03] are these barriers to getting there

[07:46:04] because they often require lots of work,

[07:46:07] lots of communication, lots of things

[07:46:09] like that. this the new tools have

[07:46:11] ability I think to reduce these barriers

[07:46:14] a lot in the way that kind of the

[07:46:16] traditional approaches um had had some

[07:46:19] kinds of benefits clearly but couldn't

[07:46:22] really transform uh this kind of

[07:46:24] communication and organization in the

[07:46:26] way that the current tools could if we

[07:46:28] use them right

[07:46:31] you want to add anything Rebecca

[07:46:32] sure I think one of the reasons this is

[07:46:34] really different this time is schools

[07:46:37] did not invite this techn technology

[07:46:40] into their classrooms like MOOs or

[07:46:43] other, you know, higher ed institutions.

[07:46:45] It showed up. It's software. Kids are

[07:46:49] accessing it through social media,

[07:46:51] through gaming platforms, you know,

[07:46:53] through edtech. Yes. Um, and we really

[07:46:56] can't uh forget that kids learn

[07:47:01] everywhere all the time and so they are

[07:47:04] learning with AI outside of school. And

[07:47:07] I do think this time is really different

[07:47:09] in terms of us having to build really

[07:47:12] strong

[07:47:13] family school partnerships if we want to

[07:47:16] try to wrap our hands around how to

[07:47:18] support kids to learn in an age of AI

[07:47:21] even more than say with a MOO. I I see

[07:47:23] it more analogous to social media

[07:47:25] frankly where we in the education

[07:47:28] community were asleep at the wheel. We

[07:47:31] didn't really talk about it when it

[07:47:32] rolled out. We sort of ignored it and

[07:47:34] it's come back to bite us. So when you

[07:47:36] when you draw a parallel to social media

[07:47:38] immediately comes into my brain where

[07:47:40] all the mistakes we made in social media

[07:47:43] about not limiting use or putting in

[07:47:45] safeguards and things like that. So you

[07:47:48] have to worry about this in the case of

[07:47:49] AI as well.

[07:47:50] Absolutely.

[07:47:51] Okay.

[07:47:52] Yeah.

[07:47:52] Nerv how about you? What do you what do

[07:47:54] you think?

[07:47:54] Um I guess I would start by saying we

[07:47:56] don't know yet. So this could be MOO 2.0

[07:47:59] and we'll find out over the coming

[07:48:01] years. I doubt it but it's possible. Um,

[07:48:04] I think just zooming out, this might be

[07:48:05] the most powerful technology that

[07:48:07] humanity's ever created.

[07:48:09] And so we should at least have some

[07:48:11] assumption and curiosity that that would

[07:48:13] have a big impact on education both on

[07:48:15] the opportunities and the risks.

[07:48:18] I think that's just kind of my starting

[07:48:19] point. This is a very big deal for

[07:48:21] humanity and it'll probably be a big

[07:48:23] deal for education.

[07:48:24] Yeah, I I agree with you. In fact, 10

[07:48:26] years ago, we had a debate at the Google

[07:48:28] board. What was going to be more

[07:48:30] important, AI or the internet?

[07:48:33] and Larry and Sergey said AI and

[07:48:36] everybody else on the board said the

[07:48:38] internet. Five years later, we had the

[07:48:40] debate again and everybody had flipped

[07:48:42] into the AI thing. So, I think you're

[07:48:44] right. It's a transformative technology.

[07:48:47] Shantu.

[07:48:47] Yeah. No, I I completely agree with that

[07:48:49] because I think now we're not just

[07:48:51] transforming education, we're

[07:48:52] transforming everything, how we work,

[07:48:55] the economy, every part of that. And I

[07:48:57] guess you know you said you wanted

[07:48:58] debate. Uh I actually disagree with the

[07:49:00] premise of the question that education

[07:49:02] technology hasn't had a transformative

[07:49:04] impact over the last 10 years. And and I

[07:49:06] and I say that in the context of what

[07:49:08] we've seen in terms of what it's done

[07:49:10] for access what it's done for

[07:49:12] personalization. You know before I was

[07:49:13] at at at Google I helped Sar Khan

[07:49:16] Academy. And one of the things that

[07:49:18] we've always believed like that

[07:49:19] personalized learning could be and that

[07:49:21] was a dream of the muks right. You know,

[07:49:22] I met I met somebody a couple years ago

[07:49:24] who was a girl in Afghanistan who had to

[07:49:27] drop out of school at the age of 11

[07:49:29] because of the Taliban and she found

[07:49:32] online learning. She found YouTube. She

[07:49:33] found Khan Academy. She found online

[07:49:35] resources. She studied, she got did well

[07:49:38] in the SAT, came to ASU, now she's

[07:49:39] studying quantum computing at as doing

[07:49:42] her PhD in quantum computing. So to me

[07:49:44] that is a transformative impact.

[07:49:46] I I agree with you. I think the muks in

[07:49:48] the developing world have had I've been

[07:49:50] to India and seen young children who

[07:49:53] didn't have a good school environment

[07:49:55] were not going to learn learning

[07:49:57] learning on MOO. So I I think

[07:49:59] particularly

[07:50:00] and just to add to I think the big

[07:50:01] difference now is I think before that's

[07:50:03] the unusual story that's not the average

[07:50:05] goal in Afghanistan but I think with AI

[07:50:07] that becomes that personalization and

[07:50:09] that access becomes so much more

[07:50:10] accessible to a lot more people.

[07:50:12] Yeah, I think the personalization thing

[07:50:13] is absolutely right. I mean it's the the

[07:50:16] old days personalization was watch the

[07:50:18] video again right that until you get it

[07:50:20] right Salan actually once told me a

[07:50:23] story somebody had to watch a video on

[07:50:24] linear algebra 16 times but they got it

[07:50:27] on the 16th time so

[07:50:29] but we can do better we we can do better

[07:50:31] uh Rebecca you've written a lot about

[07:50:34] how education has to change that we need

[07:50:36] to rethink how we're educating children

[07:50:39] for this age of AI say something about

[07:50:41] what you think what what really has to

[07:50:44] change And how do we how do we make sure

[07:50:45] they're learning?

[07:50:46] Right. Well, um I've been thinking about

[07:50:49] what the purpose of education is for

[07:50:51] some time, PreAI. And the the best

[07:50:55] analogy I've most recently come up with

[07:50:57] is with my co-author Jenny Anderson. We

[07:51:00] wrote a book last year called The

[07:51:01] Disengaged Teen. And really what we were

[07:51:04] looking at is why do so many kids hate

[07:51:06] school? Like why do they hate school?

[07:51:08] Kids love learning. They're hardwired to

[07:51:10] learn. You know, 75% of them say they

[07:51:12] love school in third grade. And by 10th

[07:51:14] grade only 25% of them say they love

[07:51:16] school. And one of the main sort of

[07:51:20] conclusions we came to was that for many

[07:51:24] years we have operated under an age of

[07:51:27] achievement. This that the purpose of

[07:51:29] school is to rank sort uh and help kids

[07:51:35] figure out who gets to go to Stanford or

[07:51:37] who gets to go to other higher ed

[07:51:39] institutions. And historically that is

[07:51:41] really in every country of of the world

[07:51:43] industrialized or not industrialized

[07:51:46] that is sort of the core spine of

[07:51:47] schooling. So could we uh think of a

[07:51:51] different purpose? And we kept again

[07:51:54] this was we were doing the study before

[07:51:56] chat GBT came out. And so uh it came out

[07:51:58] right in the middle of the research and

[07:52:00] we thought well you know what we really

[07:52:02] we're seeing the fray fraying nature of

[07:52:05] the age of achievement in low engagement

[07:52:07] in employers saying I have to spoon feed

[07:52:10] my young recruits. They need to be told

[07:52:13] exactly what to do. They can't problem

[07:52:15] solve. um mental health crisis, chronic

[07:52:18] absenteeism, you name it. And we

[07:52:20] thought, well, really what we need is

[07:52:21] sort of an education system, a new era,

[07:52:23] and sort of age of agency, age of

[07:52:25] student agency. What if if education

[07:52:28] systems really were about helping kids

[07:52:31] learn how to learn, find their passion?

[07:52:33] Yes, acquire knowledge. You can't throw

[07:52:35] I'm especially worried about throwing

[07:52:36] the baby out with the bathwater with AI

[07:52:38] on the acquiring knowledge piece. Uh but

[07:52:41] apply it. And there's lots of creative

[07:52:43] pedagogies to do it. just haven't done

[07:52:45] that at scale and we do need to reorient

[07:52:48] how we assess and AI has a potential I

[07:52:52] think assessment is an area that AI has

[07:52:54] a real potential for opening up. So S

[07:52:57] Susanna you've spent your life studying

[07:52:59] measuring education measuring impact

[07:53:02] measuring how do we approach that in

[07:53:04] this age of AI because it seems like

[07:53:06] we're going to have to rethink a lot of

[07:53:08] things.

[07:53:08] Yeah. So certainly we don't want to kind

[07:53:11] of you I I don't think there's a

[07:53:13] dichotomy really between just trying

[07:53:16] things that seem right and also instead

[07:53:19] using research approaches that we've had

[07:53:21] forever where you need five years to

[07:53:23] find out any kind of information but you

[07:53:25] can really instead engage with what we

[07:53:27] know and also build learning into what

[07:53:30] we do. And I think we need both of those

[07:53:32] things because like we in some of these

[07:53:35] AI um chat bots and things like that,

[07:53:38] the tutors uh con Kamigo or something

[07:53:41] like that where you've seen this great

[07:53:42] effect in Afghanistan. We also um we

[07:53:46] those are great because we've built in

[07:53:48] an understanding of some of the learning

[07:53:50] sciences into it kind of what kinds of

[07:53:52] supports do we need and we need to do

[07:53:54] more of those kinds of things. But we

[07:53:56] also know a lot uh from from research

[07:53:59] that 5% what do they say the 5% rule the

[07:54:02] 10% rule you get that that many students

[07:54:05] is a lot when you look around the world

[07:54:07] but we don't want to leave out the 95%

[07:54:10] of other students who aren't going to

[07:54:12] engage in this and so we really need to

[07:54:14] think about the research that's out

[07:54:16] there that that talks about how you

[07:54:18] engage uh students how do you get them

[07:54:21] to want to be on these platforms and I

[07:54:23] think we have a lot of knowledge about

[07:54:24] that that's out there too. So you want

[07:54:26] to pull that kind of knowledge in. For

[07:54:28] example, getting an adult to work with

[07:54:30] them can can allow them to work on that

[07:54:33] platform better. Maybe that adult

[07:54:34] doesn't need to be with them always, but

[07:54:36] the adult needs to be there to celebrate

[07:54:38] the successes the students have to

[07:54:40] really motivate and engage them to do

[07:54:42] it. So the first thing that we want to

[07:54:44] do is kind of take what we know and

[07:54:46] apply it. give them the kinds of

[07:54:47] experiences they want and don't make the

[07:54:49] obvious mistakes that that uh we know

[07:54:53] that from everything we've done in

[07:54:55] edtech that we've we've been talking

[07:54:57] about. But the other thing is that we

[07:54:59] really need to learn more as it goes

[07:55:00] along. So um I I will be an advocate for

[07:55:04] partnerships where we build research

[07:55:06] into what's going on. If you're rolling

[07:55:08] out a program into 10,000 schools with a

[07:55:11] 100,000 students, do it in a way where

[07:55:13] you learn something. you can randomize

[07:55:16] different parts of it so you know what

[07:55:17] engages and what doesn't engage students

[07:55:20] what helps them get to the next stage um

[07:55:22] and what doesn't what makes them

[07:55:24] disengage we have I'm sure we'll talk

[07:55:26] about you know cognitive offloading and

[07:55:28] the the worries about that so you really

[07:55:30] want to build that in and I think you

[07:55:32] also want to build in measures as you go

[07:55:34] along you want to know um what's

[07:55:37] happening so you want to do this

[07:55:38] randomization and you want to collect

[07:55:40] kind of implementation measures where is

[07:55:43] this falling off or for something like

[07:55:45] these um AI tutors, how can they be used

[07:55:48] in a classroom in a way that actually

[07:55:50] engages students in there? So, if I were

[07:55:53] thinking, I would maybe get rid of the

[07:55:55] five-year RCTs kind of stuff, but I

[07:55:57] wouldn't get rid of really really

[07:55:59] careful measurement and a lot of AB

[07:56:02] testing as you do this.

[07:56:04] Yeah, which of course on the internet,

[07:56:06] everybody who builds an internet program

[07:56:07] knows how to do AB testing and do it and

[07:56:10] scale it up as you get more and more

[07:56:11] people in. It's the way that business is

[07:56:13] run. So I think we ought to be we should

[07:56:15] be able to do this.

[07:56:15] I I think that's right. But it's more

[07:56:17] than just like what the tool does.

[07:56:18] You've got to AB test how you get it

[07:56:20] into to get it to students for the whole

[07:56:24] which is a more complicated

[07:56:25] it's a little bit more complicated.

[07:56:26] Yeah.

[07:56:27] Nerv you you you're you're working on AI

[07:56:29] at anthropic. What do you what do you

[07:56:32] think about what what what does

[07:56:33] education need to double down on that AI

[07:56:35] isn't going to do? What's the role and

[07:56:38] how do they fit together? Yeah, I think

[07:56:40] I mean one of the hard things about AI

[07:56:43] deployment is I think we have strong

[07:56:46] confidence that it'll eventually have

[07:56:47] big impacts across the society but in

[07:56:50] any given year you don't know exactly

[07:56:52] what part of society what's going to hit

[07:56:53] when. So uh one thing we think about

[07:56:55] anthropic is kind of if then statements

[07:56:58] like if the world looks like this then

[07:57:01] we should start changing in this way but

[07:57:03] trying to have some humility that we

[07:57:05] don't know when the ifs will happen. Um

[07:57:08] so in education and curriculum like and

[07:57:10] how we might change I can kind of think

[07:57:11] of maybe three phases in the current

[07:57:14] phase like the world doesn't look that

[07:57:16] different than it did a couple years ago

[07:57:18] and so I think our main focus should be

[07:57:20] just using AI tools for the existing

[07:57:22] curriculum we have there's plenty of

[07:57:24] kids who aren't getting reading math

[07:57:25] done maybe AI can help um and that

[07:57:28] should just be our focus right now and I

[07:57:30] wouldn't call for any radical change I

[07:57:32] do think it's likely in the next kind of

[07:57:34] two to 10 years that the economy will

[07:57:36] start shifting jobs will start shifting.

[07:57:38] I don't know that that'll affect

[07:57:40] elementary and middle schools that much,

[07:57:42] but it should probably affect high

[07:57:43] schools and colleges um and what they're

[07:57:46] focusing on. And so, as that data comes,

[07:57:49] I think that's an area where you will

[07:57:50] want some curricular change. And then,

[07:57:54] you know, just to be very open and

[07:57:55] direct, I do think over some period of

[07:57:58] time, uh call it decades, the world's

[07:58:01] going to look radically different akin

[07:58:03] to the farmer industrial revolution. And

[07:58:06] you know that changed education a lot

[07:58:08] and that's hard to predict. It's hard to

[07:58:10] know and so I have less confidence here.

[07:58:13] Um but one way like I personally think

[07:58:16] about it is education has three fairly

[07:58:18] core functions. Uh turning a kid into a

[07:58:21] good person, turning a kid into a good

[07:58:23] citizen and then turning a kid into a

[07:58:26] good worker is probably the wrong word

[07:58:28] but uh participating in the workforce.

[07:58:31] And my hunch is those first two things

[07:58:34] are going to become more important and

[07:58:36] the third thing is going to decline in

[07:58:38] important a bit uh when understanding

[07:58:41] what it means to be a flourishing human

[07:58:43] and how to participate in a society

[07:58:45] where maybe humans are doing less labor

[07:58:47] is going to have profound impacts on

[07:58:49] education but I think that's further

[07:58:51] down the road. Yeah, it it is somewhat

[07:58:53] ironic that the early impact on jobs

[07:58:56] appears to be on starting programmers,

[07:58:59] which is kind of a funny thing. And one

[07:59:02] of the healthy things is the CS

[07:59:05] enrollment at Stanford and Berkeley has

[07:59:07] actually dropped for the first time in

[07:59:09] 10 years because it was too many

[07:59:11] students. Too many students. And what

[07:59:12] are they doing? They're going to other

[07:59:14] fields. They're still using computing as

[07:59:16] a tool. They're using AI as a tool, but

[07:59:18] they're going to other fields. They're

[07:59:19] shifting around. Shantanu uh this is

[07:59:22] obviously as you roll this out we have

[07:59:25] to worry about that ensure that there's

[07:59:27] positive human impact right that we've

[07:59:29] got appropriate guard rails we've got

[07:59:31] the right kinds of how do you think

[07:59:32] about that part of the problem at the

[07:59:34] same time when people are saying we want

[07:59:36] this technology how do you balance that

[07:59:39] yeah and I think one of the things that

[07:59:40] you know I focus on is how do we bring

[07:59:42] this technology to education schools and

[07:59:45] in the schools and institutions K12

[07:59:46] schools and higher institutions and you

[07:59:48] know as a part of that you know a big

[07:59:51] what we already see is like students are

[07:59:52] running away with this, right? Like the

[07:59:54] youth are the earliest adopters. They're

[07:59:56] moved they've moved on, you know, and

[07:59:58] they're they're using the technology

[08:00:00] significantly. So what's critical here

[08:00:02] is that we figure out how to bring the

[08:00:05] technology responsibly responsibly to

[08:00:07] empower kind of all the the the

[08:00:09] institutions and you know as a part of

[08:00:11] that I think I actually do think that's

[08:00:13] why educators are so critical to this

[08:00:15] conversation because they can really

[08:00:17] guide the the right way to use the

[08:00:19] technology. is obviously the safety

[08:00:20] aspects. We don't, you know, we want to

[08:00:22] we want to restrict the persona so

[08:00:23] people aren't building unhealthy

[08:00:25] relationships with the technology, you

[08:00:27] know, particularly for younger younger

[08:00:28] users. Uh there's obviously content

[08:00:30] safety, but also how do we make sure

[08:00:32] that the technology is pedagogically

[08:00:35] appropriate, right? It's one thing to do

[08:00:36] well on competition math exams and and

[08:00:39] and score a gold medal on the math

[08:00:41] olympiad. It's another thing to to be a

[08:00:43] good teacher and and and really, you

[08:00:46] know, have those pedagogical attributes

[08:00:48] to it. But one of the things that we do

[08:00:49] at Google is we actually benchmark the

[08:00:50] models against that. You know we have

[08:00:52] learn lm where we're looking at it on

[08:00:54] whether it's deepening metacognition

[08:00:55] whether it's driving out to learn is it

[08:00:57] when we ask it to not give the answer

[08:00:59] and and have the productive struggle

[08:01:00] does it do instruction following well

[08:01:02] there. So I think all of those things

[08:01:04] are complex problems and you can't do

[08:01:05] them on your own. I think you have to

[08:01:07] work with institutions you have to work

[08:01:10] with all the people and it has to be

[08:01:12] that partnership to bring it correctly

[08:01:14] and really make sure it's landing

[08:01:15] appropriately in society. And do you

[08:01:18] think we can build guard rails that

[08:01:20] brilliant young people won't jailbreak

[08:01:22] somehow and get around? I I think

[08:01:25] they're really smart when it comes to

[08:01:27] using computing. And often most young

[08:01:30] people know more than their parents

[08:01:31] about how to use the technology. I I do

[08:01:34] think you can, but I think it's not

[08:01:35] easy. I think there's the point, right,

[08:01:36] which is you really need to focus on it.

[08:01:38] It can't be something where you put a

[08:01:40] little bit of effort like, oh, I tested

[08:01:41] it and it looks pretty good on these,

[08:01:43] you know, with w with with with with

[08:01:45] some of the eval. You have to really,

[08:01:48] you know, because students are pretty

[08:01:49] good, young people, as you say, are

[08:01:50] pretty good at finding workarounds on

[08:01:52] anything that that that you put out

[08:01:54] there.

[08:01:54] Yeah. And one one thing we've seen is AI

[08:01:56] produced code sometimes has complex

[08:01:59] security flaws that we were not aware of

[08:02:01] if you just look at the code. So we

[08:02:03] there are some challenges we still have

[08:02:05] to solve. Um coming back to Rebecca and

[08:02:09] and Susanna um talk about how do we

[08:02:12] think about this in the context of if we

[08:02:15] don't do something I mean there's a

[08:02:16] there's a there's AI as making a

[08:02:19] positive change but there's also well if

[08:02:21] we don't engage with it are we going to

[08:02:23] be not preparing uh kids for the for the

[08:02:26] new century here. I think I think the

[08:02:29] biggest sort of quandry that parents,

[08:02:32] educators, experts like all of us are

[08:02:35] trying to figure out is what's the

[08:02:36] balance between protect our kids and

[08:02:38] prepare our kids. Like that is a real

[08:02:41] tension lived on a daily basis in

[08:02:43] classrooms in homes. We just did a big

[08:02:46] study um at the Brookings Institution

[08:02:49] with the global task force on AI and

[08:02:51] education and we were and Isabelle and

[08:02:53] Glenn and a bunch of other people were

[08:02:55] were on our uh steering committee. Um,

[08:02:57] and we were really asking the question,

[08:03:01] are we on the right track when it comes

[08:03:03] to K12 students

[08:03:05] learning and development and generative

[08:03:07] AI? And ultimately what we found was no,

[08:03:10] we are not on the right track. The risks

[08:03:12] are overshadowing the benefits as it is

[08:03:15] rolled out today. And part of that is

[08:03:18] because we I think are at risk of like

[08:03:22] losing the forest for the trees when we

[08:03:24] focus on very single point great

[08:03:26] solutions deployed in one place in one

[08:03:29] classroom. And you don't you miss the

[08:03:31] big picture that a lot of kids um are

[08:03:34] spending lots of time with frontier

[08:03:37] model chat bots that are not designed

[08:03:40] for kids are not designed for learning.

[08:03:42] And you can put uh you know learn LM

[08:03:45] next to a not give me the answer LM.

[08:03:48] What kid is going to pick the the learn

[08:03:50] one and not the give me the answer one?

[08:03:52] It's and it's not because they're bad or

[08:03:54] they're lazy. It's just that's how the

[08:03:56] human brain is wired. So the risks

[08:03:58] really do uh come about around

[08:04:04] unscaffolded open-ended

[08:04:06] sort of a genai use largely with chat

[08:04:09] bots um often outside of school. And

[08:04:12] there are benefits, real benefits, but

[08:04:14] it's with narrow AI use really, you

[08:04:17] know, vetted tools, vetted content,

[08:04:20] embedded in good pedagogy,

[08:04:22] uh, helping to sort of do things that

[08:04:24] you couldn't do before. I have a real

[08:04:27] and I've become increasingly this way,

[08:04:30] minimalist mindset when it comes to

[08:04:32] deploying technology in schools. If you

[08:04:34] can do it with a paper and pencil or a

[08:04:36] hug or in person or give a kid a hug or

[08:04:38] just have a chat, just do that. Like use

[08:04:42] the technology for things that are

[08:04:43] really transformational. I was yesterday

[08:04:46] I can't remember where I was, but I

[08:04:47] think it was yesterday. I was um touring

[08:04:50] um u George Mason University has

[08:04:54] starting to develop AI embedded um VR

[08:04:57] simulations to train nurses. Fantastic

[08:05:00] use. They learn the content, they brief,

[08:05:02] they go in, they're they practice, they

[08:05:04] practice. The AI is making this VR

[08:05:05] simulation incredibly real life. That is

[08:05:08] a great use of AI to really do something

[08:05:11] new. So that's where I think we should

[08:05:13] be focusing our attention.

[08:05:16] Yeah. So I don't know if it really

[08:05:18] matters whether we, you know, get the

[08:05:20] least technological thing we possibly

[08:05:22] can in there. I I more worry about kind

[08:05:24] of what we're not providing right now

[08:05:27] that we should be and that maybe we can

[08:05:30] even before what you need to be to

[08:05:33] flourish in life changes. We're not

[08:05:35] focused schools on what currently kids

[08:05:38] need to flourish in life. I think there

[08:05:40] are all sorts of different capacities

[08:05:42] that that are beneficial to you. Even

[08:05:44] things like getting pleasure out of what

[08:05:46] you do. That's like a simple one. But I

[08:05:48] do think there are ones like being

[08:05:50] creative, coming up with new ideas,

[08:05:53] critical thinking, of course we think is

[08:05:54] important, being able to persuade

[08:05:56] something or some somebody of something,

[08:05:58] which we're trying to do now. But we

[08:06:00] have all these kinds of things and and

[08:06:02] we haven't really measured them in in

[08:06:04] schools very well. And I think as a

[08:06:06] result, we haven't worked towards them.

[08:06:08] And part of the reason we don't measure

[08:06:09] them is it's really hard to measure them

[08:06:11] with an assessment. If you're thinking

[08:06:13] about something like coming up with an

[08:06:14] idea, you have to measure that over a

[08:06:17] long period of time because you don't

[08:06:18] come up with an idea, you know, in the

[08:06:20] 10 minutes you have to take this

[08:06:22] assessment. So, if we could have kind of

[08:06:24] more ambient information that's out

[08:06:27] there that isn't necessarily

[08:06:29] summitative, but provides us an ability

[08:06:32] to uh help students learn towards

[08:06:35] outcomes that really help them flourish

[08:06:36] in life, I think we would be kind of so

[08:06:38] much better. And I think there are a

[08:06:40] number of things like that. Like for

[08:06:42] high school students, I think there's

[08:06:44] this great potential of career and

[08:06:46] technical education and apprenticeships

[08:06:48] or at least project-based learning to be

[08:06:50] much more engaging than it is now. But

[08:06:53] part of the problem is that it's really

[08:06:55] complicated to do that well. And so

[08:06:57] there are a couple of places that do

[08:06:58] that well, but most it's worse than, you

[08:07:00] know, just being in a classroom and

[08:07:02] hearing stuff. So um as a result, we

[08:07:05] don't do it very much. But if we could

[08:07:07] get over those barriers by kind of I

[08:07:10] again to the measurement like if we

[08:07:12] could measure when a student in

[08:07:14] apprenticeship was doing something that

[08:07:16] demonstrated they were developing skills

[08:07:18] or they were slacking off and not doing

[08:07:20] anything and we could go in and say well

[08:07:22] what about if you took made this effort

[08:07:25] I really think we could even in the

[08:07:27] world we're in now make huge progress

[08:07:30] and I don't really I don't know I don't

[08:07:32] really care if a student is using an L

[08:07:36] uh um you know an LLM or they're using

[08:07:38] paper or whatever it is if we are

[08:07:40] thinking about whether they're getting

[08:07:42] these kind of experiences and capacities

[08:07:44] that they need to flourish.

[08:07:46] So that that's a really great question

[08:07:48] because that's what we really want to

[08:07:50] make sure that they have critical

[08:07:51] thinking skills. How nerv how do we deal

[08:07:54] with that? How do we prevent them from

[08:07:56] just using the tool to get the answer

[08:07:58] without doing the hard thinking without

[08:08:00] doing the critical analysis?

[08:08:02] That's definitely a risk. I think a lot

[08:08:04] of kids are cheating on a lot of things

[08:08:06] right now

[08:08:07] or being strategic

[08:08:09] or being strategic like who who wouldn't

[08:08:12] use it like Yeah.

[08:08:14] Uh yeah. Uh I I'm definitely worried

[08:08:16] about uh the amount of cheating that's

[08:08:18] going on. Um you know, my hunch is like

[08:08:21] the thing that prevents cheating is good

[08:08:23] assessments.

[08:08:24] Um and so I'll just throw out a

[08:08:27] curiosity. uh you know like Oxford and

[08:08:31] Cambridge they did oral exams like you

[08:08:34] go in you talk to your professor and

[08:08:35] then they tell you if you like got the

[08:08:37] material or not um I think AI will allow

[08:08:40] you to scale oral exams so just like as

[08:08:43] a little side project at anthropic uh

[08:08:46] now that anybody can build an app I'm an

[08:08:48] English major um and very non-technical

[08:08:51] and I created my own oral examiner app

[08:08:54] where you could just upload a PDF and

[08:08:56] then Claude would just start asking you

[08:08:58] questions And I gave Claude a little

[08:08:59] rubric and then it would grade you and I

[08:09:02] did it and like it gave me a 3.6 out of

[08:09:04] five and that felt about right.

[08:09:08] And so I don't know my hunch is if you

[08:09:10] get the assessment right that'll be the

[08:09:12] way to get cheating down. And I think

[08:09:14] there's a lot of cool ways to do

[08:09:15] assessments. Like I'll throw out one

[08:09:17] more anecdote at the end of every week I

[08:09:19] ask Claude uh h how good of a thinker

[08:09:22] does it think I was over the past week?

[08:09:25] and it goes over all our conversations

[08:09:27] and it tells me where I've done well and

[08:09:29] not well.

[08:09:29] Wow.

[08:09:30] Um and it usually stings sometimes but

[08:09:34] but often insightful. Uh so yeah.

[08:09:37] Do you change because of that?

[08:09:39] Oh yeah. Yeah. I care what the following

[08:09:42] week. Do you like think about how

[08:09:44] you know it's gradual but uh and I guess

[08:09:46] too early to tell. There's no RCT on

[08:09:48] Europe's uh interactions.

[08:09:51] But there could be.

[08:09:52] There could be. There could be. We had

[08:09:54] an AB test and the mask.

[08:09:56] The problem is you you know you can't do

[08:09:58] this by natural experiment because it's

[08:10:00] all new technology. So you can't look

[08:10:02] back and say

[08:10:04] you're going to have to do it from

[08:10:05] ground zero.

[08:10:06] Yeah. But you do do need uh you do know

[08:10:09] some things. And for example, I wouldn't

[08:10:11] go initially to assessments. I would go

[08:10:13] to humans. I would say that's how we

[08:10:15] make it good is we build it into

[08:10:16] interactions and that's how we know if

[08:10:19] it's you know that's how we make it so

[08:10:20] kids don't cheat.

[08:10:21] You know I was thinking about something

[08:10:24] I said to my son and to my granddaughter

[08:10:26] reading is a superpower. Do I have to

[08:10:29] now say AI is this knowing how to use AI

[08:10:32] is the new superpower?

[08:10:34] Additional

[08:10:35] additional

[08:10:36] I mean it can still read

[08:10:39] but it's true like domain expertise

[08:10:41] higher level cognitive critical thinking

[08:10:44] you get you do a lot more with AI and

[08:10:47] you know you ask questions and formulate

[08:10:50] things poorly you get not so much. So I

[08:10:53] I'm actually quite worried about,

[08:10:55] you know, a new layer of cognitive

[08:10:57] stratification.

[08:10:59] I want them to understand the tools. I

[08:11:01] don't want to turn them all into prompt

[08:11:02] engineers, though. Right. That's that's

[08:11:04] the that's the thing we don't want to

[08:11:06] do.

[08:11:06] Shantrew, talk about the role h how do

[08:11:09] you think about this from a role of of

[08:11:12] industry of a company as we roll out

[08:11:14] this technology? And I'm sure Nerv will

[08:11:16] have some thoughts about this too. I

[08:11:17] mean, how do you think about their role

[08:11:20] um in doing this right and ensuring that

[08:11:22] we make progress? I mean, Notebook LM is

[08:11:25] great. I mean, I love it. It's it's

[08:11:27] amazing to just do things and you just

[08:11:29] take a 30page very dense scientific

[08:11:31] paper and you get a summary of it that

[08:11:34] you can read actually and understand.

[08:11:36] So,

[08:11:36] yeah. And I think you know and I guess

[08:11:38] going back to the thing of like are

[08:11:40] students cheating or not? Is that

[08:11:41] efficiency? like if you didn't read that

[08:11:43] and you you turn you know people used to

[08:11:44] say YouTube is that efficiency or maybe

[08:11:46] I should have read the full the full the

[08:11:48] full chapter and and I think what you

[08:11:50] have to recognize is the technology is

[08:11:52] changing this uh there's an important

[08:11:54] part of learning that is the productive

[08:11:56] struggle everybody's going to try to

[08:11:58] take the shortest path the same way like

[08:11:59] I'm not going to go down the road that's

[08:12:02] has all the traffic that's going to give

[08:12:03] me a headache I'm going to take the the

[08:12:05] quickest path and and people are doing

[08:12:06] that that's human behavior and we can't

[08:12:08] fault students for following human

[08:12:10] behavior but we have to do is make sure

[08:12:12] the pads they're taking are actually

[08:12:13] going to give them the skills that they

[08:12:15] need. And I think that's where it really

[08:12:17] does go into how do we ensure that the

[08:12:20] experiences that are happening in the

[08:12:22] classroom are around that. And I think,

[08:12:24] you know, from an industry standpoint,

[08:12:25] what I'm interested in, because this is

[08:12:27] technology, it's a general purpose, you

[08:12:29] know, technology and and it can be used

[08:12:31] for great good and it can also be used

[08:12:33] incorrect. It's like fire. You can use

[08:12:35] it to cook. You can use it for war. And

[08:12:37] with with with something like this, you

[08:12:39] really do have to identify the benefits,

[08:12:42] right? Like what are the ways we empower

[08:12:43] the educator? We already see that

[08:12:45] they're running away with it also.

[08:12:46] They're, you know, we have we had a

[08:12:48] pilot in Northern Ireland where they're

[08:12:49] saving 10 hours a week of time. So,

[08:12:52] they're embracing it. But how do we make

[08:12:53] sure that we're giving them the tools

[08:12:55] because this change is happening. And if

[08:12:57] you just leave the system as it is,

[08:12:58] you're going to have all these these

[08:12:59] negative effects. We have to empower the

[08:13:02] educators to adapt in in in this for

[08:13:04] this new world.

[08:13:05] Yeah. I think uh like one message if I

[08:13:08] can give to all the builders and amazing

[08:13:11] people in this room is the technologies

[08:13:14] I think about there but we don't yet

[08:13:16] have the product right. Um, and so what

[08:13:19] would be amazing I think and

[08:13:20] transformative from AI is if in a couple

[08:13:23] years we had a AI tutor that worked with

[08:13:26] most kids most of the time, most

[08:13:28] subjects um that we had it well

[08:13:31] researched and that it didn't degrade on

[08:13:34] mental health or disempowerment or all

[08:13:36] these issues we've talked on.

[08:13:38] And I think if we can crack that, which

[08:13:40] seems very real like uh possible that we

[08:13:42] should, then it'll just be I mean be

[08:13:44] amazing. And I think like it'll have a

[08:13:47] chance to be one of the more equitable

[08:13:50] interventions we've ever had. So I'm

[08:13:51] like generally bullish that AI will

[08:13:53] reduce educational inequality. Like the

[08:13:57] things that work the most in education

[08:13:59] are a great teacher and or a great tutor

[08:14:01] and those are just like notoriously hard

[08:14:03] to scale um here in America and

[08:14:06] particularly global. And so if we can

[08:14:07] get an AI tutor that's like approaching

[08:14:10] what a human does, I think that will be

[08:14:12] uh amazing for reducing educational

[08:14:15] inequality, but we just don't have that

[08:14:16] product yet. Um so I do think that is

[08:14:18] like the one thing that we don't have

[08:14:21] that we need. Uh but if we can get

[08:14:24] there, I think great

[08:14:26] I think we can get I mean we do have it

[08:14:27] for limited context, right? We have it

[08:14:29] for introductory programming through

[08:14:30] code in place, right? So tens of

[08:14:33] thousands of people around the world are

[08:14:34] learning to code using these AI tutors.

[08:14:37] We've got to get it broad-based though

[08:14:38] so they can handle a wide variety of

[08:14:39] things.

[08:14:40] I worry I worry much less about getting

[08:14:42] an incredible product like that and much

[08:14:45] more about are our how are our systems

[08:14:47] going to absorb them

[08:14:49] our learning systems in and out of

[08:14:50] school because you know that is actually

[08:14:54] where I think we often go off the rails.

[08:14:57] So Rebecca this is a perfect segue

[08:14:59] because that leads to my silver bullet

[08:15:01] question. And I'm giving you each a

[08:15:02] silver bullet to solve or knock out one

[08:15:06] barrier, one system level problem that's

[08:15:09] going to become the the bottleneck in

[08:15:12] getting this technology out there,

[08:15:14] Susanna. What are you going to go after?

[08:15:16] Oh, okay. So, I'll go after measurement.

[08:15:18] I think we have to measure other things

[08:15:20] besides what we measure now. And we have

[08:15:22] to measure them kind of continuously so

[08:15:25] that they can be formatively used. And

[08:15:27] with that, I'll just get into this one

[08:15:29] thing about the the tutor. I I spend

[08:15:31] most of my time thinking about tutoring.

[08:15:33] So you're not tutoring. So I love I love

[08:15:38] tutoring. I love tutoring like more than

[08:15:40] anything. But nonetheless, I think

[08:15:41] that's a very small part of what the

[08:15:43] educ what education is. I think you

[08:15:45] learn some things individually and

[08:15:48] they're factual, but that a lot of what

[08:15:50] you learn you learn through interacting

[08:15:52] with people and creating things and

[08:15:54] going back and forth. So I think we've

[08:15:56] got big barriers to getting to uh those

[08:15:59] kinds of educational experiences

[08:16:01] students which also could go use AI

[08:16:04] tools. You can use AI tools to better

[08:16:06] schedule when adults are meet with

[08:16:08] students and to to better and more

[08:16:10] flexibly schedule where students go. So

[08:16:12] that would be one or just our human

[08:16:14] resource system in general could be

[08:16:17] better done so that we can get these

[08:16:20] kind of exciting um you know beneficial

[08:16:24] experiences to students. So while I

[08:16:26] think measurement is kind of at the

[08:16:27] heart of it and that's the one I'd like

[08:16:28] to overcome, I do think we have to

[08:16:30] overcome the idea that education is just

[08:16:32] like feeding information at the right

[08:16:35] level to students because that is just

[08:16:38] and it's one important part of what we

[08:16:40] do but not a but not the main thing.

[08:16:42] Yeah, Rebecca.

[08:16:44] Um I second that. Okay.

[08:16:47] Uh so I'm I'm I'm cheating. I'm doing a

[08:16:50] comment I'm doing a comment on the

[08:16:52] tutoring the pro the Yeah. um that this

[08:16:55] is what Susanna said is exactly what I

[08:16:57] meant about how education systems absorb

[08:16:59] technology. I this is just a mere

[08:17:01] anecdote but I was talking to somebody

[08:17:03] who has was working at a large

[08:17:05] corporation in the HR department um with

[08:17:07] a lot of people who spend a lot of time

[08:17:10] in front of computers and said we have a

[08:17:12] big morale problem there was a lot of

[08:17:14] engineers and they'd rolled out AI and I

[08:17:17] said what's the issue and they said well

[08:17:18] the engineers are increasingly lonely

[08:17:21] because now when they have a question

[08:17:22] they just ask chatgpt or whatever

[08:17:24] they're using they don't go and ask

[08:17:26] their colleague for help and I just want

[08:17:29] us to I'll remember that because that is

[08:17:30] a bit your point. Like yes, a great

[08:17:33] tutor for each kid, but how are we going

[08:17:36] to absorb that? We don't want them just

[08:17:38] sitting there with their little AI tutor

[08:17:40] all day long, right? That's just a slice

[08:17:42] of how they would spend their their time

[08:17:44] because that's not how kids develop. Um,

[08:17:46] but the one thing I think that I would

[08:17:49] try to tackle is really helping

[08:17:52] educators shift their pedagogy and

[08:17:55] support them, give them access to the

[08:17:58] tools, let them lead, uh, and really

[08:18:02] help them get into this explorer mindset

[08:18:04] where they can, you know, decide when

[08:18:06] not to use it and when to lean into it.

[08:18:09] I think that pedagogical shift is going

[08:18:11] to be our biggest hurdle.

[08:18:14] Um maybe one this is great to go back

[08:18:16] and forth. I don't think I'm worried too

[08:18:18] much. I agree that like adoption and

[08:18:21] absorption like could go well or not

[08:18:24] well um or could takes uh be efficient

[08:18:27] or not efficient, but I I do think if

[08:18:29] it's like a truly amazing product, it

[08:18:32] will happen. Um, and so I I can't

[08:18:35] particularly

[08:18:36] that's what I'm worried about

[08:18:37] that that

[08:18:38] that if it's so amazing it will happen

[08:18:40] and we'll close down some of the sort of

[08:18:42] social interactions.

[08:18:44] Got it. Got it. Well, that goes to my

[08:18:46] that goes to my silver bullet. Okay,

[08:18:47] great.

[08:18:48] Um, so I do think like uh my silver

[08:18:51] bullet would be I wish it was easier to

[08:18:53] start new schools and particularly new

[08:18:55] public schools in the United States. Um

[08:18:57] because I think starting a new school is

[08:19:00] an easier way to figure out what the AI

[08:19:03] human curriculum school day should look

[08:19:06] like and I think we just need a lot of

[08:19:08] experimentation to figure that out. Um

[08:19:10] and I think new schools are the best way

[08:19:12] to do that.

[08:19:13] But that's a very big silver bullet

[08:19:14] because you'd have to start a lot of new

[08:19:16] schools. So

[08:19:18] uh uh maybe I should be more narrow. Uh

[08:19:20] the new schools will figure out the

[08:19:22] thing that then everybody else

[08:19:23] Oh, everybody else will. Okay. It's just

[08:19:25] a way to get to the answer quickly.

[08:19:27] Good. Well, that's that's a great idea.

[08:19:28] I mean, you have an you have a

[08:19:30] experimental, right? You have the point

[08:19:32] of the arrow doing the work to really

[08:19:34] discover what really works. That's a

[08:19:36] great idea, Shant.

[08:19:37] Yeah, I I'd actually agree with Rebecca

[08:19:38] on like the the training supporting

[08:19:40] educators is critical. And I think it it

[08:19:42] goes back I mean your first question

[08:19:43] about MOK didn't disrupt higher ed. You

[08:19:46] know, I would actually say like when you

[08:19:47] think about this, remember what's going

[08:19:48] to stay the same in education. We talk a

[08:19:50] lot about the change, but what stays the

[08:19:52] same is education is about the social

[08:19:54] experiences of the humans in the room.

[08:19:57] At its core, when I send my kids to

[08:19:59] school, it's it's not about consumption

[08:20:01] of of information. It's about the

[08:20:03] mentorship from the teachers, the social

[08:20:04] experience, higher ed. It's about the

[08:20:07] the campus experience. Like that is what

[08:20:08] education is about. That's not going to

[08:20:10] change. So, what we have to do is we

[08:20:12] have to really help support the

[08:20:14] educators so that they can refine their

[08:20:16] craft in this new world where all these

[08:20:18] things are changing. when information

[08:20:20] del when your assignments no longer make

[08:20:22] sense because you can just use a gentic

[08:20:23] AI and complete all of them in a minute

[08:20:25] like what what is the new experience and

[08:20:28] that really does get back to the cords

[08:20:30] will stay the same right I think you it

[08:20:32] is going to be the discussion it's going

[08:20:33] to be the critical thinking it's going

[08:20:34] to be all those things that make humans

[08:20:36] humans and how do we really support

[08:20:38] educators through that transformation

[08:20:40] moment so that they can embrace the tech

[08:20:42] technology and use it to their benefit

[08:20:44] so it's actually driving that critical

[08:20:46] thinking as opposed to being stuck in

[08:20:48] the old system where students are just

[08:20:49] bypassing them along the way.

[08:20:51] Right? It's augmentation, not

[08:20:53] replacement of the teacher and the

[08:20:54] teacher's role. Uh we have a few minutes

[08:20:56] for questions and I think there are some

[08:20:58] mics around if somebody has a question

[08:21:00] for any of our panelists.

[08:21:03] Audience questions are always the best.

[08:31:49] our panel for a great conversation.

[08:31:56] Thank you. Thank you very much.

[08:32:49] Is that what you wanted?

[08:33:08] Uh, what do you think? I guess we Well,

[08:33:11] our mic is on.

[08:33:12] Okay.

[08:33:12] Um I think they wanted

[08:33:15] wanted to do the real

[08:33:17] Oh, yeah.

[08:33:18] Yeah.

[08:33:18] Yeah.

[08:33:19] So, um as you know, we are going to move

[08:33:23] into cocktail. Um which we're all very

[08:33:26] excited about. So excited that people

[08:33:28] are already going there. Uh I get it. Um

[08:33:33] but we have a few things. Um first, do

[08:33:37] you want to thank the sponsors?

[08:33:39] Yeah. Yeah. Yeah. So, uh, a few thank

[08:33:41] yous before I know that we are the the

[08:33:43] last remaining thing standing in the way

[08:33:45] of you and drinks. So, we'll do this as

[08:33:47] quickly as we can, but many thanks to

[08:33:49] our sponsors, Google, GoodNotes, Owl

[08:33:52] Ventures, and Everway.

[08:34:00] And then, let's see. Thanks as well to

[08:34:02] all of our event staff, all of the

[08:34:04] support team that's been joining us

[08:34:05] today, um all of our speakers, um and um

[08:34:10] certainly all of you. So, thank you all

[08:34:12] for being here. It's really been

[08:34:14] wonderful to have you here for this

[08:34:15] fourth annual AI and education summit

[08:34:17] and we look forward to seeing you again

[08:34:19] next year.

[08:34:20] Yeah.

[08:34:21] And then we have a last little gift for

[08:34:24] all of you. Uh we have a little reel uh

[08:34:28] from today.

[08:34:34] I think one thing that AI can't teach us

[08:34:37] but human can is maybe empathy. Uh

[08:34:40] that's something that's uh not tangible,

[08:34:43] not something that AI can really

[08:34:44] translate into into meaning and that's

[08:34:47] something that human connection can only

[08:34:49] foster and in the realm of education I

[08:34:51] think that's what facilitators,

[08:34:52] teachers, schools, environments u can

[08:34:55] enable. I think that to have human-

[08:34:58] centered AI and education, we first need

[08:35:02] to have human- centered education, which

[08:35:04] I don't think we have. AI will never be

[08:35:07] able to teach us how to be present, how

[08:35:09] to look into our inner world, observe

[08:35:11] our breath, observe how things make us

[08:35:13] feel. That's just something that I don't

[08:35:15] think AI will ever be able to help us

[08:35:17] do.

[08:35:19] I think AI can help educators solve the

[08:35:21] problem of expressing their curriculum

[08:35:24] in a relevant way for students,

[08:35:26] especially culturally. I think a lot of

[08:35:28] times educators get thrust into an

[08:35:30] environment or community that they don't

[08:35:31] necessarily have a connection to. Um,

[08:35:34] and AI can help them tailor their

[08:35:35] content in a really authentic and

[08:35:37] engaging way to make sure that their

[08:35:39] students feel seen and heard in the

[08:35:41] classroom. So one problem that AI can

[08:35:43] help educators solve is making sense of

[08:35:45] all of the data that we collect around

[08:35:47] students and hopefully help us to

[08:35:49] surface problems before they become

[08:35:51] problems so that we as human beings can

[08:35:53] intervene directly with the students and

[08:35:55] help them.

[08:35:57] Educators really need the ability to

[08:36:00] determine between what is fact and what

[08:36:02] is not fact and to be able to know when

[08:36:05] to trust what they have learned in the

[08:36:08] past um versus when to trust the new

[08:36:10] information coming out. I think what

[08:36:12] would help the most is tools to help

[08:36:14] scaffold and differentiate learning for

[08:36:16] all different learners in the classroom.

[08:36:18] AI community should collaborate with

[08:36:20] educators to understand longitudinal and

[08:36:23] long-term effects uh of AI on

[08:36:26] well-being.

[08:36:27] Educators really need support with

[08:36:29] dealing with change, change management,

[08:36:30] uh understanding their spheres of

[08:36:32] control so that they can implement AI

[08:36:34] and evolve and adapt along with the

[08:36:35] technology.

[08:36:36] I think educators right now need more

[08:36:40] transparent involvement in decision-m

[08:36:43] capacities that are going to affect

[08:36:44] their students. Um, often educators that

[08:36:46] are practicing in a classroom aren't

[08:36:48] asked. And so we need to be talking to

[08:36:50] people in the classroom.

[08:37:02] Feel so sad. It's almost done. What are

[08:37:05] you guys going to do? Go home and cry?

[08:37:06] I mean, I'm not going home. I I think

[08:37:08] I'm going downstairs.

[08:37:09] Downstairs? What's downstairs? Join us.

[08:37:12] We're continuing the conversation

[08:37:13] downstairs after party downstairs. See

[08:37:16] you there.

[08:37:20] That ending was great.

[08:37:33] And my test might get arrested.