# The TRUTH About AI: It's NOT Intelligence, It's a TRICK! | Prof Jiang Xueqin

https://www.youtube.com/watch?v=5llffhy3SFs

[00:00] paragraph, uh Alan.
[00:01] My conversation with Brockman and Sutskever continued on in circle until we run out the clock after 45 minutes.
[00:09] I tried with little success to get more concrete details on what exactly they were trying to build, which by nature they explained they couldn't know.
[00:17] And why then, if they couldn't know, they were so confident they would be beneficial.
[00:21] Okay, so this is a really huge problem.
[00:24] You know, so Karen Hao is a reporter.
[00:27] She's working for the Technology Review at MIT.
[00:28] In 2019, she goes to Silicon Valley and she meets with Brockman and Sutskever, who are the two main scientists behind the OpenAI project.
[00:39] And she The first question she's going to ask, and anyone should ask, is what is AGI?
[00:44] What is AI?
[00:46] What are you trying to build?
[00:49] And they could not answer that question.
[00:51] Okay?
[00:54] It's not because they don't know, it's because the truth is too scary.
[00:57] And what we're going to do is
[01:00] figure out what the truth is.
[01:02] What is AI?
[01:02] What is artificial intelligence?
[01:04] What is AGI?
[01:06] And the answer, of course, is it's God.
[01:11] What Brockman and Sutskever could not tell Karen Hao is that we're trying to create God because that is insane, evil, and stupid.
[01:24] But, if you look at what they're doing, that's exactly what they're trying to do.
[01:26] Okay?
[01:26] All right.
[01:29] So, what I'm going to do now is really quickly, all right, explain to you what AI is.
[01:38] And to understand what AI is, let's go back go back to the year 1966.
[01:42] And in the year 1966,
[01:47] a man, um, named Joseph Weizenbaum, who was at MIT, he created something called a chatbot, which is basically chat GPT.
[01:56] And he calls it Eliza, okay?
[01:58] And he's
[02:01] trying to prove to people how easy it is to fool people into believing that AI can actually think for itself.
[02:11] Okay?
[02:15] So, he creates this computer, and you sit down with this computer, and you ask it questions, and then it talks to you.
[02:18] It's not really talking to you.
[02:20] The It's a program designed to fool you to believe that it's sentient, okay?
[02:28] So, the conversation goes like, Eliza says, "Is something troubling you?"
[02:31] And you're like, "Men are all alike.
[02:34] What is connection do you suppose?
[02:36] They're always bugging us about something or other.
[02:38] Can you think of a specific example?"
[02:40] "Well, my boyfriend made me come here."
[02:42] "Is it Is it important to you that your boyfriend made you come here?"
[02:45] "He says I'm depressed much of the time."
[02:47] "I'm sorry to hear that you're depressed."
[02:48] So, it sounds as though they're having a real conversation, but what What Bom did, and this is again 1966 when they didn't have that much technology and processing power, is all this very simple trick, okay?
[03:01] And like so, let's do a thought experiment where I design I design a program software, okay?
[03:08] And all it's saying is "Tell me more."
[03:13] "Or this is interesting."
[03:16] Okay?
[03:20] This This is it. "Tell me more. This is interesting."
[03:21] Okay?
[03:24] And so, what's going to happen is that we're going to do a thought experiment where you call into a psychology hotline.
[03:31] And the psychology hotline, you you think you're talking to a person, but it's actually a computer program that says two things: "Tell me more. This is interesting."
[03:40] Okay?
[03:40] So, you call the hotline, you say, "Help, I'm in a lot of trouble."
[03:45] "Tell me more."
[03:48] "Oh, my boyfriend uh broke up with me."
[03:49] "This is interesting."
[03:51] "Yeah, he's a jerk."
[03:54] "Yeah, we've been fighting for 5 months."
[03:55] Okay, you keep then you keep on going.
[03:57] And the question is, how many people will be fooled into believing that this is a real person?
[04:01] And the answer is, unfortunately, quite a lot of
[04:03] People.
[04:03] Okay?
[04:06] So, this is a
[04:07] This is a very interesting aspect of humans where we often hallucinate reality.
[04:17] Okay?
[04:19] It is not that things are real.
[04:19] It's that we want them to be real.
[04:20] So, think of hypnosis.
[04:24] I'm not sure if you've been ever to been to a magic show where people conduct hypnosis, right?
[04:28] Well, why does hypnosis work?
[04:31] Because the audience wants it to work.
[04:36] If you go in skeptical and say this is all complete nonsense, it probably will not work on you.
[04:39] But, you're not going to pay $100 to go to hypnosis show and think it doesn't really work cuz why would you pay $100, okay?
[04:49] All right, so it's almost like sunk cost fallacy.
[04:55] And again, this is all using just basic human psychology to trick people into believing something that is not true.
[05:03] Right?
[05:03] Does that make
[05:05] Sense, guys?
[05:07] All right.
[05:07] So, let me explain to you how Open AI works, ChatGPT works.
[05:15] All right.
[05:17] Okay.
[05:18] So, ChatGPT is what we call a large language model.
[05:26] Okay?
[05:27] So, in other words, what it's trying to do is trying to trick you, the user, into believing that it knows what it's talking about.
[05:43] Okay, and how it works is basically it takes all of the internet, okay?
[05:48] Right?
[05:48] All the from the internet, and then it translates it into um a um idea.
[05:59] Okay?
[05:59] So, you ask uh you query
[06:05] the LLM, the LLM then takes the query,
[06:08] and then picks out the information from the internet, and then presents it into a paragraph that tries to trick you into believing that it is true.
[06:19] Okay?
[06:22] Do you understand?
[06:24] All right.
[06:25] So, in other words, it's actually no different from a Google search.
[06:27] The only difference is that it's taking the Google search, figuring out what the most popular answer is, and then presenting it in a way that makes you think that it's talking to you directly.
[06:37] All right?
[06:40] The trick, and this is really important to understand, guys, is it's trying to trick you.
[06:43] All right?
[06:45] It's not trying to teach you, it's not trying to tell you the truth, it's trying to trick you into believing it.
[06:49] That's what we call a hallucination.
[07:00] Okay?
[07:01] You have You guys have to understand this idea.
[07:03] There's nothing truthful about what um
[07:05] ChatGPT says.
[07:08] All it's trying to do is trying to manipulate manipulate you with words,
[07:12] with pretty words, into believing that it knows what it's talking about.
[07:15] But it itself cannot judge what it's doing.
[07:24] Okay?
[07:25] All right.
[07:25] Any questions so far?
[07:27] Are we clear?
[07:29] Okay.
[07:29] All right.
[07:30] So, now the question is, how does it do that, okay?
[07:33] And um so, I'm going to teach you a little bit about artificial intelligence.
[07:39] And please stop me if I'm not being clear about how AI works, okay?
[07:44] All right.
[07:44] So, AI doesn't exist.
[07:48] What exists we can call supervised machine learning.
[07:56] This is the technical term, okay?
[08:05] All right, supervised machine learning,
[08:07] Okay?
[08:08] And how it works is this.
[08:12] Before how computer programs would work is we would write the program, the algorithm,
[08:17] and then we'd give it the input,
[08:22] and it would do the output, okay?
[08:28] So, the algorithm may be A + B.
[08:31] We give the input 1 1, the output would be 2, okay?
[08:33] For example.
[08:35] How supervised machine learning works is, okay, this is fine for simple problems, but there's certain hard problems that humans cannot figure out, okay?
[08:48] And one hard problem is the idea of facial recognition technology.
[08:53] Facial recognition.
[08:59] How do I separate faces?
[09:01] Okay?
[09:01] And so, the problem is this.
[09:03] I have about a million faces, 1 million faces,
[09:07] in a database.
[09:10] Okay?
[09:15] And I don't know how I can best differentiate these faces.
[09:21] Now, what I do know is that there's certain characteristics of a face that allows me to differentiate, okay?
[09:28] All right, so certain variables, weights.
[09:33] Okay, so for example, eye.
[09:36] For example, uh nose, chin, okay?
[09:39] About a million, okay?
[09:42] About a million weights.
[09:43] So, I know these things do matter, but I don't know how much they matter.
[09:48] So, I'm trying to figure out what the weighting is.
[09:52] And I could try to play play by myself like say 1%, 2%, 5%.
[09:54] But as you can imagine, this would take too long because there are too many possibilities.
[10:02] So, what I do is this.
[10:02] I let the computer figure out it by itself.
[10:04] I let
[10:07] I let the computer figure out the weighting by itself, okay?
[10:11] And the way I do that is using my technique called back propagation.
[10:18] All right.
[10:20] So, what So, I control the input, okay?
[10:25] The input,
[10:27] and then I control the output.
[10:29] Yes or no.
[10:33] Okay.
[10:33] All right.
[10:33] So, does the face match or does it not match?
[10:37] And what I'm trying to do is I'm trying to figure out a situation in which all million faces are matched perfectly.
[10:46] And I do that by training the computer to constantly back propagate until it gets the weighting perfectly.
[10:54] Okay?
[10:55] So, basically, what I'm trying to do if if if you want to stand um how this works is I'm trying to turn each face into a distinct mathematical model.
[11:10] All right.
[11:12] That is unique to it.
[11:14] It doesn't make sense.
[11:16] All right.
[11:16] So, it's pretty simple.
[11:18] It's not doing that much.
[11:19] But to make it sound really fancy, I give it really fancy names to trick people to believe that this is actually much more sufficient sufficient than it is, okay?
[11:31] So, what name should I give it?
[11:31] This waiting system, I call IT A NEURAL NETWORK, GUYS.
[11:35] IT'S A BRAIN.
[11:39] IT'S MAGIC.
[11:42] OKAY?
[11:43] And back propagation, I don't call it back propagation, I call it deep learning.
[11:49] You see?
[11:56] AND I DON'T CALL IT SUPERVISED machine learning, I call IT AI.
[11:58] AH!
[12:00] THERE YOU GO.
[12:03] MAGIC, YOU SEE?
[12:06] All I've done is taking a very simple process and giving it like a really,
[12:10] Really fancy names.
[12:14] Then listen to this, like, why do I do that?
[12:15] And some people say, "Oh, it's for marketing purposes.
[12:17] It's to get more money from investors.
[12:20] It's to trick people.
[12:22] No, no, no.
[12:25] The real reason is you're trying to, with these names, create God, okay?
[12:30] It's what we call the occult.