# Bayesian Statistics vs Epistemology, with Vaden Masrani

https://www.youtube.com/watch?v=St9EnAnBlEA

[00:02] Let me show you how to be a good Bayesian.
[00:05] Today, we are doing something a little different, a proper philosophy of science episode, and one of my favorite conversations in a while.
[00:15] My guest is Vaidyanathan Madhavan, a machine learning researcher turned consultant and the co-host of the Increments podcast with Ben Chuck.
[00:24] Andreas Munk from episode 155 put us in touch, and I am so glad he did.
[00:29] Vaidyanathan has a sharp and unusual position.
[00:31] He loves Bayesian statistics, and he's deeply critical of Bayesian epistemology.
[00:35] So, we spend the episode walking that line.
[00:40] Why is it wonderful to use Bayes' theorem when you have data and a model you can falsify, and why does it go wrong the moment you start assigning probability to one-off future events with nothing to count?
[00:54] From there, we get into the problem of induction, Karl Popper and critical rationalism, why criticism rather than falsifiability sits at the
[01:02] foundation of knowledge, and how [music] all of these shows up in stand-up comedy, stoicism, and even the downfall of Sam Bankman-Fried.
[01:11] [music] It is playful, contrarian, and it made me think a lot.
[01:14] I had more questions, [music] actually, than we had time for, so I am going on Vaidyanathan's show to keep the conversation going.
[01:20] Stay tuned [music] for that.
[01:23] This is Learning Bayesian Statistics, episode [music] 160, recorded May 20, 2026.
[01:32] Let [music] me show you how to be a good Bayesian and change your predictions after taking information.
[01:39] And if you're [music] thinking I'll be less than amazing, let's adjust those expectations.
[01:45] What's a Bayesian?
[01:47] It's someone who cares about evidence.
[01:48] Welcome to Learning Bayesian Statistics, a podcast about Bayesian inference, [music] the methods, the projects, and and people who make it possible.
[01:57] I'm your host, Alex Andorra.
[01:59] You can follow me on Twitter at Alex_Andorra,
[02:02] like the country, for any info about the show.
[02:04] learnbystats.com is the place to be.
[02:07] Show notes, becoming a corporate sponsor, unlocking Bayesian merch, supporting the show on Patreon, everything is in there.
[02:13] That's learnbystats.com.
[02:15] If you're interested in one-on-one mentorship, online courses, or statistical consulting, feel free to reach out and book a call at topmate .io/alex_andorra.
[02:22] See you around, folks, and best Bayesian wishes to you all.
[02:27] Baden M'sereni, welcome to learning Bayesian statistics.
[02:34] Nice to be here.
[02:35] Yes, I'm I'm very happy to have you on the show.
[02:41] Um you were recommended to me by Andreas Munk from episode 155.
[02:45] So, of course, Andreas' episode will be in the related episodes for that one.
[02:47] Um but that's how we met, and I'm I'm very happy that we did, because I think we have a fun episode in front of us today.
[03:03] Um before that though, as usual, what is your origin story?
[03:09] You know, what are you doing nowadays, and how did you end up doing that?
[03:13] Yeah.
[03:15] So, I know Andreas from grad school.
[03:16] We did our PhDs together at UBC in a probabilistic machine learning uh reading group.
[03:21] Uh not reading group, uh lab, rather.
[03:23] Um and how I got there is kind of an interesting story.
[03:26] So, I um my undergrad was in physics, um and so, near the end of my um undergrad, I was working for um the ATLAS experiment, which is the one that discovered the Higgs boson, gosh, in 2012, 2013.
[03:43] To be extremely clear, I had nothing to do with the discovery of the Higgs boson.
[03:47] I was just a just an undergrad making uh making plots.
[03:50] Um but my uh kind of my summer project was to train neural networks to detect a particular kind of particle decay.
[03:58] And through that process, I realized I was way more interested in the neural networks side of things than the particle decay side of things.
[04:03] So I kind
[04:06] of pivoted and then did my masters in ML
[04:08] at UBC as well, more on the applied side
[04:11] of
[04:11] side of the research.
[04:14] And then that led me to working in Tokyo
[04:16] in a Bayesian ML group there with empty
[04:19] Ozcan. I met Riken doing variational
[04:21] inference
[04:23] kinds of
[04:24] of work.
[04:25] And I just love that experience so much
[04:27] that I decided as
[04:30] a masochist to do the whole PhD
[04:32] experience. And that was a lot of fun.
[04:35] And so that kind of takes us to a couple
[04:38] years ago.
[04:39] So after that I was doing some research
[04:41] at one of the big companies and
[04:43] didn't love industry research as much.
[04:45] So I jumped from that about two years
[04:48] ago, a year and a half ago. And started
[04:50] a consulting company.
[04:52] And that's kind of what I've been doing
[04:53] ever since. And so a lot of industry
[04:54] projects, not so much Bayesian
[04:56] statistics anymore.
[04:58] But lots lots on the philosophy side of
[05:00] things.
[05:01] I should probably also mention so in the
[05:04] sometime during the PhD,
[05:07] maybe first year of COVID, I started a podcast with a friend of mine.
[05:11] And in that space we talk a lot about Bayesian things, also the philosophy of science and history and and all these various subjects.
[05:19] And And so that's been just a extremely valuable source of intellectual nourishment because you don't have any of the publisher perish constraints.
[05:27] You can just freely explore ideas at your following your interests and that's why I'm spending most of my time these days doing cuz consulting kind of the 9-5 and then podcasting on the 5-9.
[05:42] Yeah, we'll we'll talk a bit more about your podcast.
[05:44] That's the the Increments podcast with Ben Czech.
[05:47] Um so definitely folks should check it out but we'll we'll talk about about it in later in the show and of course um you will have everything in the show notes.
[05:58] People, if you want to check out uh Ferenc's podcast, I definitely encourage you to.
[06:05] And on YouTube also,
[06:08] well, you'll have that on the on his
[06:11] This episode will also be on his
[06:14] YouTube channel. Thanks to the magic of
[06:16] YouTube collaborations.
[06:18] So yeah, definitely check that out. On
[06:21] the consulting side, actually, so a firm
[06:23] member country, your company is
[06:26] is called Sophia Consulting, right?
[06:29] It's Sophia AI.
[06:31] Yeah, Sophia AI Consulting. So two
[06:33] questions.
[06:34] What do you guys do? And also
[06:37] e Is my guess right that the Sophia is
[06:42] in relationship to philosophy and coming
[06:45] from the Greek word for wisdom or is
[06:48] that it?
[06:49] and related. No, it's totally related.
[06:51] Nice catch. You're one of the few to
[06:53] catch that. Yeah.
[06:54] So yeah, Sophia means lover of wisdom.
[06:57] And I kind of liked
[06:59] introducing a nod to wisdom compared to
[07:01] say intelligence, which is a vastly
[07:04] overused word these days. Um
[07:06] And so yeah, what we do is essentially
[07:08] just working with companies on their
[07:10] real-world problems.
[07:13] And so maybe these ways is to explain one of the projects we just concluded.
[07:17] So there's a local company called Aqua I and they make an underwater sonar gun that detects drowning swimmers.
[07:25] Um and so you have a diver out at 5 m, 10 m, up to 50 m.
[07:31] And you're bouncing sonar off of them.
[07:33] And then you do a lot of data cleaning and then you train a fast CNN so that it works on device essentially.
[07:41] Released the Aqua I Pro model, which um has moved over from random forests to CNNs.
[07:47] And so that's been one of the major projects I've been working on for the last couple years almost.
[07:54] Um, that's kind of in the space of machine learning in AI.
[07:58] So I know that these terms are overloaded and mean many different things to different people, but I am these days using AI to refer to chatbots and anything that's generative.
[08:09] Whereas ML is more of the traditional
[08:11] make a prediction.
[08:12] Um, so the Aqua Eye was one of the major ML projects we've been working on.
[08:17] On the AI side of things, I was working with another local company called Huge that's making a smart home assistant that lives on device entirely locally,
[08:24] so you can trust it with more personal data because none of that data will have to leave leave [clears throat] the the box.
[08:34] Um, so these are just kind of two of the projects that I work on, but in general, I just really love the um, real-world problem-solving component of consulting, which is very different than publishing papers, for instance.
[08:47] Um, as I'm sure many of your listeners know, publishing papers is extremely important and difficult, but there's all sorts of problems that come online when you're dealing with a real data set that you've actually had to collect and actually have to clean up.
[09:00] Um, and so those kinds of challenges have been just loads of fun to work on these days because you're actually able to save lives and and make a positive impact on the world.
[09:09] Wow, okay.
[09:09] This is super cool.
[09:11] And is Yeah, I'm not I'm not going to argue with you that podcasts are a great way to stay on up-to-date with the research with a lot more freedom and and same for that kind of research that you just talked about.
[09:28] Um, definitely resonates with me.
[09:30] And also I'm curious for Sophia AI, do you like what are you always seeking these kind of projects on or is that something like that's like is that project representative of what you do or do you usually do other kind of consulting?
[09:50] Yeah, I'm just curious how that works.
[09:52] I used to do a professional statistical consulting.
[09:56] So I'm always curious about other people.
[09:58] Yeah.
[10:00] Um
[10:03] So the word consulting can mean a lot of different things from
[10:04] Right.
[10:04] charging a hundred grand for a PowerPoint presentation to working in the in the weeds and programming.
[10:12] And then obviously there's like the strategic direction side of things if the company is newer.
[10:19] And so it kind of depends on the project.
[10:21] I like to I think I like to think of myself as a bit of a jack of all trades in the sense of having the research backgrounds to be able to um determine what kind of projects are feasible compared to we want to train a bot to predict the stock market kinds which just aren't feasible.
[10:38] And so sometimes the the projects are more at the high level of working with the founder and offering like strategic direction about Okay, if you want to do X, well you're going to have to do A, B, and C first and what's the the risks associated with A, B, and C.
[10:51] If the company is a bit more advanced if it's been around for a long time and they have a more concrete research problem then often I'm the person that they reach to when they need to figure out exactly how to run the experiments, how to validate the models, how to make sure that you haven't accidentally overfit and given yourself extremely um unrealistic positive results which which is
[11:13] something that's easy to um make that kind of mistake when you're new to this.
[11:17] So it very much depends on the client and and what their needs are.
[11:21] But one of the things I just love so much about let's say machine learning data science is that data gives you a
[11:29] Sorry, knowing how to work with data gives you an all access hall pass into everyone else's industry.
[11:34] So, I don't need to know much about how sonar works, but if I know enough about Fourier and signal stuff, then I can learn that.
[11:44] Um there's another company I was working for um that was doing predictive analytics for heavy machinery in the Amazon rainforest.
[11:50] And so, it was super cool to learn all about trucks and how trucks and and stuff work.
[11:54] Um so, I I like the I like the fact that each job is different.
[11:59] Each job requires really understanding um the unique problem situation that the company is in and what their goals are and then being able to offer some uh both strategic direction, but then also highly technical um instructions to to
[12:15] their team if if uh if they already have engineers or doing the first round of POCs and and prototypes to demonstrate that the idea is viable.
[12:21] So, it totally ranges um I have yet to charge 100 grand for a PowerPoint presentation, but I'm hoping such opportunities will present themselves.
[12:29] But uh but in general, it ranges from the actually doing it uh programming POCs all the way up to offering recommendations about which direction the the company should um uh should pursue.
[12:40] Okay.
[12:42] Okay.
[12:43] Oh.
[12:43] Basically, the whole range you you could do, but you you of course do a lot of of modeling because that's that's all part of your background.
[12:49] Yeah.
[12:49] Okay.
[12:51] Interesting.
[12:51] Um and and that makes sense also with my with my personal experience.
[12:56] Um yeah, sounds like you guys are doing very very interesting work.
[12:58] Well done on Yeah, that's a lot of fun.
[13:01] on on doing that.
[13:02] How many how many are you now in the company?
[13:06] Just one, baby.
[13:06] Um so, we um there's another idea which it's a bit too early to discuss right now, but um for that idea we're
[13:17] I say we because I'm an academic at heart and I cringe saying I, but um um but so, there's an idea that I'm potentially pursuing VC funding for and I'm going to go the more startup route and if we go that direction, then uh definitely we'll be hiring and stuff.
[13:31] But um but right now I like being agile, versatile, and being able to pick up projects as they come.
[13:38] Um there is a bit of a difficulty when you want to scale in the consulting world because you need to have enough steady clients coming in that you can um then guarantee steady work for the people you hire.
[13:48] And right now it's been just a lean kind of uh just me, my laptop, and just a bunch of networking and stuff.
[13:54] But um but if we decide to do the VC route, then I'll be kind of going in more of the hiring hiring direction.
[14:00] Yeah.
[14:01] Yeah, makes sense.
[14:04] Um so good luck with that and uh hopefully you get uh you get what you want.
[14:08] Yeah.
[14:10] If we focus on the Bayesian side of things now, um what drew you to to Bayesian methods in the first place?
[14:18] Yeah, so that's a good question.
[14:20] Um what drew me to Bayesian methods was partially just how beautiful the math was.
[14:27] Um Mhm. So there's loads of applications in Bayesian statistics as you and your audience are quite familiar with, but what I found personally exciting is just how deep the research opportunities are.
[14:38] Um and it's interesting cuz in industry I found that there've been less applications than I would have kind of guessed being in academia.
[14:45] But the applications in industry are are very much like the fundamental scientific level.
[14:49] So protein discovery or obviously um diffusion models are a a big one.
[14:58] Um and so I liked liked the um the coupling of pretty math with real-world applications.
[15:06] Um and my master's was um in a bit of a different domain.
[15:08] It was um about predicting whether someone Sorry, predicting the stage of um a patient's Alzheimer's uh diagnosis based on uh speech analysis.
[15:16] Um and so very
[15:20] kind of just uh black and white ML problem.
[15:23] You got a bunch of text, you compute features, you do parse trees, and you count morphemes and and all this, and then you train classifier.
[15:30] Um and so that was a great um experience to to kind of get uh a good understanding of like the applications of ML.
[15:37] But through that experience, I didn't feel like I had a good understanding of how these algorithms actually worked.
[15:43] Um there's a major difference between uh pulling off, say, a logistic regression model from scikit-learn versus actually like deriving the updates and and doing it on the whiteboard.
[15:54] Um and so the the pivot into Bayesian stuff was largely motivated by liking the applications and liking the opportunity to really dive deep into into how a lot of the stuff worked.
[16:04] Um and then that's uh that's what led to the PhD and and yeah, and so that's kind of how I got into it.
[16:09] Okay.
[16:10] Okay, so that was pretty pretty early in your career, right?
[16:16] This was even be- before even working.
[16:20] Yeah, so it was at the end of my master's, I took a late internship.
[16:26] And that was my opportunity to decide if I wanted to continue in the academia route or make the jump to to industry.
[16:34] Um actually did two back-to-back internships.
[16:35] One was at Samsung and the other was at uh in Tokyo.
[16:40] Actually, opposite way around, Tokyo then Samsung.
[16:43] Um and those two back-to-back experiences definitively uh convinced me that I want to stay in academia land.
[16:47] I didn't like the working at a megacorp uh aspect so much.
[16:52] Um It's nice to have like well-funded coffee machines and lunch stations and stuff, but um but I much preferred like the the difficult whiteboard sessions that um that the Tokyo experience um offered.
[17:06] And so um that kind of sealed the deal, and then I decided to do the PhD, and um uh and then never looked back.
[17:10] Yeah.
[17:13] Okay.
[17:13] Yeah, very cool.
[17:13] Yeah.
[17:18] Um and actually something you you distinguish a
[17:20] lot in your own work and and podcast is distinguishing Bayesian statistics from Bayesian epistemology.
[17:31] So, this is a very practical podcast, so we don't I mean, sometimes we talk about it like episode 50 with David Spiegelhalter, episode 51 with Aubrey Clayton.
[17:42] So, we do mention philosophy and epistemology from time to time, but we are usually more practitioners podcast.
[17:50] So, I'm very interested to hear about you today because um can you for listeners who haven't heard framed that way before, can you tell us what the actual difference is between statistics and epistemology?
[18:08] Yeah. Um so, that's a big question. Um I'll just lay my cards and my biases out on the table um cuz I am extremely in favor of Bayesian statistics because of the myriad of applications of Bayesian stats um has um
[18:23] just in researching this episode, I was just like ChatGPT, give me some just examples.
[18:26] So, obviously slam and common filters and VAEs is all well understood.
[18:32] Um MCMC, that's all well understood, but there is a nice example of I guess airline Air France 447 went down about a couple decades ago and Bayesian stats was used to be able to recover the the the wreckage.
[18:44] So, huge amounts of applications, that's great.
[18:48] Um Bayesian epistemology, I'm extremely critical of um and [snorts] I don't like it as an epistemology.
[18:55] Um I obviously haven't defined it yet, but um but that's kind of the the position I'm coming from.
[19:03] And so, um I'm going to give you an opinionated take on what Bayesian statistics or Bayesian epistemology is.
[19:07] Um and then we should probably talk about um epistemology in general because I think Bayesian epistemology is best understood um contrasted against another epistemology.
[19:19] Um But so if I can take a bit of a
[19:23] roundabout way to get there, the there's a podcast episode with Toby Ord and Sam Harris.
[19:33] Um and this podcast episode was uh came out maybe five or six years ago.
[19:38] Um and in this episode um Ord was talking about his most recent uh book uh called The Precipice at the time.
[19:45] And he was talking about existential risks.
[19:47] Um and so he was talking about uh so what's the probability that we will die from an asteroid um uh or what's the probability that we'll die from a volcanic eruption.
[19:57] Um and so you can assign numbers to these because you have a data set to um to look at.
[20:02] So you can look at the number of um super volcanoes in the geological record over the last couple hundred million years and then you can start doing some basic um uh uh take account divided by time to get the probability associated with it.
[20:19] Uh but then he switched and he started talking about what's the probability that we will all die from superintelligence in
[20:24] the next hundred years.
[20:26] Um and then he assigned a probability to
[20:28] that too.
[20:30] But if you are a statistician, if you're
[20:32] a Bayesian statistician, I think your
[20:33] first question should be, well, where
[20:35] are the statistics? What's the what's
[20:38] what are we putting into these models?
[20:39] How can you come up with a probability
[20:42] of something that's going to happen in
[20:44] the future when there's nothing to
[20:46] count, right? Um and in particular uh
[20:50] how is it a valid move to compare
[20:53] probabilities derived without data
[20:56] um against probabilities that come from
[21:00] data? Because in this episode um Ord was
[21:03] essentially making the point that um we
[21:05] have a one in I believe in this book,
[21:07] it's one in 10 chance of extinction due
[21:11] to superintelligence. Uh he contrasted
[21:13] that against one in, I'm just going to
[21:15] make up a number, one in
[21:17] a billion chance of of earthquake.
[21:19] Um so that got my little radar
[21:22] beinging, my little [ __ ] detector
[21:24] beinging, because
[21:26] um if you have a model and you train
[21:28] that model on zero data,
[21:30] how is that a model? Where like where is
[21:32] the statistics?
[21:34] So
[21:34] the answer to your question, the kind of
[21:36] the tongue-in-cheek answer is that
[21:37] Bayesian epistemology
[21:38] is Bayesian statistics minus the
[21:40] statistics. It's just the equations. Um
[21:43] how that actually works in practice and
[21:44] why some philosophers, um
[21:47] particularly ones that come out of
[21:49] Oxford these days,
[21:50] why that they consider that to be a
[21:52] valid move, where
[21:54] all other statisticians, Bayesian or
[21:56] frequentist, would say,
[21:57] "Hold on, if you are drawing a scatter
[21:59] plot and you do a line of best fit
[22:02] on an empty whiteboard
[22:04] with no dots, you're just doing lines on
[22:06] a whiteboard, how is that a model? So
[22:10] the answer to your question, um
[22:13] at a surface level, is that Bayesian
[22:14] epistemology is what happens when you
[22:16] take Bayes' theorem way too seriously
[22:18] and you forget about the data. And then
[22:19] you start coming up with numbers and you
[22:21] kind of convince yourself that these
[22:23] numbers are legitimate because you're
[22:24] calling them probabilities, when in fact
[22:26] they're made up they're made up out of
[22:28] your head and you then are making
[22:30] decisions based on made up numbers which
[22:32] you've called a probability. Now,
[22:34] um if there was someone who was a
[22:36] Bayesian epistemologist on this podcast,
[22:38] they would strongly disagree with my
[22:40] description of this.
[22:42] Um
[22:42] and so
[22:43] uh from their perspective, which we we
[22:45] can maybe steel man in in in a bit here,
[22:48] they have a different story to tell. So
[22:50] I I do want to be clear to the audience
[22:51] that I'm giving a biased
[22:53] accounting. Um and so maybe I'll I'll
[22:55] pause there to let you kind of
[22:58] poke and prod at some of the stuff I
[22:59] said, but
[23:00] but it is an interesting question. Why
[23:03] are some philosophy departments okay
[23:05] with using
[23:06] probabilistic estimates that don't come
[23:08] from any data? And the answer to that is
[23:11] due to 100 years of literature that can
[23:14] all be kind of considered to be Bayesian
[23:16] epistemology.
[23:18] Mhm.
[23:18] Mhm. Okay. Yeah, thanks. That's very
[23:21] clear.
[23:22] Um
[23:24] It's very clear
[23:26] explanation of of the difference.
[23:29] I really find that interesting.
[23:31] So, I spend most of my time thinking
[23:33] about statistics
[23:35] and models and not not so much about
[23:36] philosophy and epistemology. But
[23:39] so, I do have a few questions. It you
[23:42] know, I'll mainly re- try and rephrase
[23:45] what you told me to make sure I
[23:46] understood. So, please let me know if I
[23:49] didn't. Um
[23:52] The main thing that
[23:55] bothers you about Bayesian epistemology
[23:59] is that
[24:01] is in the scenarios where you don't have
[24:04] data to
[24:07] update your beliefs.
[24:08] Is that right?
[24:10] Um I would frame that as that's the most
[24:13] immediate example of the problems with
[24:16] Bayesian epistemology. That's one that
[24:20] I think a listener can
[24:22] grok without having to talk about Cox's
[24:24] theorem or why, for example, um we
[24:29] conflate
[24:30] probability distribution with the
[24:32] psychological phenomenon of having
[24:34] beliefs in the first place.
[24:36] There's questions to about why that
[24:38] connection even makes sense. So, there's
[24:40] a lot of um
[24:42] philosophical reasons why I don't think
[24:44] it's going to work.
[24:45] But I like to start with a clear example
[24:49] of of how it can get us into trouble.
[24:51] And
[24:53] so, I guess my my my comment there is
[24:55] that that's just the tip of the iceberg.
[24:57] But um but that's one one reason because
[24:59] I think it can
[25:00] um
[25:01] uh
[25:01] trick people into making decisions that
[25:03] are poorly informed because they're
[25:05] using a lot of math. And um and using a
[25:07] lot of math is not sufficient to making
[25:09] a good decision. But Bayes' theorem
[25:13] s- divorced from the data, when it's
[25:15] just Bayes' theorem, that I think can be
[25:17] quite um enticing to people. Um and it
[25:19] can lead them lead them astray. So, that
[25:21] would be how I
[25:22] Yes, so I do I do fully agree with the
[25:24] danger of basically math washing. I
[25:27] think it's used a lot in in politics and
[25:30] and basically like people who want to
[25:34] convince you of something they think is
[25:36] true.
[25:37] And then they just try to
[25:41] um
[25:43] to just shoehorn
[25:45] what they already believe into a math
[25:47] package to give it the
[25:50] um
[25:51] the illusion of objectivity.
[25:54] [sighs]
[25:56] Um
[25:57] what I'm So, I'm less disturbed by you
[26:01] are
[26:02] by the fact of using Bayes' theorem
[26:06] with our priors, without data. Sorry.
[26:10] Because
[26:11] so, one thing I
[26:13] thought I understood, but I may be
[26:15] completely wrong on that,
[26:17] is
[26:19] in the Bayesian
[26:21] epistemology framework,
[26:22] you define probability not as a
[26:26] long-term frequency,
[26:28] but as a degree of belief.
[26:30] So, if you accept that definition of
[26:32] probability,
[26:34] then using Bayes' theorem without data
[26:38] is just regurgitating your priors.
[26:42] And
[26:43] to me, that's not too much of a problem
[26:45] of an issue
[26:47] if you say so.
[26:48] Like, so I think my my problem is more
[26:51] when people do that thing and for
[26:53] instance your your example I think with
[26:55] the Sam Harris podcast and the 10%
[26:58] probability that we get wiped out by 10
[27:01] AI. I think I listened to that podcast
[27:03] actually. Is that a recent one?
[27:05] No, it was older. That was the one of
[27:07] the conversations that got me interested
[27:10] in this subject in the first place.
[27:11] Okay.
[27:12] But I would say it's probably four or
[27:14] five years old. Uh but
[27:15] Mhm. Okay.
[27:16] Yeah.
[27:17] So
[27:18] Yeah, because I I've seen I've also seen
[27:20] a report like that but what the impact
[27:22] of 10 AI would be in 2027 or things like
[27:25] that. And they were doing a lot of
[27:27] things like that where it's like I don't
[27:29] know what like you don't have any data
[27:31] to really back that up. So you're
[27:33] basically taking an assumption
[27:35] um and then just thinking about how we
[27:38] could get there, which is fine as So and
[27:41] again to me this is fine as a thought
[27:45] experiment.
[27:46] The problem is that
[27:49] sometimes you can get convinced by you
[27:51] can convince yourself that actually this
[27:54] is going to happen and forget that you
[27:55] base that on an assumption.
[27:57] Or
[27:58] in a less
[28:00] um let's say um
[28:03] in a less nice way of thinking about
[28:06] that uh it would be more like Molly's
[28:09] where it's be I'm going to package that
[28:12] as something that is actually objective
[28:15] and instead of saying it's just
[28:18] reflecting my decrease of belief about
[28:20] something that can happen i.e. my priors
[28:24] I'm going to say this is actually a
[28:25] probability of long-term event. So you
[28:27] basically in the communication
[28:30] of the output you swap the definition of
[28:33] of probability that you had at the
[28:35] beginning
[28:36] and I would say that is what really
[28:39] bothers me here.
[28:41] Because if you stop that well, I just my
[28:44] priors are are that and so I think the
[28:46] probability of turning out wiping us out
[28:48] is 10%.
[28:49] Uh and it's my belief.
[28:51] Okay, like you you didn't really apply
[28:54] base theorem. This is just the basis of
[28:56] base theorem, but
[28:58] Okay, but I mean, why not? Um
[29:02] So, yeah, I'll stop here because I
[29:05] talked a lot already.
[29:06] Yeah, no, totally. So, I think we're
[29:08] we're um
[29:09] We share intuition.
[29:12] I would maybe agree and sharpen what you
[29:15] said, which is that
[29:17] um
[29:18] I've no problem whatsoever with someone
[29:20] saying, "My opinion is that we will be
[29:23] wiped out next year
[29:25] with one in 10 chance." Or
[29:28] um my gut feeling is that we will be
[29:31] wiped out. Or my
[29:33] "What's your prior on the question of
[29:35] whether or not you're going to get the
[29:36] job next week?" Well, my my hunch is
[29:39] that it's going to be it's pretty
[29:41] likely. That's all totally fine and
[29:43] great.
[29:44] Mhm.
[29:44] Um the difficulty or the where the the
[29:47] trap lies, I think, is that when one
[29:49] starts talking in terms of priors, and
[29:51] then when when starts talking in terms
[29:53] of actual probabilities,
[29:55] [sighs and gasps]
[29:56] now you're comparing two things that
[29:58] have extremely different origins. So,
[30:00] for instance, um we can come up with
[30:03] a probability that um COVID-19
[30:07] is going to um
[30:10] let's say uh
[30:11] cause 10% of the population to be
[30:13] hospitalized, right? And so, how can we
[30:14] come up with that? We can use Bayesian
[30:15] methods. So, we could use like
[30:16] approximate Bayesian computation. We
[30:18] could get a giant simulator. Um so, it's
[30:21] actually a project I worked on, um where
[30:23] you get a big simulator that literally
[30:25] has a computer program to
[30:27] say, "What's the probability that you're
[30:28] going to go to work and you're going to
[30:29] touch something?" And uh so, we're
[30:32] talking like 100,000 lines of C++ code
[30:35] with like deep assumptions based on the
[30:38] spread of disease and how the city
[30:41] works, and it's something that grad
[30:43] students have been working on for 15
[30:45] years to be able to come up with these
[30:46] models. And that gives you a
[30:47] probability. And then you say, "Okay,
[30:48] the probability is one in 10." I'm just
[30:51] going to keep using the same numbers.
[30:52] And then someone else says, "Well, my
[30:54] prior is one in 13."
[30:57] And now the fact that the first group is
[31:00] is using Bayes' theorem and uh
[31:04] practicing Bayesian statistics, but
[31:05] they're still counting stuff. They're
[31:08] still coming up with frequencies, and
[31:11] the frequencies are Monte Carlo runs,
[31:13] right? Um
[31:15] and that number is then compared to
[31:18] someone's gut feeling.
[31:20] But people forget that they're switching
[31:21] between these two extremely different
[31:24] assumptions about what the nature of
[31:26] probability is. And so if you license
[31:28] yourself to use a subjective
[31:30] um interpretation of probability,
[31:33] um and then you kind of forget that, and
[31:35] then you compare it to counting
[31:36] asteroids or or looking at the MCMC runs
[31:39] from a giant test simulator. Um now
[31:42] you're comparing someone's gut feelings
[31:44] to
[31:45] 30 years of research, right? Um and this
[31:48] is exactly what Toby Ord did in his
[31:51] book, and it's escapes even someone as
[31:53] as uh wise as Sam Harris. Um because uh
[31:57] if I just tell you what the probability
[31:59] is, then you think, "Okay, got it.
[32:01] That's some objective thing that people
[32:02] have have calculated." When in fact, no,
[32:06] no, no, probabilities are not all made
[32:07] equal, and probabilities derived from uh
[32:10] models um derived from data. Like
[32:13] even when we talk about a prior, like
[32:15] you can absolutely do empirical priors
[32:18] where your priors are informed by data.
[32:19] So nothing is um intrinsically
[32:23] um data independent about priors because
[32:27] there are techniques to inform your
[32:28] priors um if you're taking a Bayesian
[32:30] statistical um approach. Uh and so um
[32:35] my my main rub against Bayesian
[32:38] epistemology is that I just don't think
[32:39] that is how knowledge is produced and
[32:41] how we actually make scientific
[32:43] progress.
[32:44] Um I think we make that through
[32:46] um Karl Popper's um
[32:48] philosophy conjectures and refutations
[32:50] which
[32:51] uh I'm I'm sure we'll go into it at some
[32:52] point. Um so my my biggest problem is is
[32:55] that uh I just don't think that this
[32:56] actually describes how knowledge is
[32:58] produced. But my more immediate problems
[33:01] and the more applied problems are that
[33:03] it um
[33:04] allows people to compare numbers that
[33:05] have extremely different origins and
[33:07] then potentially make extremely uh bad
[33:09] decisions based on this. So one
[33:11] extremely salient example is uh Sam
[33:13] Bankman-Fried. So if you read um
[33:17] uh the book that came out um by
[33:19] uh
[33:20] Going Infinite. I I forgot the author's
[33:21] name at the moment. Um
[33:23] Same guy who did uh The Big Short. Um
[33:25] you can just
[33:26] read that Sam Bankman-Fried just viewed
[33:29] everybody and everything as just walking
[33:31] probability distributions. Um and now
[33:34] he's in jail. And he's in jail because
[33:35] he has a certain um he made certain
[33:38] decisions based on uh
[33:40] faulty epistemology um that uh that
[33:43] wouldn't have happened uh had he not
[33:45] been um
[33:46] uh drunk off probabilities, shall we
[33:48] say?
[33:49] [snorts]
[33:49] Yeah.
[33:50] Yeah. And so let's try and make that
[33:54] a bit concrete.
[33:56] So basically it's
[33:58] let's go back to the part the part you
[34:00] like, the Bayesian statistics part.
[34:03] When a working statistician fits a model
[34:06] with priors and a likelihood,
[34:08] what work is the Bayesian machinery
[34:11] doing that you think is legitimately
[34:13] useful?
[34:14] Oh. Um yeah, so Bayesian statistics for
[34:17] your audience will know this for sure,
[34:19] but um
[34:20] uh [snorts] Bayesian statistics is
[34:21] differentiated from say a frequentist
[34:23] statistics in the sense that it uses
[34:25] random variables as your parameters,
[34:27] right? Um and so uh
[34:30] that gives you um a lot of, uh,
[34:33] freedom to incorporate in knowledge that
[34:37] doesn't really
[34:39] um, uh, have an entry point, um, uh, via
[34:42] other other means. So, for example, if
[34:45] we are trying to, um,
[34:49] get an estimate of the average height of
[34:53] people in Canada, let's say. Um, well,
[34:55] that's a really nice domain for Bayesian
[34:56] statistics because you can put a prior
[34:58] over what you think the height's going
[34:59] to be. You know the height's not going
[35:00] to be zero. You know the height's not
[35:02] going to be 100 ft. Um, and so that's
[35:04] just a really nice way to to incorporate
[35:07] in some just domain knowledge into your
[35:09] problem such that the the answer you
[35:11] get, um, is is making use of of this
[35:14] common sense information we have about
[35:16] the problem.
[35:17] Um,
[35:18] and that's all fantastic and in
[35:19] particular it's fantastic because you
[35:21] can get a model, a Bayesian model, and
[35:23] then you can compare it to your data and
[35:25] you can be like, "Oh, my model is wrong.
[35:27] This was a bad model. I need to go back
[35:29] and I need to reassess." Um,
[35:31] and so to the extent that Bayesian
[35:33] methods when used in statistics are
[35:35] falsifiable and you can use data to
[35:38] figure out if your modeling assumptions
[35:39] were, uh, bogus or or or not bogus, I
[35:42] think it's it's fantastic. Um,
[35:44] so that's just one example, but you can
[35:47] talk about how variational autoencoders
[35:49] work. Um, you can talk about uh, MCMC
[35:53] and and there's all sorts of real-world
[35:55] problems for which, um, this assumption
[35:58] that your random variables uh, sorry,
[36:00] that your parameters are random
[36:01] variables makes a lot of sense and it's
[36:03] been validated and it works in practice.
[36:06] Um,
[36:07] I wouldn't say that it's better or worse
[36:09] than frequentist statistics. I would say
[36:10] it's just a different kind of statistics
[36:12] and any statistician these days is going
[36:14] to be practicing both and they're going
[36:16] to know, uh, about the kinds of problems
[36:19] for which, say, a bootstrap estimator
[36:21] or, um,
[36:23] uh,
[36:23] or other frequentist estimators of
[36:25] uncertainty are appropriate and in other
[36:27] problem domains where Bayesian
[36:29] statistics are appropriate. So, from the
[36:31] perspective of like the Bayesian
[36:32] frequentist debate as applied
[36:36] to like engineering problems,
[36:38] no dog in that fight. I think both have
[36:40] their pros and their cons.
[36:42] Um where I have more of a dog in that
[36:44] fight is when it comes to epistemology
[36:45] and when it comes to saying things like
[36:48] um the way that Newton was able to come
[36:50] up with his theories was using uh Bayes'
[36:52] theorem in his head somehow. Um and so
[36:55] when it's used as a model of scientific
[36:57] discovery,
[36:59] um that's where I think one can start
[37:01] running into uh difficulties quite quite
[37:03] um
[37:04] um [snorts]
[37:05] quite severe ones, in fact, yeah.
[37:07] Mhm. Yeah, so let's
[37:09] let's talk about that. What's the
[37:12] central thing
[37:14] concretely on
[37:16] that that Bayesian epistemology is a
[37:18] general framework for reasoning under
[37:20] uncertainty gets wrong, according to
[37:22] you?
[37:24] If I may take a bit of a historical lens
[37:27] um on that. Um so on our podcast, we
[37:30] talk about Karl Popper a lot. He's uh
[37:33] philosopher that had a big impact on me.
[37:35] Um and he also um
[37:37] was uh writing
[37:39] basically he's born in 1902 and he died
[37:42] in 1994,
[37:43] meaning he was writing throughout the
[37:45] development of uh measure theory.
[37:47] He was writing when all of these
[37:49] probabilists were were really working on
[37:51] on um probability theory. That was when
[37:54] like the Banach-Tarski paradox was
[37:55] discovered. That's when Kolmogorov
[37:57] started working.
[37:58] Um and so uh throughout the 20th
[38:01] century, there was a huge amount of
[38:03] development uh on probability theory. Um
[38:06] and
[38:07] a sizable percentage of the these uh
[38:10] economists and philosophers, um not so
[38:12] much the mathematicians, they were kind
[38:13] of just working on their measure theory,
[38:15] um wanted to use this as a framework to
[38:19] explain where knowledge comes from.
[38:22] Um, and
[38:24] and that's essentially where Popper
[38:27] says, "Eh, hold hold your horses here,
[38:29] fellas. Um, I don't think it works that
[38:31] way." Um,
[38:33] and
[38:34] so
[38:37] maybe one way to to enter this
[38:39] conversation is to
[38:41] um, [snorts]
[38:42] think about
[38:44] Newton's theories when they were
[38:46] first uh I guess developed in the 16
[38:48] 1700s. Um,
[38:50] when they were developed, they were
[38:52] basically seen as like
[38:54] we have finally found true knowledge.
[38:58] Like this again this is pre-Einstein. So
[39:00] this is before we realized that Newton's
[39:02] theories were wrong. And Newton's
[39:03] theories just gave us this like insane
[39:05] amount of predictive
[39:08] um power and the same amount of
[39:10] engineering capabilities that allowed us
[39:11] to uh
[39:13] figure out how to shoot cannonballs, it
[39:15] allowed us to explain um how the moon
[39:18] stays up and and it's um its orbit. Um,
[39:21] and it so just it told us
[39:23] so much not only about the things you
[39:26] can immediately see, but it told us
[39:28] stuff about stuff we can't see, stuff
[39:29] that's like in another galaxy. It
[39:31] explains how stars would planets would
[39:34] orbit stars. Um,
[39:36] and so this was the problem that a lot
[39:37] of philosophers are trying to figure
[39:38] out. So how is it possible that human
[39:41] beings
[39:43] coming up with scribbles on their piece
[39:45] of paper in some place in Europe um are
[39:49] able to discover uh general truths about
[39:52] the entire cosmos, truths that they
[39:54] haven't themselves seen or don't have
[39:57] immediate access to. So you can't touch
[39:59] a force. You can't see a force. You can
[40:02] see a
[40:03] a bowling ball fall. You can hypothesize
[40:06] that it falls because of a force, but
[40:09] you can't actually see it, right? Um,
[40:11] and so uh
[40:13] the traditional answer for how this
[40:16] comes about Um, is via this process
[40:19] called induction.
[40:20] Um, and so induction in the Popperian
[40:24] worldview is like a a dirty word. You
[40:27] induction is something that the we
[40:29] Popperians don't don't like very much.
[40:30] But what is induction? Induction in this
[40:32] case simply means knowledge is acquired
[40:36] via repetition. So, um, the way that uh,
[40:39] Newton
[40:41] um, was able to discover his laws was by
[40:44] repeatedly observing um, apples falling
[40:47] from trees, say.
[40:49] Um,
[40:50] now Bayesian statistics
[40:52] uh, Bayesian statistics is just a form
[40:55] Sorry, Bayesian epistemology. Bayesian
[40:56] epistemology. Um,
[40:58] is just a a
[40:59] a form of induction. It's it's this
[41:02] um,
[41:03] uh, it was an attempt to fix what is
[41:05] called the problem of induction. And the
[41:07] problem of induction is just that
[41:09] um,
[41:10] all these philosophers said that you
[41:11] gain knowledge by repeated observations.
[41:14] And yet it's logically impossible that
[41:17] seeing an apple fall a whole bunch of
[41:18] times can tell us a general truth about
[41:20] what's happening in the Andromeda
[41:22] galaxy. Um, and so uh, philosophers from
[41:26] like Francis Bacon to all the way to
[41:28] Bertrand Russell, um, were wrestling uh,
[41:30] and and David Hume is the guy who uh,
[41:32] who named the problem of induction.
[41:35] They were wrestling with this apparent
[41:36] contradiction. And the contradiction is
[41:38] on one hand all the philosophers saying
[41:40] that we get knowledge because of
[41:41] repeated observations. And yet logically
[41:44] it's impossible to get any knowledge
[41:45] about something far away based on
[41:47] immediately what you see.
[41:49] So then Bayesian epistemology um, was an
[41:52] attempt to patch this.
[41:54] So it was an attempt to say, "Okay,
[41:55] sure. Maybe we can't be absolutely
[41:57] certain
[41:59] um, that uh, Newton's laws are true
[42:03] given the things we've seen.
[42:05] But maybe we can be probably certain. Or
[42:07] maybe we can be um, less uncertain.
[42:10] Um, and this is uh,
[42:12] what a lot of work has been um,
[42:15] spent on is trying to basically use
[42:18] probability theory to patch the problem
[42:20] of induction.
[42:21] Um,
[42:22] Now, so your question was like what's
[42:25] what's the big problem with this and why
[42:26] why don't why don't I think that this
[42:29] works?
[42:30] Um,
[42:31] [snorts]
[42:31] Well, so for a number of reasons. So,
[42:33] um,
[42:34] the in the Bayesian framework, this is
[42:38] um,
[42:39] repeated observations, so you call them
[42:40] repeated so you see a bunch of evidence,
[42:42] right? Um, and now when I say evidence,
[42:44] I'm not meaning a row of a CSV, right?
[42:47] So, a row of a CSV is totally legit.
[42:50] That's that's data. What the Bayesian
[42:52] epistemologist meant is just
[42:54] you just see it. You just open your eyes
[42:56] and you just see evidence. Um,
[42:58] but the question is well, what evidence?
[43:00] So, what counts as evidence
[43:03] only counts in light of the theory. So,
[43:06] you have to have your theory first
[43:08] and then based on your theory, you can
[43:10] figure out what counts as evidence. But,
[43:13] it's a circularity or it's an infinite
[43:15] regress. If you're going to say that
[43:17] your theory comes from evidence,
[43:19] then that just doesn't make any sense.
[43:22] Um,
[43:22] so there's one major issue which is that
[43:25] um,
[43:26] uh, when Bayesian epistemologists try to
[43:28] explain
[43:29] where Newton's theories come from, they
[43:31] start with a prior over H um, and then
[43:35] you have your likelihood evidence given
[43:37] H, but you've already
[43:40] kicked the can down the road because the
[43:41] question is where does the where does H
[43:42] come from? And so H can't come from
[43:44] evidence. It has to come from some other
[43:47] process. Um, so that's one problem
[43:49] [snorts]
[43:50] with it. I think the other major problem
[43:52] with Bayesian epistemology is that it
[43:54] encourages you to find evidence in
[43:58] support of your hypothesis.
[44:01] Um, every astrologer
[44:03] on the planet, every conspiracy theorist
[44:05] on the planet looks to find evidence in
[44:08] support of their theory. Um it's
[44:09] extremely easy to find evidence that
[44:12] supports your view that homeopathy is
[44:14] going to cure your cold. Um it's
[44:16] extremely easy to find further evidence
[44:19] that
[44:20] um 9/11 was an inside job. If all you're
[44:23] looking for is evidence to support your
[44:26] preferred view, then you're going to
[44:27] find a lot of that quite easily.
[44:29] Um what Popper did, and we can
[44:32] punt on talking about Popper um as long
[44:34] as possible, but he inevitably rears his
[44:36] rears his uh his head when um the
[44:39] subject comes up. Um but Popper's whole
[44:41] point was that no, no, no, the
[44:43] scientific mind is not looking to find
[44:45] evidence in that confirms the
[44:48] hypothesis. It's always trying to find
[44:49] evidence to disconfirm your your
[44:51] preferred hypothesis. Um and so I think,
[44:55] to summarize, my main problems with
[44:56] Bayesian epistemology is that one,
[44:58] logically it just doesn't make any
[45:00] sense. Um you can't explain where a
[45:03] hypothesis comes from by talking about
[45:05] evidence in support of it because the
[45:06] evidence only works given the
[45:08] hypothesis.
[45:09] Um
[45:10] and two, it
[45:12] produces
[45:14] dogmatism.
[45:15] Um it produces this this idea that you
[45:18] should only look for stuff that supports
[45:20] your your view.
[45:22] Um and then three, which I haven't
[45:23] mentioned yet, but we can go into it, is
[45:25] that as soon as you start trying to
[45:26] patch the problem of induction with math
[45:29] on top of it, you're basically building
[45:31] an edifice on a logical impossibility,
[45:34] and then infinite numbers of paradoxes
[45:36] emerge. So, one of the major paradoxes
[45:38] in Bayesian epistemology is something
[45:40] called um Hempel's paradox or the raven
[45:43] paradox,
[45:44] uh which some of your listeners may be
[45:45] familiar with this, but uh it's this
[45:47] idea that um
[45:49] if we're trying to find evidence in
[45:50] support of the theory that all ravens
[45:52] are black,
[45:53] um all ravens are black. So, that is
[45:56] equivalent to if a thing is a raven,
[45:59] then it is black. And as soon as you
[46:00] have an if-then statement, you can take
[46:02] the contrapositive, which is if a thing
[46:04] is not black, then it's not a raven.
[46:06] And then that would um count
[46:09] logically
[46:11] it would be equivalent to um seeing a
[46:13] black raven. So, Hempel's paradox or the
[46:16] raven paradox is just the idea that um
[46:20] any green shoe you look at or any orange
[46:24] um
[46:26] ball or any uh
[46:28] what is it? San Pellegrino cup that's
[46:30] not black would be in support of your
[46:32] hypothesis that all ravens are black.
[46:34] So, how can you possibly
[46:37] claim to have knowledge and
[46:38] understanding about how ravens work by
[46:41] looking at green shoes.
[46:43] Um
[46:43] [snorts]
[46:43] so, Bayesian epistemology, because it is
[46:45] so fundamentally subjective, because it
[46:48] really encourages you to go inwards and
[46:50] think about your beliefs, and it doesn't
[46:52] encourage you to go outwards and
[46:54] experiment, and it doesn't encourage you
[46:56] to go outwards and actively find
[46:58] information that goes against your
[46:59] beliefs, and actually try to disconfirm
[47:01] your beliefs.
[47:02] Um I think it causes people to run in
[47:05] circles. I think it causes people to
[47:07] become extremely dogmatic, as Sam
[47:09] Bankman-Fried I think is um evidence of.
[47:11] Um and I think it logically just doesn't
[47:13] make any sense. Um and so, those are the
[47:14] main main reasons uh
[47:16] why I am not a big fan of Bayesian
[47:19] epistemology, but still I'm a huge fan
[47:20] of Bayesian statistics, and that line is
[47:22] something that uh
[47:23] I constantly have to walk in the other
[47:24] direction
[47:25] uh with like our audience on the
[47:27] podcast, because some people who are
[47:29] extremely anti-Bayesian now will want to
[47:31] write off all of Bayesian stats, too,
[47:32] and I'm have to be like, "No, no, no,
[47:34] hold on. Bayesian statistics is great,
[47:36] because you have data, because you can
[47:37] be falsified by your data and your and
[47:39] your models, but Bayesian epistemology
[47:41] is less um less good, shall we say? So,
[47:44] yeah. Yeah.
[47:45] Right. Right. So, the main like one of
[47:47] the most potent points
[47:50] in your critique is that
[47:52] what it whatever the name actually of
[47:54] your framework of epistemology
[47:57] whether that's Bayesian or anything
[47:59] else,
[48:00] it needs to be falsifiable, and it needs
[48:03] to have falsifiability as a first-class
[48:06] citizen in it.
[48:07] Otherwise,
[48:09] you can't really know
[48:12] when you're producing knowledge or
[48:14] you're just producing more evidence for
[48:17] your prior beliefs.
[48:20] Is that correct?
[48:21] With
[48:23] a few asterisks. Um so,
[48:26] the
[48:28] um the first asterisk is that that's
[48:29] absolutely true for what we consider to
[48:31] be science. So, a scientific hypothesis
[48:34] has to be falsifiable.
[48:36] Um
[48:37] but philosophy is not falsifiable. Um
[48:40] there's all sorts of ideas that aren't
[48:42] falsifiable, but yet are still extremely
[48:43] valuable.
[48:45] Um and so, Popper's philosophy, he
[48:47] called it critical rationalism.
[48:49] Um or
[48:51] inverted, you can say it's all about
[48:52] rational criticism. So, the
[48:55] generalization
[48:57] of falsifiability is criticism. Um
[49:00] and so, Popper's epistemology is
[49:04] takes criticism as a first-class
[49:06] citizen, not falsifiability. So,
[49:09] falsifiability um would be a kind of
[49:13] criticism. In particular, it's an
[49:14] empirical kind of criticism. Um where
[49:17] um if you are so lucky as to come up
[49:20] with a theory that can in fact be
[49:23] tested, then your whole goal is to try
[49:25] to falsify that theory. Um
[49:27] but there's all sorts of theories for
[49:30] which they just don't have tests. Um
[49:33] The integral calculus. How can you
[49:35] falsify the integral calculus? Well,
[49:37] um you
[49:39] can't run experiments to falsify it, but
[49:41] you can absolutely uh criticize it. You
[49:43] can absolutely
[49:44] um
[49:46] discover things inside it that don't
[49:47] make any sense, such as um
[49:50] uh
[49:50] what I mentioned earlier, the Banach
[49:51] Tarski paradox and and so
[49:54] Popper's epistemology takes criticism as
[49:58] the fundamental
[49:59] unit, the most important thing.
[50:01] Um, and then various types of criticism
[50:05] um, apply to various kinds of domains.
[50:07] So in the scientific domain,
[50:09] um, experimental tests are are one of
[50:11] the the most important kinds of
[50:12] criticism, but also peer review. Also
[50:14] um, the peer review process, that just
[50:17] explaining your idea to a friend on a
[50:19] podcast and having them be like, "Huh,
[50:21] what does that mean?" That's a kind of
[50:23] criticism that's entirely valid because
[50:25] the whole goal is to error correct um,
[50:27] all of the bugs in your thinking. It's
[50:29] not to
[50:30] demonstrate that what you believe to be
[50:33] true has high probability, for example.
[50:35] Um, and so falsifiability is is what
[50:37] people know Popper for um, most because
[50:40] that's one of the his like
[50:42] most important contributions, but that
[50:43] contribution he made when he was 16.
[50:46] Um, and that was kind of like one of the
[50:47] first things he did, but then what he
[50:49] went on to do was um,
[50:52] uh,
[50:53] generalize that concept so that it
[50:55] applies not only to to science, but to
[50:58] um,
[50:58] philosophy and and metaphysics and um,
[51:01] art as well. Art criticism is just as
[51:03] valid or stand up comedy. Um, so you can
[51:06] you can find domains for which they get
[51:08] immediate real world feedback and
[51:09] criticism
[51:10] um, and those are the domains that are
[51:12] the ones that are going to um,
[51:14] uh, uh,
[51:16] make much more progress because they're
[51:17] the ones that have some sort of feedback
[51:20] mechanism that tells them when they're
[51:21] wrong. And so that's the most important
[51:23] thing. Experimental tests are a kind of
[51:25] that, but it's not the only form of
[51:27] that.
[51:28] Mm, okay, okay. That's fascinating. So
[51:31] basically
[51:32] critical and I was going to ask you what
[51:34] about Popper and critical rationalism.
[51:37] So so that's great that you did this
[51:39] already.
[51:40] Um,
[51:41] and basically
[51:43] that philosophy includes
[51:46] falsifiability with experiments in
[51:48] science, but it's not reduced to that in
[51:51] the sense that criticism will critical
[51:55] rationalism will be helpful for other
[51:59] domains like art, stand-up comedy as you
[52:01] were saying.
[52:02] Where you cannot falsify, but you can
[52:05] criticize rationally.
[52:07] Um
[52:09] What I would like to understand better
[52:11] is
[52:13] how does critical rationalism work
[52:16] concretely when you cannot make
[52:17] experiments? So,
[52:20] that can have to do with stand-up comedy
[52:23] for instance, which I'm a big fan of.
[52:25] It's one of my hobby. So, definitely
[52:27] take that as an example, please.
[52:30] But also
[52:31] still in science, you have
[52:35] Um you have cases where you cannot run
[52:37] experiments, but you can make
[52:39] inferences. Causal inference on quasi
[52:41] experiments,
[52:42] uh for instance, is is an example.
[52:46] How would that work here?
[52:48] Um so,
[52:50] inferences is a word that
[52:53] has a specific meaning for machine
[52:55] learning people. Um basically, when
[52:57] you're not in training mode, you're in
[52:59] inference mode. But maybe for the
[53:01] people outside of ML,
[53:03] often inference just means like um
[53:07] some
[53:09] secondary conclusion that is not
[53:11] immediately apparent, but yet can be
[53:14] inferred or guessed or deduced or
[53:18] somehow you get from A, which is the
[53:20] immediately obvious thing, to some B,
[53:23] which is less obvious. Um so, I'm going
[53:26] to take
[53:27] take inference to mean that more general
[53:29] um
[53:30] notion if if if that's a fair way to
[53:32] understand your your question.
[53:34] Um
[53:35] And so, if your question is how
[53:39] do we get from A to B? How do we um uh
[53:44] uh start with some information like it's
[53:48] raining outside to an inference like I
[53:51] should bring my umbrella, right? Um
[53:54] what Popper says is essentially he
[53:57] doesn't care.
[53:58] Um and that's going to not be as vacuous
[54:00] as I as as I as initially sounds. So,
[54:03] you could get from A to B because B
[54:06] logically follows from A. So, it could
[54:09] be
[54:10] A is I think that DNA is wound as a
[54:13] double helix. Uh I'm going to think
[54:15] really hard and deduce that that would
[54:17] lead to an experiment where I shine a
[54:19] light on it in a particular way and it
[54:20] make a particular shadow pattern um on
[54:22] the wall. So, that would be like a
[54:24] deductive inference and that works in
[54:26] some cases it's
[54:28] not most cases. Um sometimes uh
[54:30] you're lucky enough that you can
[54:31] logically deduce um secondary
[54:33] information. Uh other times you just
[54:35] need to make a wild crazy guess. Um
[54:38] there is no mathematical or formulaic uh
[54:42] way to get from A to B. It's just a
[54:45] conjecture. Um I can you can take a lot
[54:48] of acid and go into a float tank for all
[54:50] Popper cares. Um come up with some idea.
[54:53] Um but what
[54:55] but Popper did and what he has his um
[54:58] his move here was that everyone at the
[55:01] time was trying to figure out like a
[55:04] reliable way to go from A to B always.
[55:07] And Popper said I don't care. All I care
[55:11] about is once you have B, how can we see
[55:12] if B is correct or if it's not correct?
[55:15] So, his whole framing is on error
[55:18] correction and it's not on a specific
[55:22] method that will reliably produce truth,
[55:25] right?
[55:26] [snorts]
[55:26] Um and so, once you make that reframing
[55:28] then all of a sudden a whole bunch of um
[55:31] moves open up to you, which is one of
[55:34] them is that
[55:35] however you get your idea, I don't
[55:36] really care as long as you can present
[55:38] your idea in such a way that it's clear
[55:40] to the person you're speaking to what
[55:42] you're trying to say, so they can
[55:43] understand it, and then they can
[55:45] criticize it. Um,
[55:46] so kind of the the the moves that are
[55:48] are um
[55:50] looked down upon by by Popper would be
[55:53] to have an idea that is so vague such
[55:56] that the person who you're explaining it
[55:58] to can't understand what you're saying
[55:59] and then can't even criticize it. Um,
[56:02] so if you reframe
[56:04] the way that we get from A to B as being
[56:08] less important than figuring out how to
[56:11] remove the errors from B,
[56:14] um, then you can start making a huge
[56:16] amount of progress um
[56:18] in all sorts of different domains. So,
[56:21] to go to stand-up comedy for instance,
[56:23] um
[56:25] you could use stand-up comedy as an
[56:26] experiment and this the experiment is
[56:27] quite simple. Are people going to laugh
[56:29] when I say this thing? If they don't
[56:31] laugh, the experiment has failed. If
[56:33] they do laugh, then experiment has
[56:35] uh not been proven to be always funny.
[56:38] Um, it it it might not be funny in
[56:39] another context, but at least
[56:41] um you've attempted to falsify your view
[56:45] or criticize your view of the humorous
[56:48] nature of your your um joke.
[56:50] Um, the Bayesian epistemologists would
[56:52] instead think of it as, "Okay,
[56:55] given um all the information in a
[56:57] person's head, how can you reliably
[56:59] produce funny jokes over and over and
[57:00] over and over and over again?" Like, how
[57:02] can we come up with a a process to
[57:04] produce that funny joke? Um,
[57:06] and that's I think uh the one of the
[57:08] central reframings um that uh Popper's
[57:11] work um
[57:12] uh provides the the the the reader,
[57:15] which is that it doesn't matter. Doesn't
[57:17] matter how you get from A to B. Like,
[57:18] there obviously are differences between
[57:19] deductive logic and um uh conjectural um
[57:24] um
[57:24] inferences, but at the end of the day
[57:26] all that matters is
[57:28] uh if you can criticize, if you can
[57:29] critique, if you can falsify, if you can
[57:30] error correct, if you can um uh remove
[57:34] the mistakes with the new piece of
[57:36] information. And so,
[57:41] Popper sometimes is described as either
[57:43] a fallibilist, i.e. the idea that all
[57:45] people are always fallible, or a
[57:48] negativist,
[57:49] the idea that all that matters is the
[57:51] elimination of error,
[57:53] contrasted against the positivists who
[57:56] are positively trying to come up with a
[57:57] formula, a mechanism,
[58:00] an AI,
[58:01] a training procedure that will reliably
[58:03] produce new new knowledge. And so,
[58:05] that's the the difference.
[58:06] Yeah. Okay. Okay. Super interesting.
[58:09] So,
[58:10] yeah, for the stand-up example, that's
[58:12] interesting.
[58:14] The problem is that
[58:16] so, the main the main issue I would have
[58:18] with that is
[58:20] I mean, the main question, it's not an
[58:21] issue. My main question is feasibility
[58:24] in the sense that
[58:26] it's not that easy to run that
[58:27] experiment when you do stand-up, like if
[58:30] you're on stage and you're not a known
[58:33] comedian,
[58:34] you go to open mics. Open mics are very
[58:38] weird sample of people who go there in
[58:41] the audience. And the audience is very
[58:43] variable.
[58:44] So,
[58:45] someone who is very well known, like
[58:47] let's say Ricky Gervais, if he goes to
[58:49] an open mic to test
[58:52] to test material,
[58:55] it's somewhat random because usually
[58:58] they don't say Ricky Gervais is going to
[59:00] be there.
[59:01] But, he knows his audience and he
[59:03] already knows
[59:06] what
[59:07] people who finds his joke funny are, who
[59:11] people
[59:12] who find his joke funny are.
[59:15] The problem, if you're a completely
[59:17] anonymous stand-up comedian who just
[59:19] does random open mics, is that you
[59:22] cannot really repeat the experiment
[59:24] because
[59:26] you don't have the same people in front
[59:28] of you or at least the same
[59:31] the
[59:31] the same audience with somewhat similar
[59:35] audience characteristics.
[59:37] So, you kind of repeat the experiment,
[59:40] but not really. Like, the only thing
[59:42] that doesn't change is the joke.
[59:44] But, the audience changes and it really
[59:46] depends on that. So,
[59:48] yeah, here I'm I'm a bit
[59:50] like, my question is, okay, like, I see
[59:53] what you what Popper mean Popper means
[59:55] on paper. I'm not sure how feasible it
[59:58] is in all cases.
[01:00:00] And a second question I have is
[01:00:03] okay,
[01:00:04] so, it's critical rationalism. I think
[01:00:06] we have covered the critical part.
[01:00:08] And I think the rational part is very
[01:00:10] important, too.
[01:00:11] And so, what makes
[01:00:13] a
[01:00:14] critique rational? Because I'm guessing
[01:00:16] this is extremely important to Popper's
[01:00:18] framework knowing
[01:00:20] um
[01:00:21] the rest of I mean, some of the rest of
[01:00:24] his proposition.
[01:00:25] Mhm.
[01:00:26] Uh second question is easier and then
[01:00:28] let's do the first question after that.
[01:00:31] Uh what makes a person rational is the
[01:00:33] willingness to listen to criticism.
[01:00:36] Uh that's it. It's
[01:00:37] criticism is not easy to receive. I
[01:00:40] don't like getting it. No one likes
[01:00:41] getting it. Um
[01:00:43] but the the rational mind is the one who
[01:00:47] will
[01:00:48] um earnestly listen to
[01:00:51] someone tearing their life's work apart
[01:00:53] and say thank you afterwards. Um
[01:00:56] and that's it. So, rationality is is in
[01:01:00] Popper's world view um and I would say
[01:01:03] in my world view, too, in the sense that
[01:01:04] I'm extremely influenced by him. Um
[01:01:06] is is simply the willingness to receive
[01:01:08] criticism lovingly and um in a spirit of
[01:01:12] appreciation. Um
[01:01:14] and so, someone uh
[01:01:16] I think Popper's um
[01:01:18] was able to distill his entire
[01:01:20] philosophy into one simple motto, which
[01:01:21] is
[01:01:23] "I may be wrong and you may be right,
[01:01:25] but together we'll get closer to truth."
[01:01:27] Um and that's it. And so, it's all well
[01:01:31] and good to talk about um so now to move
[01:01:33] here
[01:01:34] your first example, which is that um
[01:01:38] [snorts]
[01:01:38] if Ricky Gervais goes to a
[01:01:41] show of his, he's getting a biased
[01:01:43] sample, right? That was I think part of
[01:01:45] part partly what you're saying. Um and
[01:01:48] how can this be a good experiment um if
[01:01:50] he has a biased sample the people who
[01:01:52] are listening to him
[01:01:53] are already biased to like his work and
[01:01:56] and so, it's not really a great
[01:01:58] experiment. Um I would say 100% agree
[01:02:01] with you, not a great experiment from
[01:02:03] the standards of rigor that say a
[01:02:07] um
[01:02:08] microbiologist would have to undergo in
[01:02:11] order to determine if their drug works
[01:02:14] for the whole population. Of course, for
[01:02:15] that kind of problem, you absolutely
[01:02:17] need much more rigor than a biased
[01:02:19] sample. But
[01:02:21] if you reframe Gervais and the stand-up
[01:02:24] comedian is not
[01:02:26] running a thorough unbiased AB test as
[01:02:29] like meta would have to do or
[01:02:31] or anyone doing drug discovery would
[01:02:32] have to do, but just getting some more
[01:02:34] criticism. Um then all of a sudden
[01:02:38] the biased sample
[01:02:41] doesn't matter too much. Um what matters
[01:02:43] is that it's just yet another round of
[01:02:46] criticism and so
[01:02:47] um maybe to fill it out a little bit. So
[01:02:50] you were uh
[01:02:52] bringing up this idea that it's not very
[01:02:54] feasible in practice because the
[01:02:56] stand-up comedian can't randomly sample
[01:02:58] every person from every
[01:03:01] demographic and every geographic
[01:03:03] location.
[01:03:04] Um
[01:03:05] Absolutely, they can't. Uh but what they
[01:03:07] can do is set up their life so that
[01:03:10] every conversation they have, they're
[01:03:12] doing a little bit of trial and error.
[01:03:13] They're trying out a new joke here and
[01:03:14] there and they're saying, "Okay, that
[01:03:15] one landed a little bit. Oh, didn't
[01:03:17] land. Okay, so I just tried to make my
[01:03:19] 4-year-old kid laugh. Didn't work. Tried
[01:03:21] it on my wife. That didn't work either.
[01:03:23] Oh, but this one made both of them
[01:03:24] laugh. Okay, sweet. Now maybe I'm going
[01:03:25] to take this joke and I'm going to try
[01:03:26] it at a few
[01:03:28] um like off-Broadway
[01:03:30] comedy sellers like like the the Comedy
[01:03:33] Cellar. Um
[01:03:34] and this is in fact how these comedians
[01:03:36] do it. They they are constantly
[01:03:38] workshopping their jokes. Um and then
[01:03:40] they'll go to a few small places where
[01:03:41] they're not really well known. And then
[01:03:43] over about a year of this they'll
[01:03:45] whittle down 10 hours of material into
[01:03:48] like a tight 45 minutes. Um
[01:03:51] and that whole year is a whole paparian
[01:03:55] year of subjecting themselves to um
[01:03:59] criticism from wherever it comes. And so
[01:04:03] um thinking about experiments as a kind
[01:04:05] of criticism
[01:04:07] rather than criticism as a kind of
[01:04:08] experiment, which I think is what how
[01:04:10] you were thinking about it, um is I
[01:04:12] think very important because when it
[01:04:15] comes to criticism
[01:04:16] um
[01:04:17] doesn't really matter
[01:04:19] if your sample sizes are biased as long
[01:04:22] as you have a lot of criticism. Um and
[01:04:24] just the more of it you get, the better
[01:04:26] your work will become.
[01:04:28] Of [snorts] course, um nothing I'm
[01:04:30] saying would apply to to drug discovery
[01:04:32] because there one kind of criticism is
[01:04:34] like
[01:04:35] uh you can't just test this drug on like
[01:04:37] your nephew and your wife because people
[01:04:39] will kill you'll get people killed. And
[01:04:41] so one of the kinds of criticism there
[01:04:42] is that you absolutely need to be
[01:04:44] thinking about like unbiased samples and
[01:04:46] having large sample sizes and stuff
[01:04:47] because the stakes are so much higher,
[01:04:49] right? Um and and so uh so that that was
[01:04:52] like one of the big aha moments um
[01:04:55] for me because I've always loved
[01:04:57] science. I've been like a science nerd
[01:04:59] from from day one. And like I remember
[01:05:02] learning about like the scientific
[01:05:04] method in like I don't know grade five
[01:05:06] or grade six and that little like
[01:05:07] flowchart thing where you have like a
[01:05:09] hypothesis and then you do a little
[01:05:11] experiment and then you analyze your
[01:05:13] results, and then it comes up with a new
[01:05:14] hypothesis. Um
[01:05:15] that is like the
[01:05:18] uh
[01:05:19] the same level of resolution as like the
[01:05:22] Bohr model of an atom, where atoms are
[01:05:23] like electrons are floating around like
[01:05:25] this. Um obviously it doesn't actually
[01:05:27] work like a solar system. You have like
[01:05:29] a quantum cloud. Um and science doesn't
[01:05:32] actually work in this little circle. Um
[01:05:33] science works by criticism first and
[01:05:35] foremost, where experimental testing is
[01:05:38] just one mode of criticism. But but the
[01:05:40] reframing
[01:05:42] um that I found so valuable was just
[01:05:43] this like oh [ __ ] I need to set up my
[01:05:46] whole life to be able to receive
[01:05:48] criticism from as many people as I can,
[01:05:50] and that means like not being a jerk,
[01:05:53] because if you're being a jerk, then
[01:05:54] people aren't going to want to talk to
[01:05:55] you. And if they're not going to talk to
[01:05:56] you, then you're not going to get
[01:05:58] criticism very well. Or setting up a
[01:06:00] podcast or in a Discord environment such
[01:06:02] that everyone feels comfortable like
[01:06:04] tearing each other's ideas apart in a
[01:06:06] constructive way, right?
[01:06:08] Mhm.
[01:06:08] Um and so how you set up your life and
[01:06:10] how you tune these dials such that you
[01:06:13] don't have so much criticism that
[01:06:14] everything is is destroyed and the
[01:06:15] community is broken and and your life
[01:06:17] just sucks because
[01:06:19] everything is not working very well um
[01:06:21] to the opposite, which is even worse,
[01:06:23] which is
[01:06:24] like a Trumpian figure, where no one can
[01:06:26] criticize him because he's gotten rid of
[01:06:27] every possible person in his orbit who
[01:06:29] would ever be offer a critical ear. And
[01:06:32] so that's the kind of the the the
[01:06:33] reframing, which I think is extremely
[01:06:34] important.
[01:06:35] Mhm.
[01:06:36] Yeah. Completely agree. This is this
[01:06:39] refacing thing.
[01:06:41] Really love it, and I think I
[01:06:43] I got much closer to to what you want to
[01:06:46] mean. Um so a standard example was was
[01:06:49] really really helpful, because like this
[01:06:51] is close this is literally what I do
[01:06:53] actually what you describe is that
[01:06:56] I go to open mics, but I also workshop
[01:06:58] jokes on friends. They just don't know
[01:07:01] it.
[01:07:01] Exactly.
[01:07:02] Yeah.
[01:07:03] And then you and you're looking at their
[01:07:04] facial expressions, right? And you're
[01:07:05] like oh, that one didn't land. That's
[01:07:07] critical, right? Um you're definitely
[01:07:09] not intending it. It's just it's it's
[01:07:10] Like another way is like feedback, just
[01:07:12] some sort of information signal from the
[01:07:14] world. And note how this is
[01:07:17] intrinsically um
[01:07:19] an
[01:07:20] antithetical to the Bayesian
[01:07:22] epistemologist because they
[01:07:24] think that the most important thing is
[01:07:26] making sure your beliefs are well
[01:07:28] calibrated, right? Your priors are well
[01:07:30] calibrated.
[01:07:31] Mhm.
[01:07:32] This is This is a intrinsically um
[01:07:35] inward
[01:07:37] view of epistemology. It's not outward
[01:07:40] view. Um it's not a let's collect
[01:07:42] feedback from the world. It's
[01:07:44] let me go away into my study for 16
[01:07:46] hours and
[01:07:48] think really, really hard about all the
[01:07:49] evidence I've seen last 15 years and
[01:07:51] then write a little probability on a
[01:07:53] piece of paper and then publish a book
[01:07:54] based on that. Um it's it's not about
[01:07:57] feedback from the world. It's it's again
[01:07:59] epistemology, not statistics. Yeah.
[01:08:00] Yeah, yeah, yeah. Yeah, I I have to say
[01:08:02] I don't care too much about Bayesian
[01:08:04] epistemology.
[01:08:05] [laughter]
[01:08:05] Yeah, totally.
[01:08:06] Because like here's my goal my goal is
[01:08:08] usually not to calibrate my beliefs, but
[01:08:11] to update them and to see if I need to
[01:08:13] update them. And so basically having an
[01:08:16] evolving mental model.
[01:08:18] That's like basically what I care about
[01:08:21] on whatever thing I do. So
[01:08:24] and to evolve that mental model, you
[01:08:27] have to test it and do small experiments
[01:08:30] as controlled as you can, as we were
[01:08:32] talking about, but sometimes you can't.
[01:08:35] And also jokes, for instance, are
[01:08:36] extremely
[01:08:38] weird objects because it also it not
[01:08:40] only depends on the receiver, it also
[01:08:42] depends on
[01:08:44] the person making the joke, what he
[01:08:46] looks like,
[01:08:48] uh how he says the joke, where he says
[01:08:51] it, at what time. So it's like it's it's
[01:08:53] very it's a weird object, but you know,
[01:08:56] like you can you can make progress
[01:08:57] anyways. It's just it's going to take a
[01:08:59] bit more feedback loops
[01:09:01] um than than a controlled experiments.
[01:09:03] Um but
[01:09:05] yeah, so and I think so
[01:09:08] I think it's actually very close to
[01:09:10] patient statistics what what you're
[01:09:12] saying because the whole idea is we
[01:09:15] start from somewhere. It's a bit like um
[01:09:17] um MCMC sampling, right? We start from
[01:09:20] somewhere in the space in the posterior
[01:09:22] space, but as iterations go, we get
[01:09:25] feedback and we get much closer to
[01:09:28] what the posterior distribution actually
[01:09:30] is. Um
[01:09:32] so, that's how I
[01:09:33] I think about that and and review it in
[01:09:36] my head in a very simplified way. Um
[01:09:40] but, yeah. Like, Popper is really is
[01:09:43] really a fascinating author. I I got
[01:09:45] introduced to him mainly through
[01:09:47] falsification and
[01:09:49] basically the foundation of
[01:09:51] epistemology, right? Especially through
[01:09:54] um as you mentioned earlier,
[01:09:57] pseudoscience and basically trying to
[01:09:59] understand
[01:10:01] how you can um let's say handle people
[01:10:06] who come with this kind of of claims.
[01:10:09] Um and because it's so
[01:10:12] so intuitive. Like, the problem of
[01:10:14] pseudoscience is that they often have
[01:10:16] extremely
[01:10:18] good and intuitive thesis. It's like,
[01:10:21] that's why it's so attractive to most of
[01:10:25] you. You're like, "Oh, yes, right. Of
[01:10:28] course that that would work like that."
[01:10:30] It I mean, it makes sense.
[01:10:32] And
[01:10:33] So, yeah.
[01:10:34] just going to say, and and note how much
[01:10:35] um evidence they have for their
[01:10:38] hypothesis when evidence from the
[01:10:42] the pseudoscientists or from the
[01:10:43] Bayesian physiologist
[01:10:44] They have a lot.
[01:10:45] Uh
[01:10:46] it's it's an infinite amount. You can
[01:10:48] always come up with more examples of why
[01:10:50] your theory is correct. And to the
[01:10:53] extent that um Bayesian like Bayes'
[01:10:55] theorem encourages Bayesian philosophy,
[01:10:58] which then encourages Bayesian
[01:10:59] epistemology which then encourages
[01:11:00] people to just act this way in practice.
[01:11:02] This is the exact antithesis of of um
[01:11:05] the the the critical um mindset where
[01:11:08] Mhm.
[01:11:09] Um so it I guess it's just an
[01:11:10] interesting example of how some dusty
[01:11:13] old equations written 400 years ago can
[01:11:17] turn into um the practices and behaviors
[01:11:20] that I claim lead to pseudoscience and
[01:11:23] to um
[01:11:25] um conspiracy theories.
[01:11:26] Mhm. Mhm. Yeah, I don't So, I'm going to
[01:11:29] butcher the example
[01:11:31] but the
[01:11:32] but the the point stands. That's why I'm
[01:11:34] talking about it but
[01:11:36] I don't remember if it's Popper or if
[01:11:38] it's
[01:11:40] maybe Carl Sagan, maybe Sagan who took
[01:11:42] that example of like to show you that
[01:11:45] you cannot prove a negative.
[01:11:48] Like uh taking the example of you cannot
[01:11:51] prove to me that there is that there
[01:11:54] doesn't exist a giant teapot orbiting
[01:11:58] I think it was Neptune or whatever
[01:12:00] planet. Uh so yeah, you seem to be
[01:12:02] knowing that example so please take it
[01:12:04] away because you're going to make it uh
[01:12:06] much more justice than I I will.
[01:12:09] Yeah. Um
[01:12:11] So, uh that's Bertrand Russell's teapot
[01:12:14] famous example.
[01:12:16] Russell.
[01:12:17] And I believe he was um
[01:12:20] bringing that up in relation to his um
[01:12:24] atheism. I'm very much an atheist as
[01:12:26] well. And uh
[01:12:28] the
[01:12:29] arguments that he was arguing against
[01:12:31] were well, prove to me, Bertrand, birdie
[01:12:34] boy, that God doesn't exist. Prove that
[01:12:37] this thing isn't there. And he basically
[01:12:40] said, well, you you can't prove a
[01:12:41] negative. You can't prove the
[01:12:42] non-existence of something. Um
[01:12:45] And uh and so that's related to um
[01:12:49] um critical rationalism in in
[01:12:51] a a number of ways. One is is taking it
[01:12:54] in this positive
[01:12:56] framing is that many ideas, many of
[01:12:59] which are good and valid, are not
[01:13:02] falsifiable. You can't prove them one
[01:13:04] way or or the other.
[01:13:06] Um
[01:13:06] in a more negative framing, it's this
[01:13:09] idea that if someone
[01:13:13] doesn't want to hear criticism, if
[01:13:16] someone puts forth challenges that are
[01:13:18] impossible to solve, then you can never
[01:13:20] puncture through the dogma.
[01:13:22] And so in [clears throat] that in that
[01:13:23] example, I think he he kind of
[01:13:25] highlights two nice Popparian concepts,
[01:13:27] even though
[01:13:28] Bertrand Russell himself was
[01:13:30] inductivist. And so he was someone that
[01:13:32] Popper argued about or argued with a
[01:13:34] bunch on on those that regard. But yeah.
[01:13:38] Um actually another thing with Bertrand
[01:13:39] Russell is this interesting. Um so uh
[01:13:42] Russell himself
[01:13:44] I think is a beautiful example of the
[01:13:46] critical mindset because him and
[01:13:50] Alfred North Whitehead, I believe his
[01:13:52] name is, um came up with
[01:13:55] this like
[01:13:56] 1,000-page tome that was trying to
[01:14:00] um
[01:14:01] put all of mathematics
[01:14:03] on a firm foundation of set theory. And
[01:14:06] so they
[01:14:07] put hundreds and hundreds of hours into
[01:14:10] trying to
[01:14:12] prove that all mathematics can be
[01:14:14] derived from a small set of axioms. Um
[01:14:17] and then this annoying guy named Kurt
[01:14:19] Gödel showed that their entire life's
[01:14:21] work was impossible. Um and what they
[01:14:24] did was say thank you. That's it. You've
[01:14:28] just answered my question. They didn't
[01:14:31] start attacking Kurt Gödel, like imagine
[01:14:32] if they were on Twitter these days. But
[01:14:34] it's it's a beautiful example of um the
[01:14:39] falsif- the fallible attitude and this
[01:14:41] idea that
[01:14:43] you should be grateful when someone is
[01:14:45] kind enough to spend time to think about
[01:14:47] your ideas and offer you criticism of
[01:14:49] them. Um Um and yeah, and their
[01:14:52] jettisoning of 1,000 pages of work given
[01:14:55] a proof that showed they were wrong is I
[01:14:57] think a really nice example of that.
[01:14:59] Yes. [snorts] Yeah, yeah, for sure. Uh
[01:15:01] and actually something that reminds me
[01:15:03] from what you were saying about the
[01:15:05] rationality
[01:15:07] of the criticism
[01:15:09] defined by Popper.
[01:15:11] If I understood correctly, that's only
[01:15:14] on
[01:15:16] the eyes of the criticized, right? Not
[01:15:20] the criticizee. So, basically
[01:15:24] it becomes rational just because I take
[01:15:26] it rationally if somebody criticize me
[01:15:29] like criticizes one of my joke for
[01:15:30] instance.
[01:15:31] And
[01:15:33] not because
[01:15:37] the critique is not rational. Is that
[01:15:41] correct?
[01:15:42] It is correct. Yeah, you you
[01:15:45] as much as I would like to be able to uh
[01:15:47] control the dogmatism of other people.
[01:15:50] As much as I try, I can't do that
[01:15:52] unfortunately.
[01:15:53] All I can do is control the way that I
[01:15:55] act and I receive criticism. Um
[01:15:57] and sometimes when you're speaking with
[01:16:00] a fine gentleman like yourself, both
[01:16:02] participants are just naturally trying
[01:16:03] to get towards the truth and
[01:16:05] um and all criticism is offered in that
[01:16:07] in that spirit.
[01:16:09] Um other times, the interlocutor is not
[01:16:11] like that and they just hate you. Or
[01:16:14] I was [snorts] on uh the Doom Debates
[01:16:15] podcast, which if any listeners want to
[01:16:18] get a good example of what Bayesian
[01:16:19] epistemology looks like these days in
[01:16:21] practice, I would listen to uh Doom
[01:16:23] Debates.
[01:16:24] Um
[01:16:25] I was a bit of a jerk on that podcast. I
[01:16:27] was in a bad bit of a bad mood and there
[01:16:29] was 300 uh YouTube commenters who were
[01:16:32] all tearing into me.
[01:16:33] Um and it hurt. It definitely hurt cuz I
[01:16:36] thought I did well on that podcast, but
[01:16:37] my attitude was kind of just a bit of a
[01:16:39] bad attitude. Um and so, yeah, it would
[01:16:41] be so nice if I could shout at each of
[01:16:43] these commenters and tell them to be
[01:16:45] more rational, but I'll take criticism
[01:16:47] where I can get it. Um, and even if it
[01:16:49] was a bit harder to, uh, to hear, it was
[01:16:51] still extremely valuable for me, um, and
[01:16:53] was one of the episodes that, uh, I
[01:16:55] learned from the most, um, because of
[01:16:58] how much that particular criticism hurt,
[01:17:00] but again, criticism in any form is is
[01:17:02] is valuable, and so, uh, so I definitely
[01:17:04] adopted Yeah. I changed a lot of things
[01:17:07] based on that, but, um, but yeah,
[01:17:08] unfortunately you can't control other
[01:17:09] people. You can just control yourself.
[01:17:10] And I think, um,
[01:17:12] realizing that criticism is like the
[01:17:14] most valuable thing that you could
[01:17:15] possibly receive from anyone in any
[01:17:16] form, um, I think is is such a powerful,
[01:17:19] uh, reframing, and it, um, it's been,
[01:17:21] uh, extremely, um, important in my own
[01:17:23] life. Yeah.
[01:17:24] Definitely. Yeah, I mean couldn't agree
[01:17:26] more. And so, I'm obviously asking you
[01:17:28] that because of, yeah, the like social
[01:17:31] media, basically, and the tendency of
[01:17:33] people to be extremely
[01:17:35] uh, radicalized on there.
[01:17:38] And since like, yeah, but what do you do
[01:17:40] if somebody's just giving you a critique
[01:17:43] which is ad hominem, you know, like they
[01:17:45] have nothing interesting to tell you,
[01:17:47] and they just attack you instead of
[01:17:49] attacking the idea. Um, sure. So, in
[01:17:52] the, you know,
[01:17:54] the scientist in me wants to find a
[01:17:56] framework to handle that and just like,
[01:17:58] you know, have an answer to that. But
[01:18:00] then, I think
[01:18:02] Popper's answer to it is actually
[01:18:05] much better and much more useful. And
[01:18:07] actually, I think here probably the
[01:18:09] Stoics had that before Popper. I see a
[01:18:12] lot of resemblance with Stoic philosophy
[01:18:14] here, like Epictetus, where it's like,
[01:18:17] an insult just becomes an insult because
[01:18:20] you're interpreting it that way. Um, but
[01:18:23] if you don't take the ins- what's meant
[01:18:26] as an insult as an insult, it is not an
[01:18:29] insult.
[01:18:30] And actually, it's going to be much
[01:18:31] better because it's going to make the
[01:18:32] insulter, uh, much, much more, uh, angry
[01:18:35] at you because, uh,
[01:18:37] they'll see they basically cannot
[01:18:40] cannot get at you.
[01:18:41] Um and
[01:18:43] and also the idea in Stoic philosophy
[01:18:45] that well, actually, you know, um the
[01:18:47] obstacle is the way. So, um
[01:18:50] the critique you got is actually always
[01:18:53] useful because it's a critique, but it
[01:18:55] it's not binary. The usefulness of a
[01:18:57] critique is not binary. It's a spectrum,
[01:18:59] and maybe it's at zero if it's just an
[01:19:01] ad hominem attack. Uh maybe it's at 1.5%
[01:19:05] where it's like, "Okay, most of it is
[01:19:07] garbage. Oh, but here there is one part
[01:19:10] that's actually that's interesting. Um I
[01:19:12] could actually get better with that."
[01:19:14] Wherever whatever the source or the form
[01:19:18] of the criticism was. And I think that
[01:19:20] that goes directly into into Popper's uh
[01:19:23] philosophy, as as you were saying.
[01:19:25] Yeah, Stoicism um and critical
[01:19:28] rationalism or fallibilism
[01:19:31] pair beautifully together, absolutely.
[01:19:33] Um Stoicism is less about like where
[01:19:36] does knowledge come from and more about
[01:19:37] just how do you live the good life? How
[01:19:39] do you do with less? How are you okay
[01:19:42] Um if you were to lose your house
[01:19:44] tomorrow, are you going to be okay? Um
[01:19:47] and I think the kind of equanimity that
[01:19:49] Stoicism can teach um allows you to
[01:19:52] receive criticism much better. Um my
[01:19:54] favorite line of from one of the Stoic
[01:19:57] philosophers um is something like uh
[01:20:00] Stoicism teach I got this from a a
[01:20:02] lecture online whose name I'm blanking
[01:20:03] on at the moment. I can try to add the
[01:20:05] show notes. Uh but at the end of the
[01:20:06] lecture he said uh Stoic the Stoics
[01:20:09] teach us that every man will die, but
[01:20:11] not every man will die whining.
[01:20:13] And I just [laughter]
[01:20:15] I love that.
[01:20:16] Yeah.
[01:20:16] Um it's awesome.
[01:20:18] yeah, every man will be criticized, but
[01:20:19] not every man will be criticized and
[01:20:21] whine about it afterwards, right?
[01:20:22] Um yes.
[01:20:23] And uh and
[01:20:25] And yeah, and so um
[01:20:27] yeah, with regards to like ad hominems
[01:20:29] and um and just when people attack you,
[01:20:32] um
[01:20:33] of course, nothing prevents the critical
[01:20:35] rationalist from criticizing the
[01:20:36] criticism.
[01:20:37] Mhm.
[01:20:38] ad infinitum. And like [snorts] like
[01:20:40] where does the phrase like critical
[01:20:41] thinking come from? Well, it's a
[01:20:42] derivative of of Popper's work, right?
[01:20:45] Um
[01:20:46] but
[01:20:48] yeah, so sometimes the criticism you
[01:20:49] receive just is low quality and they're
[01:20:52] just attacking you as
[01:20:54] you as a person. Um so often that I will
[01:20:57] tend to ignore or
[01:20:59] like
[01:21:00] I guess one danger of talking about
[01:21:01] critical rationalism
[01:21:03] um and trap that I've arguably fallen
[01:21:05] into in this conversation is like
[01:21:08] deifying criticism and pretending like
[01:21:10] it's this otherworldly thing that never
[01:21:12] hurts. But no, of course it's going to
[01:21:14] hurt. Like the the deeper the criticism
[01:21:16] is, like sometimes it really hurts. Um
[01:21:19] and so especially when you're in like a
[01:21:21] extremely critical debate with somebody
[01:21:23] um and emotions are high and tempers are
[01:21:26] are are overflowing and this has
[01:21:28] happened to me many times on on the
[01:21:30] podcast. Um listeners of Increments will
[01:21:33] will know what I'm referring to. Um
[01:21:35] but uh uh
[01:21:37] but yeah, a little bit of stoicism added
[01:21:39] on top and just a little bit of a
[01:21:40] realization that everyone is human and
[01:21:42] that like even if someone has given you
[01:21:44] a big paragraph online and three
[01:21:46] quarters of it is ad hominems, but maybe
[01:21:48] like one quarter of it is actually valid
[01:21:51] then you the person receiving it can
[01:21:54] focus their conversation on the germ of
[01:21:56] truth
[01:21:57] um and get them to sharpen their own
[01:21:58] criticism of you um if you are able to
[01:22:02] control your emotions um and I say that
[01:22:04] as someone who is extremely not able to
[01:22:07] control my emotions sometimes. But uh
[01:22:08] it's a lifelong project. But uh but
[01:22:10] yeah. And then so um so basically in a
[01:22:12] nutshell Popper's philosophy is
[01:22:16] criticism is at the foundation.
[01:22:18] Everything you think about the
[01:22:19] scientific method comes out of that.
[01:22:21] Peer review
[01:22:22] posting your like your paper on Twitter.
[01:22:25] Like I remember I was much more scared
[01:22:28] to post a paper on Twitter than I was to
[01:22:29] submit it to peer review because I knew
[01:22:31] that Twitter was going to give me way
[01:22:32] harsher criticism. And so, like, you
[01:22:35] figure out how to um set up your work
[01:22:38] such that you get like the right amount
[01:22:40] of criticism at the right stages because
[01:22:42] sometimes an idea is so early that,
[01:22:45] like, if I was to explain it to you, you
[01:22:46] would be like, "Ah, I think it does it's
[01:22:48] not going to work at all." And so, you
[01:22:49] kind of want to think about titrating in
[01:22:51] the criticism at the right stages. You
[01:22:53] want to think about, like, okay,
[01:22:56] there's a wall of YouTube commenters
[01:22:57] that are all telling me I'm a terrible
[01:22:59] person. Um but, there's like four or
[01:23:01] five people whose opinions I really do
[01:23:03] respect and I'm going to ask them to see
[01:23:04] what's valid there. And so,
[01:23:07] framing your entire life around how do I
[01:23:10] um take advantage of the fact that the
[01:23:12] world is going to criticize me whether
[01:23:13] or not I like it. And so, I may as well
[01:23:15] harness that that energy in a
[01:23:17] effective way um is the game of figuring
[01:23:20] out how to do science. Um and that's
[01:23:22] that's uh
[01:23:23] um one of the his central insights in
[01:23:25] why he calls his life's work
[01:23:27] critical rationalism.
[01:23:29] Yeah. Yeah, I would I would even argue
[01:23:31] not only do science uh the way right,
[01:23:34] but mostly life.
[01:23:35] It's exactly exactly. Yeah, no, I just
[01:23:38] if I can echo that cuz I don't want to
[01:23:40] give the impression that this is just
[01:23:41] what us dodgy academic-y science people
[01:23:44] do. It's just as much what the the
[01:23:46] roofer does when they're trying to
[01:23:47] figure out why your roof is leaking.
[01:23:49] Just as much what your kid does when
[01:23:51] they're trying to learn to ride a bike
[01:23:52] and then they fall and that's a little
[01:23:54] bit error correction. They should have
[01:23:55] done it differently. Um and so, there is
[01:23:58] no distinction in the Popperian
[01:24:00] worldview between the scientist, the
[01:24:01] human being, the artist, the comedian,
[01:24:03] the UFC fighter. Uh it's all it's all
[01:24:05] just learning from the world is is um
[01:24:07] uh as as best you can.
[01:24:09] Yeah. Yeah, yeah.
[01:24:10] Love it. Love it. Um
[01:24:12] fascinating. Um
[01:24:13] and actually, you know, um
[01:24:16] so, most of my work is trying to take
[01:24:20] the latest state of the art science and
[01:24:22] trying to distill it to
[01:24:24] people who either don't have the time or
[01:24:26] the inclination to
[01:24:28] read the papers or talk to the
[01:24:30] researchers and
[01:24:31] and and it's just yeah, like doing that
[01:24:35] and I think it's very important because
[01:24:37] uh
[01:24:38] papers are fine, but nobody reads them
[01:24:40] only scientists and they just read each
[01:24:43] other, you know, but uh normal people
[01:24:45] don't and so we need people to take the
[01:24:48] science and just push it out there
[01:24:49] because otherwise um most of the
[01:24:52] population is just going to say you stay
[01:24:54] with the old beliefs and that's not what
[01:24:57] you want in the end. You want the
[01:24:58] science to be applied and not only
[01:25:00] written. And
[01:25:02] so that's why also I love a lot stand-up
[01:25:05] comedy. You've got a lot of very good
[01:25:07] stand-up comedians who actually kind of
[01:25:09] do, you know, um
[01:25:11] epistemology like that. I think Ricky
[01:25:14] Gervais is actually a good one. Uh Jimmy
[01:25:16] Carr I think is a good one. They often
[01:25:18] are very rational, so that's interesting
[01:25:20] because I think they have to be very
[01:25:22] analytical.
[01:25:23] And a a very good show for that also
[01:25:26] that I often cite is The Big Bang
[01:25:28] Theory.
[01:25:29] Uh and Sheldon in particular has some
[01:25:32] very good snippets, you know, sometimes
[01:25:34] and and I think it's great like and like
[01:25:37] the
[01:25:38] the two which comes to mind for our
[01:25:41] conversation which is one
[01:25:43] um
[01:25:44] that
[01:25:45] he so Penny
[01:25:47] makes a joke about Nebraska as she's
[01:25:50] from Nebraska and I don't know like
[01:25:52] another state that Nebraska Nebraska
[01:25:55] looks down on
[01:25:56] and she makes that joke in the apartment
[01:25:58] of Sheldon in Pasadena, California and
[01:26:02] nobody laughs. [clears throat]
[01:26:03] And and then she's like, "Oh well, I
[01:26:07] guess that joke is only funny in
[01:26:08] Nebraska."
[01:26:09] And then Sheldon says,
[01:26:11] "How are you going to say that? With the
[01:26:12] data at hand, you can only say that that
[01:26:14] joke is not funny here."
[01:26:17] Which is exactly what you were saying.
[01:26:19] Exactly. You There's no way to
[01:26:21] generalize beyond the local. in a
[01:26:23] reliable
[01:26:24] Yes. And I think that's awesome because
[01:26:26] everybody can understand that.
[01:26:29] And yet it's a very profound concept of
[01:26:33] scientific thinking. And I think this is
[01:26:35] a great distillation of scientific
[01:26:37] principle for everybody.
[01:26:39] Very nice.
[01:26:39] Which is extremely extremely precious.
[01:26:43] Um
[01:26:43] And another one I really love that's
[01:26:45] also related to what we're saying, you
[01:26:47] know, about the with the teapot. And the
[01:26:50] burden of the proof is basically on the
[01:26:52] people saying that God exists.
[01:26:54] is um
[01:26:55] So Sheldon's mother in the show is very
[01:26:57] religious. So she basically tells him at
[01:27:00] some point that uh to
[01:27:03] prove to him that God doesn't exist. And
[01:27:07] then Sheldon is like, "Well,
[01:27:08] imagine that I told you that I believe
[01:27:10] there is a giant invisible man in this
[01:27:14] room that influences your behavior. Um
[01:27:18] the burden of proof wouldn't wouldn't be
[01:27:20] on you to disprove me. It'd be on me to
[01:27:23] prove you that I'm actually right."
[01:27:26] And it's ex- it's uh yeah, another
[01:27:29] example of of of a great way to
[01:27:32] distinguish what we've been talking
[01:27:33] about.
[01:27:33] Exactly. Totally.
[01:27:36] I love it. Yeah. Um
[01:27:37] so actually I still had a lot of
[01:27:39] questions for you, but it's it's getting
[01:27:42] late. Uh so I don't want to I don't want
[01:27:44] to take too much of your time. And um
[01:27:48] since I'm going to come on your show at
[01:27:50] some point, actually I can keep some of
[01:27:53] those questions. And if you want I can I
[01:27:55] can extend to you and it's just going to
[01:27:57] start spark some some interesting
[01:28:00] interesting conversation between us.
[01:28:02] What would you say?
[01:28:04] Yeah, sounds great. Yeah, I'm I'm in no
[01:28:05] rush. So Yeah.
[01:28:07] Awesome. So I let's do that. I'll keep
[01:28:09] these
[01:28:10] these questions on my computer. And then
[01:28:12] when I come in your show I can I can
[01:28:13] actually ask them to you and then we'll
[01:28:15] just talk about that. I have some some
[01:28:18] some good one about a bit more concrete,
[01:28:20] you know, one about a paper by Andrew
[01:28:22] Gelman. So, yeah, just
[01:28:25] interesting stuff.
[01:28:26] Mhm.
[01:28:27] And
[01:28:29] also yeah, and two other questions that
[01:28:31] actually I think you're going to like
[01:28:33] about falsifying your beliefs. So, but
[01:28:36] this is a teaser for you.
[01:28:38] But so before you go
[01:28:39] Come on next time I'll ask you.
[01:28:41] Exactly.
[01:28:43] So, before you go though,
[01:28:45] I do want to talk a bit about your show,
[01:28:48] Increments podcast with Ben Chug. You've
[01:28:51] been running that for a while now. And
[01:28:53] so I can definitely say congratulations
[01:28:56] because it's not an an easy job.
[01:28:59] Um
[01:29:00] what's the elevator pitch? Who's the
[01:29:02] audience? And what made you want to
[01:29:05] start a philosophy of science podcast in
[01:29:07] the first place?
[01:29:08] Yeah.
[01:29:09] [snorts]
[01:29:09] Yeah, what's the elevator pitch? I don't
[01:29:11] know what the elevator pitch is. I'm
[01:29:12] still trying to figure that out. Um
[01:29:13] typically I just say it's applied
[01:29:15] philosophy or it's a philosophy
[01:29:19] and science and history and everything
[01:29:21] in between.
[01:29:22] Um
[01:29:23] the elevator pitch is that so I
[01:29:24] mentioned a little bit about Popper how
[01:29:26] he was writing from in the 20th century.
[01:29:29] And if we just think a little bit about
[01:29:31] all of the things that happened in the
[01:29:33] 20th century. So, we have Marxism, we
[01:29:36] have Freudianism, we have Einstein, we
[01:29:39] have Turing, we have Kolmogorov, we have
[01:29:43] Dawkins and the selfish gene, and we
[01:29:45] have um
[01:29:47] Turing if I already mentioned him. Um so
[01:29:50] to study Popper is to study everything
[01:29:52] because he was in the weeds with all of
[01:29:54] these subjects.
[01:29:56] And so I like to think of epistemology
[01:29:59] as like the master subject, this the
[01:30:01] center of the Venn diagram. Um and so on
[01:30:04] the podcast we are kind of just
[01:30:07] exploring outwards from the center. So,
[01:30:09] uh um we have an ongoing series where
[01:30:11] we're um going through chapter
[01:30:13] uh by chapter of Conjectures and
[01:30:15] Refutations.
[01:30:16] Um
[01:30:17] in between that, we spend a lot of time
[01:30:18] talking about um AI and in particular
[01:30:22] reasons why um
[01:30:24] artificial general intelligence is not
[01:30:26] um going to happen in our lifetimes. Um
[01:30:29] controversial claim. See you next
[01:30:30] episode. Um we spend a lot of time
[01:30:33] talking about effective altruism um in a
[01:30:35] bit of a negative way. Uh effective
[01:30:37] altruism is a strong hotbed for Bayesian
[01:30:40] epistemology and for a newer philosophy
[01:30:43] called long-termism, which I'm quite
[01:30:46] critical of and I can maybe share a few
[01:30:48] blog posts um to put in the show notes
[01:30:51] where where I um
[01:30:52] criticized one of William MacAskill's
[01:30:54] papers and then he ended up taking it
[01:30:55] down and removing all the quotes that I
[01:30:57] had quoted from him. Um we talk about um
[01:31:02] uh open societies. So, we haven't talked
[01:31:03] at all about Popper's political
[01:31:05] philosophy, but um that's a huge
[01:31:08] current. Um
[01:31:09] and then we just have random episodes
[01:31:13] discussing kind of
[01:31:15] current uh cultural trends. So, um
[01:31:19] Jonathan Haidt's work um
[01:31:21] this idea that uh if you give screens to
[01:31:23] your kids, you're going to rot their
[01:31:24] brains. Um my daughter has an iPad and
[01:31:28] she's loving it and it's been the best
[01:31:29] thing ever. So, extremely pro screen
[01:31:31] time, for example. Um we had an episode
[01:31:34] about uh recycling and why recycling
[01:31:36] doesn't actually work very well. Um so,
[01:31:38] we try to uh go from the the extreme
[01:31:43] niche philosophy discussions about what
[01:31:46] truth is, how probability works, what
[01:31:49] certainty means,
[01:31:50] how does logic work, how do you go from
[01:31:52] logic to um to the real world, like
[01:31:54] what's the connection between logic and
[01:31:56] the real world, all the way down to the
[01:31:58] super mundane, like does recycling work?
[01:32:01] Um what's this notion of the patriarchy?
[01:32:03] Is the patriarchy a
[01:32:05] valid concept?
[01:32:07] And just everything in between. And um
[01:32:09] both Ben and I
[01:32:10] um I should just shout out Ben. So Ben's
[01:32:13] uh finishing his PhD at CMU. He's like
[01:32:16] statistician extraordinaire. He's doing
[01:32:18] fundamental work on e-values, which uh
[01:32:21] you should totally have Ben on cuz I
[01:32:22] think your um um your audience, which is
[01:32:25] who's primed for technical discussions,
[01:32:27] would love to to to hear some of the
[01:32:29] stuff that he's working on is um uh
[01:32:31] [snorts]
[01:32:32] Mike Jordan uh for example is on his
[01:32:34] committee. So he's deeply into the the
[01:32:36] ML um community [snorts] there. Um and
[01:32:39] him and I have extremely short attention
[01:32:41] spans. And so we get bored of subjects
[01:32:42] pretty quickly. And so we like to keep
[01:32:45] moving on to new things. And the world
[01:32:48] right now is um
[01:32:50] not boring, shall we say. So there's
[01:32:51] plenty of stuff to talk about. Um and
[01:32:53] those are the kinds of uh subjects we
[01:32:55] touch on. Why did I start the podcast?
[01:32:57] Why do we start the podcast? Uh it was
[01:32:58] started um during COVID when all of the
[01:33:02] sudden I couldn't do my favorite thing
[01:33:03] in the world, which is going to the bar
[01:33:05] and arguing with all my friends about
[01:33:06] politics. Um and so when that stopped, I
[01:33:10] was like, well, this is not acceptable.
[01:33:12] And so Ben and I started our little
[01:33:14] conversations uh gosh, almost 5 years
[01:33:16] ago now. And uh we just had our 100th
[01:33:18] episode a couple weeks ago. And that was
[01:33:21] a nice um accomplishment. And yeah,
[01:33:23] we're looking forward to the next 100. I
[01:33:25] have no idea what the future's going to
[01:33:26] hold, but just chasing our interests as
[01:33:28] they come along.
[01:33:30] Well done. Well done. Yeah. Um
[01:33:32] this is a very interesting show. And I
[01:33:34] didn't know about it before Andreas put
[01:33:36] it in put us in in touch.
[01:33:39] Uh but yeah, I have to say um definitely
[01:33:42] have a new listener here. And this is
[01:33:43] exactly the kind of uh of uh
[01:33:46] nerdy series and entertaining content
[01:33:51] that uh I love, that I try to do here,
[01:33:54] and that uh I hope the world had uh more
[01:33:59] of.
[01:34:00] Definitely thanks a lot for
[01:34:02] doing that. And I'm sure a lot of my
[01:34:06] listeners are are going to enjoy it. So,
[01:34:09] folks, it's going to be in the show
[01:34:10] notes. Um make sure to give a listen to
[01:34:15] to Vaden's and Vaden's, sorry.
[01:34:18] And and Ben's podcast. That's that's the
[01:34:21] French guy in me, you know, wanted to to
[01:34:23] get out because I can never let let
[01:34:25] He mouth when I'm in the US.
[01:34:26] Um
[01:34:27] So, sometimes I have to talk like that
[01:34:30] like a real a real French. Um [snorts]
[01:34:34] [laughter]
[01:34:35] And yeah, so look at that.
[01:34:38] I'm going to be on the show, so it's
[01:34:40] it's the it's the mark of a show that's
[01:34:43] not really high quality with guest
[01:34:45] screening, you know, but still other
[01:34:48] other people who've been on Vaden's
[01:34:50] Vaden's show are are really I vouch for
[01:34:53] them, so.
[01:34:54] Yeah. So, yeah.
[01:34:55] Well, thank you so much.
[01:34:56] No, no, you bet you bet. You're doing a
[01:34:58] lot of good work, so so I'm really happy
[01:35:00] to uh
[01:35:01] to help you guys on that. And
[01:35:05] well,
[01:35:06] let's call it a show, Vaden. I'll come
[01:35:08] on your show and ask you my other
[01:35:10] questions. Before you go though, I have
[01:35:12] two questions ask you
[01:35:15] at the end of the show
[01:35:17] uh
[01:35:18] before you go, you know, I just connect
[01:35:20] to to all my guests. So,
[01:35:24] so that I have distribution of of
[01:35:26] answers, right? Um So, first question,
[01:35:29] what's your favorite thing about
[01:35:31] Bayesian epistemology?
[01:35:34] [laughter]
[01:35:35] I'm kidding.
[01:35:37] It ain't No, I They got to ask the
[01:35:38] question. It encourages people to learn
[01:35:40] more about math. I think math is
[01:35:42] awesome. And if that's a good entry
[01:35:43] point in for some people, then I think
[01:35:45] it's a great way to to to start learning
[01:35:47] about
[01:35:48] math and equations. Yeah.
[01:35:50] Okay. Love it. Love it.
[01:35:53] Um and yeah, that a good answer.
[01:35:56] But that was a joke. The actual first
[01:35:58] question is if you had unlimited time
[01:36:00] and resources, which problem would you
[01:36:02] try to solve?
[01:36:04] Ooh, nice question.
[01:36:06] Um
[01:36:08] I
[01:36:09] If I had unlimited time and resources,
[01:36:12] I think the problem of education is
[01:36:14] really interesting. I think that um
[01:36:17] figuring out how to um
[01:36:19] create environments where people are
[01:36:22] free to explore their own interests and
[01:36:25] explore their curiosity
[01:36:27] while also being realistic about the
[01:36:28] fact that there's a short supply of
[01:36:31] qualified teachers and that parents need
[01:36:34] a place to put their kids while they go
[01:36:36] to work
[01:36:37] is extremely interesting question and
[01:36:39] one that I think if we were to
[01:36:42] make progress [clears throat] on that,
[01:36:43] it would just unlock
[01:36:45] infinite amount of unbounded potential.
[01:36:47] But I think there's a lot of um
[01:36:49] a lot of wasted opportunities um that
[01:36:52] come from
[01:36:54] people not having the freedom to just
[01:36:56] explore their their own interests and
[01:36:58] curiosity
[01:36:59] as much as as I think that everyone
[01:37:01] should be able to. So, maybe that. Yeah.
[01:37:04] Yeah. Yeah. Um very
[01:37:07] very on point answer with with your
[01:37:10] brain, too. I'm not surprised.
[01:37:13] And second question,
[01:37:15] I think I know the answer to that one,
[01:37:17] but we'll see. If you could have dinner
[01:37:19] with any great scientific mind, dead,
[01:37:21] alive, or fictional, who would it be?
[01:37:24] That's a proper easy answer, but I'm not
[01:37:26] going to say proper.
[01:37:28] Okay. Just just to make me wrong.
[01:37:30] Yeah, Richard Feynman because I also
[01:37:33] play music on the side and I heard he's
[01:37:34] an insanely good
[01:37:37] bongo player or was an insanely good
[01:37:38] bongo player.
[01:37:40] I'm attracted to the
[01:37:42] the kind of mind that loves science but
[01:37:45] also loves people and human interaction
[01:37:48] and humor and joy and life and drinking
[01:37:52] and just having a good time.
[01:37:54] Um
[01:37:54] [snorts]
[01:37:54] and I find that that intersection is in
[01:37:57] short supply and so Richard Feynman
[01:38:00] would be definitely somebody I'd love to
[01:38:01] go for dinner with.
[01:38:03] Okay. Yeah. Yeah. Love it.
[01:38:06] Um and yeah, the dinner table is getting
[01:38:10] crowded. We've we've got a few guests
[01:38:12] already. So we'll we'll need to find a a
[01:38:15] bigger restaurant but but let's make it
[01:38:17] happen nonetheless.
[01:38:18] Sounds great.
[01:38:20] Awesome.
[01:38:21] Awesome. Well,
[01:38:22] Vaden, thank you so much for for coming
[01:38:25] to the show.
[01:38:27] There will be a few links
[01:38:29] folks in the in the show notes for that
[01:38:32] episode also a few related episodes.
[01:38:34] Feel free to to check them out. I think
[01:38:36] they're going to be very interesting. I
[01:38:38] hope you enjoyed that episode. That was
[01:38:41] somewhat of a
[01:38:42] of a change of pace uh pace from the
[01:38:46] from the the
[01:38:49] classic technical episodes which I like
[01:38:51] to do from time to time.
[01:38:53] And and I hope you [music] you did too.
[01:38:56] And on that note, Vaden,
[01:38:59] well, I'll I'll see you soon but mainly
[01:39:02] thank [music] you so much for taking the
[01:39:04] time and being on this show.
[01:39:05] Thank you for everything. This is a
[01:39:07] fantastic [music] conversation and had a
[01:39:09] blast and looking forward to part two.
[01:39:12] Happy to hear that.
[01:39:14] [music]
[01:39:15] Excellent.
[01:39:19] This has been another episode of
[01:39:21] learning Bayesian statistics. Be sure to
[01:39:23] rate, [music] review, and follow the
[01:39:25] show on your favorite podcatcher and
[01:39:28] visit learnbasestats.com for more
[01:39:30] resources about [music] today's topics
[01:39:32] as well as access to more episodes to
[01:39:34] help you reach true Bayesian state of
[01:39:37] mind. That's learnbasestats.com.
[01:39:39] [music] Our theme music is good Bayesian
[01:39:41] by Baba Brinkman with MC Lars and Mega
[01:39:44] Ran. Check out his awesome [music] work
[01:39:46] at bababrinkman.com.
[01:39:47] I'm your host Alex Andora. You can
[01:39:50] follow me on Twitter at alex_andora like
[01:39:53] the country.
[01:39:53] [music]
[01:39:54] You can support the show and unlock
[01:39:56] exclusive benefits by visiting
[01:39:58] patreon.com/learnbayesstats.
[01:40:01] Thank you so much for listening and for
[01:40:03] your support. You're truly a
[01:40:04] Good Bayesian, change your predictions
[01:40:07] [music] after taking information. And if
[01:40:09] you think it'll be nothing less than
[01:40:11] amazing,
[01:40:12] [music]
[01:40:12] let's adjust those expectations.
[01:40:16] Let me show you how to be a good
[01:40:18] Bayesian, change calculations after
[01:40:20] taking fresh data. [music]
[01:40:22] Those predictions that your brain is
[01:40:24] making, let's get them on a solid
[01:40:27] foundation.