Learning Bayesian Statistics

Proudly sponsored by PyMC Labs, the Bayesian Consultancy. Book a call, or get in touch!

I love dresses. Not on me, of course — I’m not nearly elegant enough to pull it off. Nevertheless, to me, dresses are one of the most elegant pieces of clothing ever invented.

And I like them even more when they change colors. Well, they don’t really change colors — it’s the way we perceive the colors that can change. You remember that dress that looked black and blue to some people, and white and gold to others? Well that’s exactly what we’ll dive into and explain in this episode.

Why do we literally see the world differently? Why does that even happen beyond our consciousness, most of the time? And cherry on the cake: how on Earth could this be related to… priors?? Yes, as in Bayesian priors!

Pascal Wallisch will shed light on all these topics in this episode. Pascal is a professor of Psychology and Data Science at New York University, where he studies a diverse range of topics including perception, cognitive diversity, the roots of disagreement and psychopathy.

Originally from Germany, Pascal did his undergraduate studies at the Free University of Berlin. He then received his PhD from the University of Chicago, where he studied visual perception.

In addition to scientific articles on psychology and neuroscience, he wrote multiple books on scientific computing and data science. As you’ll hear, Pascal is a wonderful science communicator, so it’s only normal that he also writes for a general audience at Slate or the Creativity Post, and has given public talks at TedX and Think and Drink.

Our theme music is « Good Bayesian », by Baba Brinkman (feat MC Lars and Mega Ran). Check out his awesome work at https://bababrinkman.com/ !

Thank you to my Patrons for making this episode possible!

Yusuke Saito, Avi Bryant, Ero Carrera, Giuliano Cruz, Tim Gasser, James Wade, Tradd Salvo, William Benton, James Ahloy, Robin Taylor, Thomas Wiecki, Chad Scherrer, Nathaniel Neitzke, Zwelithini Tunyiswa, Bertrand Wilden, James Thompson, Stephen Oates, Gian Luca Di Tanna, Jack Wells, Matthew Maldonado, Ian Costley, Ally Salim, Larry Gill, Joshua Duncan, Ian Moran, Paul Oreto, Colin Caprani, Colin Carroll, Nathaniel Burbank, Michael Osthege, Rémi Louf, Clive Edelsten, Henri Wallen, Hugo Botha, Vinh Nguyen, Raul Maldonado, Marcin Elantkowski, Adam C. Smith, Will Kurt, Andrew Moskowitz, Hector Munoz, Marco Gorelli, Simon Kessell, Bradley Rode, Patrick Kelley, Rick Anderson, Casper de Bruin, Philippe Labonde, Michael Hankin, Cameron Smith, Tomáš Frýda, Ryan Wesslen, Andreas Netti, Riley King, Yoshiyuki Hamajima, Sven De Maeyer, Michael DeCrescenzo, Fergal M, Mason Yahr, Naoya Kanai, Steven Rowland, Aubrey Clayton, Jeannine Sue, Omri Har Shemesh, Scott Anthony Robson, David Haas, Robert Yolken, Or Duek, Pavel Dusek, Paul Cox, Trey Causey, Andreas Kröpelin, Raphaël R and Nicolas Rode.

Visit https://www.patreon.com/learnbayesstats to unlock exclusive Bayesian swag 😉

Links from the show:

Abstract

by Christoph Bamberg

In our conversation, Pascal Wallisch, a professor of Psychology and Data Science at New York University, shared about his research on perception, cognitive diversity, the roots of disagreement, and psychopathy. 

Pascal did his undergraduate studies at the Free University of Berlin and then received his PhD from the University of Chicago, where he studied visual perception. Pascal is also a TedX, Think and Drink speaker, and writer for Slate and Creativity Post. 

We discussed Pascal’s origin story, his current work on cognitive diversity, and the importance of priors in perception. 

Pascal used the example of “the Dress” picture that went viral in 2015, where people saw either black and blue or white and gold. He explained how prior experience and knowledge can affect how people perceive colors and motion, and how priors can bias people for action. 

We discussed to what extent the brain might be Bayesian and what functions are probably not so well described in bayesian terms. 

Pascal also discussed how priors can be changed through experience and exposure.

Finally, Pascal emphasized that people have different priors and perspectives, and that understanding these differences is crucial for creating a more diverse and inclusive society.

Automated Transcript

Please note that the following transcript was generated automatically and may therefore contain errors. Feel free to reach out if you’re willing to correct them.

Transcript

0:00

can hear you are you there? Out of battery All right, very good problem on this on my end somehow exceptional Hello. Hello Hello. Yes, I can hear you

0:26

even can you mean the other one?

0:29

The sorry.

0:31

You can hear me in the Zen cast sandcast

0:34

I think so I can hear you now. So let me I'm going to mute you. I'm going to mute the Google meat tab on my end. Okay. See? All right. I see. Okay. Ah, okay. Okay, I see. Okay, that works. Well. It's all good. So I'm gonna mute again Google meet because they hear you twice. You should have some ego on your end, don't you?

1:01

I have what

1:03

do you do you hear me well or

1:06

not? No, no, very well, very well. No echo, no echo. Okay,

1:09

if you have some echo, it's because basically is in cancer. And meet are giving you the sound at the same time. So you need to mute the tab. Not not yourself, but mute the tab on on Google Chrome. I can show you how to do that. Actually. In case you have a problem. One moment just did Can you hear me? Yeah. You just like you see my you see my mouth or no? I do. I do. I do. Yes. Okay, so here if you go to the tab here of Google, you right click, and then you can submit in Spanish but

1:48

I didn't. I didn't. I did. I did. Yes.

1:51

Perfect. Okay, good. All right. Okay, yeah. Thanks for taking the time. Today is just like, crazy stuff happened in Argentina. In all it's beautiful. So basically, I had this internal problem that was the whole block, but now it's back. Right and if you can hear but they decided today to do some workings in the staircase. Now building.

2:21

Now every every is good. Every setback. So let me ask you this. When do I turn on Audacity?

2:28

Yes, I will tell you so you didn't have any issues with Udacity?

2:33

No, I You made a very clear how to video. The only thing the only thing I was honest I thought I installed about it must have been on a different laptop. So I had to reinstall it.

2:44

Oh, yeah. Okay, so that's awesome. So you're already in Zen caster? That's perfect. So thank you, sir. You don't have to do anything. It's just me. Yeah, it's just a backup normally I use with SVG. But sometimes technology and then the otter bots you seemed to know radio button so we're good. Sometimes they tell people because it freaks them out to see about

3:12

Okay, now it's I understand i i know what it is. I sometimes use it to not not for this but for transcripts.

3:20

Exactly. For transcripts. It's it's three Greek. It's awesome. Okay. Cool. So do you have any questions on the questions?

3:31

I think they're good questions. Um, actually, no, it's great. It's really good.

3:38

Cool. Yeah. Well, thanks for taking the time in and like Greece. super looking forward to that. But as you can see, I had like a ton of questions.

3:47

Yes, no, they're great questions. Good questions.

3:51

Cool. So we can start where are you right now? You're in New York.

3:55

I'm in one of the suburbs of New York. Okay. No, no, no, everything's good.

4:03

Okay, awesome. Well, if you're ready, you can start to their city. Okay. Okay, I did okay. And I'm starting also Zen caster. So we are. Great. Guys, Kalish. Welcome to Learning Bayesian statistics.

4:23

Thank you for having me on the show.

4:25

Yes, thanks a lot for taking the time. This is this is already one of my favorite episodes because I loved the way it happened. It was very serendipitous. As you probably can guess, I'm a voracious podcast listener and I particularly love the you are not so smart podcast, David Mcrainey. And I heard you on MacBook cast in first, like the content of these episodes was really really good in my head was, you know, like, the edge of my seat and wanted to know what what's going to happen. And David did amazing work at that. And but also then, like it appeared to me that it's actually really related to Bayesian thinking or very relevant, very relevant. And so it's like, okay, we need to dive deeper into that on the show. And I can take to do and you say, yes, very, very enthusiastically, so they're

5:28

happy to do it. Yes. Happy to do it.

5:30

Yeah. Well, let's do the dance. But first, as usual. What's your origin story? How did you how did you come to the world? of neuroscience and psychology and how seniors over with that?

5:45

Well, I think it was neither scenarios nor straight. It was kind of a random walk. And it is quite a long story. Sure. I'll throw I'll try to make it as short as I can. Yes. And that is I was a math and physics major in high school. In Germany. Yes. But from that, I mostly realized I just don't don't I don't want to do math or physics, for reasons we can get into later if you want to, but I could not see myself doing it for the rest of my life. Now, interestingly enough, there was this was like in the late 90s. So computers were really big. And I was actually very interesting computers and computer science. So naturally, I decided to study psychology, which was a big surprise to everybody, including myself. doing that for a couple years in Berlin. I realized that I want to go deeper. And I want to go deep meaning into the brain. And so I decided to do a PhD in neuroscience in Chicago. And so that's that's how that happened. But in hindsight, again, in the 90s cupboards were big. So naturally, I go into psychology in the early 2000s The interest started get big, so naturally, I go into neuroscience. So strange, yes. But okay, so it was a very secured Securitas and kind of a random walk, but I like it. I like luck.

7:08

Yeah, I mean, from the from what I know from you, which is like basically the articles on your blog that hey, I read through and, and also, of course, the podcast episode 80. That was very that I can feel your passions and it's

7:27

very, very, very passionate and byways, so So that's actually what I was lacking. So you know, I definitely appreciate the I guess beauty of math Yes, and do rigor of physics, but the passion was not there. Yes. So I you know, I do have passion in some aspects. I really like number theory, for instance, but not this all inclusive passion for the whole field. Yes.

7:51

Yeah. Yeah, I feel that can be you seem to be a very curious person. I'm kind of the same, like very intellectually curious into it can be a curse also because then you you're interested in a lot of things. But you have to choose and I think you have following the things that you're most passionate about is actually a good compromise, because otherwise you can spend months and years and years in things that actually at some point, you're like, Yeah, I'm writing the fire.

8:24

You know? Yes, that's correct. And it's actually very dangerous. Like you could, you know, if you are curious about many things, you could spend years pursuing, you know, avenues that in the end turn out as dead ends. Yes, I agree. It's very

8:35

dangerous. Yeah. And that reminds me of like, something I saw I don't remember where but I'm pretty sure it was a quote attributed to Warren Buffett or Albert Einstein. You know, any, any one of these quotes is one of those. Yeah, exactly. Right. Actually, I think the content was interesting is like, right down to 25 things the 25 most important things you want to do in your life before you die. And by decreasing order, then pick the five the first five, and, like focus on that, but most importantly, the other 20 ones. Also take them and really throw them away and avoid them like play, because that's what you don't want to spend your time on, because you're going to be interested in them. But that's not exactly what you want to do. And so you can lose a lot of time on those.

9:29

That sounds sounds more like Warren Buffett than than Einstein.

9:32

Yeah, for sure. Yeah.

9:35

But probably Charles Munger, right like the brain behind Berkshire Hathaway, Charlie, Charlie Munger, something like that.

9:44

So yeah, actually lets you tell people what you're doing nowadays. And because that's that's what we're gonna dive into in a few minutes. Basically, what what's your work and what's not the topics you're particularly interested in?

10:00

Right? So I would say, the work I'm doing right now, all of it broadly falls into the category of cognitive diversity in other words, recognizing that minds are very complex, that minds are very different from each other. In many ways. Yes. And so if that is true, and I do believe that's true, that minds are very different and they're very complex and you will need a lot of data to fully map out the space of possible minds. Yes. So that's what we do in our lab we do. Investigations of cognition. So broadly conceived, you know, can be perception can be attention, things like that. But highlighting the complexity, highlighting the diversity of different minds at and to really print it out. You have to have a lot of data. So that's what we do. Probably conceived.

10:55

Yeah, okay. I see. And how big of a team are you? I'm working on that. Um,

11:01

that's a good question. It's always a little bit of flux, but I think in my lab right now about 10 people, so you know, we work in small small teams. So something different people work in different things. So 10 people, but yeah.

11:15

Okay. And so you seem to know already about Bayesian inference. So I'm curious if it's something I asked you a guest. Do you remember how you first got introduced to Bayesian methods and how frequently do you use them today?

11:33

Oh, it's very good question. So first of all, I Yes. I remember vividly when I was first introduced to Bayesian methods. Now this was a very long time ago. It was in fall, probably October of 1998. Were you even alive back then? Were you were your life it?

11:51

Anyways, I was eight years old and be celebrating the first World Cup of the French football team. Without knowing what was happening.

12:01

Yes, that's correct. That was that summer 98 was the what's the World Cup in France yesterday? That's I remember that. Yeah. Anyway, so I was that was my first semester in college. And if you can believe it, in the first semester in the very first lecture, a professor GERD Gigerenzer became a professor that day. So he was the head of the Max Planck Institute in Berlin. But that's not necessarily Professor he's being appointed a professor at the Free University of Berlin where I was a student right. And so he gave his like inaugural lecture, and in his inaugural lecture of all things, he talked about Bayesian methods he talked about, I'm sure you know about this, like, you know, the AIDS detection, like if you if you know, what if you know, one of the 1000 people that has AIDS and your your test is 99% positive or something like that, what's your bill that you have AIDS if you test positive? And the answer is, of course, surprisingly, it's actually very low. It probably was very low, like one and 100 or something like that. And that was mind blowing. Because that, you know, as as you said, I was very curious student in high school, study many things, but that never came up. That so I was like, oh my god, this is amazing. I think I was actually thinking right then and there. That's when I decided to, you know, study this more deeply. Oh, ironically, yes. The most important thing, you know, maybe not ironically, but it makes sense, right, that the one of the most important things would come up in the very first lecture. Yeah, so as fortunately as GERD, Gerd Gigerenzer, he's still think you might be retired now, but he was the head of the Max Planck Institute in Berlin at the time. And he gave he gave a very engaging lecture on like, basically, judgment and decision making like how if you if your test is positive, what does it mean? And he had done research showing that doctors had a status ignore the patient prior more or less to go with the seed it goes the likelihood they don't they don't go to posterior or they don't take they don't take the pride to account to get to get to the procedure. They just confused the perceived likelihood and that hidden a lot of research on that. And so again, that was just mind blowing. And I was like, why don't we teach this in high school? You know, but we don't do to reach that an Argentinian High School. No, well,

14:16

I'm French, so or France or in France In France? No, the base theory No, it was after high school.

14:25

Really? Well. By the way, of course, should be called Laplace is theorem, right? Yes. And even even though that should be called that front differential teacher in high school, no,

14:38

no no, no. Unfortunately, not. Like, we have a bit of probability probability at the very end of high school. I'm not sure we see already the Bayesian, Bayesian, the Bayes formula, but we do see probabilities for sure. conditional probability.

14:57

Yes. Yes. Same thing in Germany, by the way, in high school. In Germany at the very end last semester. So we get a little bit of probability, maybe even condition probability. And as you just said, it's then only one step from conditional probability to Bayesian probability, but they don't make that step. So So I was very excited to enter college because I was encountered that in my first lecture, and I think the SEC the second part of your of your question was, how do I use them? And well, depends what you asked me for using them but I use it a lot in teaching. I mean, this comes up every semester at least once, once, actually multiple times. At least once. Right? And also my work as you as you know, so I would say quite a bit, and yeah, quite

15:43

a bit. And when you mean in your work, that means like you're using Bayesian models or

15:51

Bayesian modeling because, as we'll talk talk in a moment, as you just said, like we want to go deeper into what was touched on in that podcast by David McCraney. Yes, I mean, these these potential things are, in my opinion, probably best model in a Bayesian framework. Yes.

16:06

Okay. Even more exciting. Super cool. Yeah. I mean, and we'll dive into that right now. Just want to make the comments that Yeah, completely and share your enthusiasm within it. You know, past medical test example. me even though it's like in to me it was one of the first times I really understood like the power of the Bayesian methods. Because, of course, I made the mistake. You know, it's like, it's like those visual illusions where it's like, cord, like your first response from the brain is gonna be, oh, yeah, it's 99% because the test is super accurate. So you just take the likelihood,

16:45

and that's what that's what I don't tell you by the way, so I've completed my first experience like if you're a doctor, that's what they will tell you so yeah, right.

16:56

Yeah, exactly. Like I remember those studies. Then I dug into that I was like, oh, surely like they did studies was like really well known. Doctors. And like, guys who are smart, and they made the mistake. They don't know. They don't know. It's always like, wait, but why do we do that? And so then, yeah, sorry, I just dug into the Bayesian framework. It's

17:19

true. And I actually use that in my in my own teaching, like, I tell the students you know, you cannot rely on the experts knowing the doctors are experts on the medicine. They're not experts on the statistics or Bayesian methods. They don't know. You have to know yourself.

17:33

Yeah. So okay, let's begin. So I'm gonna have enough time. So let's, let's dive into the meat of the episode. And so yeah, basically, as I said, I discovered your work through the, you're not so smart podcast. listeners should go to the show notes. I have a link there to that episode in particular, and I don't want to remake that episode because it'd be boring for you. And like for me to in the air, you talk about your research that you did on the on the dress, from 2016 You know, the one that people so mainly either in black and blue, which is the real color of the real dress. Or lots of people saw it in white and gold. And so you bet whole episode is about that dress, what the work you did to understand what was at the root of those problems. And in that of that illusion like he will tell us actually how we should call them and then how you need to reproduce them. So well here, right? Can you just remind listeners basically what these dress was about and the problem that it was raising?

18:50

Well, so very briefly very briefly, as you just said in February 2015, a picture of a dress that was taken in a shop in England, went viral. And the reason it went viral is because even though as you just said the dress itself is black and blue, and if it's bright, light, bright, white light, bright, normal, light, regular light. Everybody agrees that yes, the dress is black and blue. But that particular image, the Internet was pretty much split down the middle, although there was a slight bias towards white and gold. So most people just thought it was built like two thirds or something like that. I would say 60% So as white and gold, third as black and blue and then the rest everything else so so. So yeah, so and, and so that's what really, I think that's why it was one of the most still to this day. And one of the most like, you know, groundbreaking things that just everybody was talking about. Yeah. And so, yeah, so that's how I remember to and spirit please briefly, there was this fundamental disagreement between what the color of the picture of the dress is and I guess what makes us remarkable is, you know, it's not subtle, okay? It's, it's good with white and gold, or black and blue. Those things are not next to each other on the color wheel. And I think that's what that's What's What's so striking. By the way. Early on, just briefly early on, people dismissed this they said, well, it's different people have different phone settings, like color sensitive phone and stuff like that. But now you can look at the same screen and different people will still disagree so that that explanation could be dismissed very easily.

20:34

Yeah. And I did, of course, that experiment before before the show. I mean, I'm not scientific but I have that feature now on my phone. After listening to your episode with David, it was like, Okay, this is super interesting. I'm gonna be a nerd and talk about that too, ever. But yeah, I encounter great notes and Tina. And so that was like, Oh, I'm gonna interview that guy, by the way. So he's an ad that I was showing that teacher I was like, so how do you see right now? And of course, it was not the same as me was just like amazing.

21:07

Early on in the first day, there was articles even like Scientific American, I can send them to you afterwards. That said, I was just screen settings, but I knew that I was not. I knew that was not true. Because I had done what you did, like, look at the same screen. Different people. Yes.

21:23

Yes. And it wouldn't make clear also that this isn't like a real picture that completely randomly appeared like it's not an experiment that was planted and afterwards, we discovered it was one, right. It's just as with someone taking your feature,

21:39

it came out of nowhere. Yes, it was not a scientists. It was I mean, literally, the people themselves who took the picture also didn't expect that. It's just that they've had gone they were buying dresses for a wedding and they couldn't agree. What the buyers were to do. To pictures of the dresses, and then later, they were looking at them and they just couldn't agree what the color was. And then they were like what's going on here? Like maybe you see need to see the eye doctor. So someone else I think someone from the band, posted that online and then just took off

22:11

yeah, yeah. And I remember and and awesome. Keeps like that index tweet with David where it's like, it's funny because at some point when seeds become it became like it blew up. It almost became a polarizing issue, or like, like in blue versus white and gold, and I remember people who weren't like seeing it in black and blue, but because I know Beyonce, or someone else was saying, No, I see it. Go then we're like, wait, but that was I want to see it in white and go to like, would it be like Beyonce,

22:41

you know? Exactly like so basically. I mean, initially, I thought he was just trolling me, you know? Like, but it took for me to meet somebody who I trust completely. Not to troll me that I started taking seriously. Yes.

22:57

Okay, yeah. So let's, let's take it seriously. And so let me play the complete beginner, which is not that hard, because it's a very new field. So, like, but basically, isn't that something we already know? Because to me, like illusions, visual illusions are common knowledge, right? If I just go to any science museum, I can be blown away by the amazing illusions where my brain sees something, my eyes see something, but actually my brain interprets it in another way. So what makes that dress particular, if anything makes it particular?

23:37

Oh, yes, it does. So let me tell you why the dress is special and LTU give you at least three reasons why the dress is special. To me, I mean, not just to vision scientists, again, different The first one is prior to that day. So prior to February 2015. Most vision scientists would probably agree with the fact that color is very simple color perception is very simple. The idea is if I measure your human response transfer function to three basic lights, yes, I can predict your response to any light. This is quite the statement but there's a field in vision science called psychophysics, where you can show that if you if I measure what's called your transfer function or your response function to three elementary lights, I can predict your response to any light. It's quite amazing. It's quite the achievement of vision scientist to to be able to do that. Yes. I mean, that's what that's what science is right eye. I can basically break down the complexities of color vision into its element forms just building blocks. And then if I understand those, I can basically protect your response to any light any mixture of lights. That's, that's quite amazing, right? Yeah. And it's essentially basically by the way, the essence of like TVs or screens, that from the elements in your LCD screen, if you watch this LCD screen or your computer or whatever, you can constitute any light mixture you want. It's amazing, right? And you can do that. Well, what the dress shows is no, that's just not true. There's more to color perception than this simple light mixture. Yes. So basically, the brain takes other things into account. That's number one. The second thing is you're absolutely right, and I'm not sure I apologize. I haven't listened to your podcast as much as should. So I don't know. I don't know what else you have covered. But so for instance, in envision, it's very it's been very well known for many years, that Bayesian priors play a role in motion perception. Yes. So let me give an example. Like give an example. Let's say you see something moving and I ask you how fast it is, and you give me your risk estimate. If I then show you the same image or the same moving stimulus at a lower contrast, yes, you will then proceed a slower why? Because your likelihood his estimate is more noisy. So your brain will put more emphasis on the prior and the prior is for slower speeds, because most things are not moving at all. There's no slow moving things in an environment right.

26:19

So basically, when you say more contrast or less contrast that means for instance, less like

26:25

practically Yes, but But contrast is basically the difference between the brightest point in the darkest point. But if you dial the Contrast down, you can show that the brain will put more emphasis on the prior and then you will see things slower but

26:41

yes, you have less because you get less information, correct data. So you have to rely more on the prior and the

26:48

brain is smart like that. And you can show the same thing is true for the motor system. So they showed basically with you know, the analogy is tennis players right. If if you see the ball go to where the ball is, but if I add fog and by the way, I'm happy to send you those papers. I'm not sure if you've covered them on your podcast yet. Yeah, no for sure. We should put that in the show. Yeah, in the show notes, yes. If you add fog basically, then the tennis player will will go into prior like where the ball tends to go. It was shown in the motor system has been shown in the motion system. But to my knowledge, nobody ever shown that in the color system.

27:26

Okay, so that would be for instance, why wait harder for us to drive at night, I guess. Yeah, he's dangerous.

27:35

things moving slower than they actually are. Yes. Correct. Yeah,

27:38

exactly. Okay. Yeah. And I'm actually super amazed still, that we can, you know, estimate quite accurately. The speed of a car moving really fast down the road, can close our node because I mean, if you think from an evolutionary perspective, our system is not at all made for that, like cars are really, really recent. So it's pretty amazing that we have that plasticity to be able to do that already.

28:05

And I will be happy to send you some show notes on that or that some things they could put in the show notes. Yes. Yeah, I'm pretty sure. Now anyway, the third thing is the third reason why it's amazing is you notice illusions in the museum, like you said, like for instance, the Necker cube that you see the cube going this way or that way or the vase, the faces Yes. But most people can see both things very quickly. Yes. Or spinning spinning dancers, paintings, clockwise and counterclockwise all of that. Yes. If the dress the dress, because it has to do with what prayer you have and refurbished prayer I have. It's not something I can easily override. So you might never be able to see by the way, what do you see black and blue or black and gold or black and blue? Or white gold?

28:48

So this is super weird, because depending on the moment of the day, I look at the picture I see differently.

28:54

Okay, so 1% on the way 1% Of the people right? In my dataset. I have now almost almost 50,000 people who did this my data set, right I said That dude is a big scale. Yes. About 1% of people like you dare dare their priors are so close that whatever the mood of the day is, or the lighting of the day is day they switch but most people are not like that. So

29:17

I want to make clear, I cannot switch on demand. It's super weird, like, and I actually had the page on wiki page on my phone up and for a while. And and then it was like, Wait, it was any change color. I was like, Wait, that's super weird. Maybe the image is dynamic on Wikipedia. It was like I don't know maybe it's not the exactly the same image. So I downloaded the image on my phone to be sure that it was the same. And then it was still it was still like from time to time like glue. And then why didn't go I was like what's going on?

29:49

It's fine to mention that because there was a lot of conspiracy theories about that early on that these new sites had switched to image. Yeah, yes. So So yeah, so So anyways, so And the fourth thing that makes it so so those are the three things so far, the third one being that, that you can't switch, whereas in most of these illusions that are in the museum, you can switch? Yes. And the fourth thing is that it's so strikingly different. So, for instance, every so often like a new dress, quote, unquote, breaks, yes. But it's like okay, is this blue? Is it gray or something like that? You know, if there was like really judgment calling? Yeah, I don't know. Yeah, but this is like, black and blue or why to go like, whoa, you know, it's like, this is not at all close. Yeah. If you look at the color wheel, it's not it's the complete opposites.

30:37

Yeah, yeah. It's like, I see a banana. And you see spinach.

30:42

Yeah. What's that? Is that different? Yes. So I would say those are those are the four things that does make the dress special. Yes, it does.

30:49

Okay. Okay, great. Awesome. Thanks so much for that very structured answer to a very broad question. Of course, of course. Okay, so I'm very impatient. So I can craft cannot take at suspense anymore. So there is why some of us see that dress as it is so black and blue, and some of his seaters weight and gold. And then you said like in the experiment you did 1% of the people could could see both Right, right, right. Yeah, and again, like so tell us that but again, listeners want to deep dive deeper. They should listen to that you're not so smart episode for the details of exactly how you discover these in how you will produce the data once we have Yes, dutiful company combination of socks and crocs.

31:40

I see. Well, there's a lot here. So I'm gonna make this as Bayesian as I can, because this is a Bayesian part. So let me let me contrast this with the 50 motion stuff are explained in the context of the motion stuff. So basically, in the motion in the motion case, yes. Everybody has the same prior because everyone has some experience because in the real world, slow things or even not moving things are just much more common and fast moving things. Yes. Which by the way, I promise you, I will send you this paper and I will, but basically, that's actually one one hint that people can see fast moving stuff because it's so unusual, right? So delight likely is very strong. But my point is that so in motion and emotion I mean everybody's same experience. Let's talk about the dress. So the picture of the dress in the picture of the dress. The illumination is not defined. Basically, this was taken inside. Yes, but it's unclear if it's artificial light or daylight because it was overexposed with cell phone, Samsung. Cell phone, flash. So let me ask you, let me ask you as to Bayesian, what does the brain do? If the likelihood is unclear if that's missing and missing information is brain say I don't know. That's the brain say that? Yeah. Well, if the likelihood is if there's missing information, what does the Bayesian brain do?

33:02

So I'm not a special specialist of the brain. But I can tell you what the model would do my model my model. The data is not informative enough. You rely on the priors

33:12

Wonderful, wonderful. So in other words, I cannot so normally to see a color lighting has to be taken into account that's called Color constancy. Yes, I'm not sure if you have a photographer if your audience is photographers, but basically you have to color correct the image right. So to make sure that you know your shirt looks the same way outside in natural light in the daylight sorry, in artificial light in the daylight. You have to take the illumination into account. Yes. But if you don't know the information, if you don't know the elimination, because it's ill defined, what are you going to do? So here's the cool thing. How do I say this? Some people have more experience with artificial light, and some people have more experience with natural light, everything else being equal. Of course, everything else will not be equal, which is why we need large groups. But the idea is that would you agree with me that someone gets up at break of dawn with the sun the sun goes up, that everything else being equal, they will be more exposed to daylight sunlight than somebody who gets gets up late and stays up late? Who will then be more exposed to artificial light? Would you agree with that?

34:19

Yeah, okay, great. I would agree with that. Like, especially in me who is I'm not an early bird. Okay, like I would, I would have difficulties telling you what the you know what the color, consistency ease of the of the dawn of the very beginning of the days like and I can be very more detailed about the sunset glide, which I love and wonderful watch a lot.

34:41

So to make a long story short, people who are morning people who are exposed to more daylight in our Bayesian account would predict that ever daylight prior that their prior of eliminations daylight prior and they will then mentally subtract that I want to be clear, not explicitly not not not like deliberately but unconsciously their their perceptual system will assume the illumination is daylight and then we'll subtract that now. And the light is blue blueish. Okay, if you look at the sky, the sky is blue, right? So if you if you subtract bluish light, a more yellowish image will remain. Yes. If you subtract from gray that's exactly what those people see. And if you have people who get up as low as they are like a night owl, they are more exposed to incandescent light and I will have to show this to you in the show notes. Incandescent Light is more warm, has more long wavelengths, and if you subtract these long wavelengths from gray, then you will get a shorter wavelength image which is bluish and that's exactly what to see. And we did in our study closed the loop so basically people who are morning people see it as white and gold people who are night night people see as black and blue and people who are morning people are more likely to assume it was daylight and people who are night people are more likely to see as far as artificial lights was closed loop. And finally, one more thing, obviously let's say you are a night person but you have neon lights as opposed to ingress light then this wouldn't work. Or let's say you are a night person but you are forcing up in the morning because of your job. I don't know what your job is. But let's say you do, then this wouldn't work. So this is not likely going to happen on the individual level. But in large groups, we found a very large, statistically reliable result and I've now replicated this in like 50,000 people, but you did raise an important point which is so what you know, maybe the dress is just a fluke. Who knows what's going to dress right? As you just said, If you truly understand something scientifically, you should be able to reproduce it. Yes. So we did so. So

36:44

before you go into that because I think it will be relevant. Just a very short question. Why did you start from gray to subtract because you said we have gray, we subtract blue and we have great was great.

36:57

Good. Good point. Gray is just the perceptual neutral points. But basically, if the elimination is ill defined, that's basically been some perception.

37:04

That having basically no data okay, no data. No data is correct. Yes. Okay.

37:10

It's not black. It's important. It's great. Okay, anyway, so to close the loop. So I did address that by myself. And I started to think about the principles. Yes. And then, together with my colleague, Michael College, we implemented this with Crocs and socks, as she said. So we took crocs, which is a kind of shoe which could be any color and be illuminated them with complimentary light, so either green so let's say if it's a green Croc, they illuminate with pink light on a black background. So we look gray or we took a pink Croc illuminate, illuminate with green light, so it will look gray. But then the person or our model, Michael College, wore socks that are white, and those white socks would then if the light was green wouldn't look green. Or if the light was pink with a will look pink. And now comes the big, which we'll call it reveal. And that is if you go with first impressions like what it looks like. Then the crocs will look gray and the stocks will look either pink or green. And maybe you know, there are green green socks, there's pink socks, right. And as grey crocs it's possible. I guess. There's some people and we were able to show that in the in our paper that no, this kind of sock is white and those people will then be able to recover the original color of the croc even because it mentally subtract that light even if it's not there. It's pretty crazy actually. Yeah. So the pigs are gray, but the appearance and the mind has to be you know, this is white. It's going to be pink or green. Yeah, and that's when that's when we knew we basically could show that that's the mechanism because we can we can create

39:07

it. Yes. Yeah, that's amazing. I loved it when you when you went to the detail in David's podcast. This is as I can. Okay, that's when it also like that's where you see, for me that was my first real exposure to Mike. Let's say color science, because me what I know about wavelengths from color and some come from physics and astrophysics, and I know, like we use the wavelengths stuff like that, like you know things that are like objects that are further from US, Russia. Yes, yeah, exactly. With the red chief concern. I know it's very important for that, but it was funny for me to see it also, like basically here on Earth, we can use that. That was super cool.

39:49

And just just to mention the Bayesian aspect of that croc sock thing. So basically, some people have a white sock prior. And I know that sounds crazy, but where does it Where does it come from? We asked in our study, and it's basically people who have more experience with that kind of sock have a more likelihood stronger likelihood to have that kind of prior? Yeah, again, again, this sounds pretty wild, but we have the data.

40:12

Yeah. I mean, this, to me doesn't make sense. Because like, maybe it's because I'm Bayesian. But in so basically, all of these what's really super important here is people's priors, right? Yes, basically, it's these kinds of experiments are a way to reveal people's priors about the phenomenon in question, right? Correct. Correct. Nice. I think it's like kind of the first time I've listened to that in a way that because prior elicitation is really very hard. Like it's notoriously hard in the Bayesian framework, correct? Because, because there is still that, you know, illusion that we don't have priors, like, you know, like people are like, but I don't have any progress in that. And then what I say is, now I don't I don't say of course your prior, you stupid. I'm like, Okay, let's say you don't have an AI just ask questions. You know, right. Like it. Let's say that. I don't know where if you try to have a model about, like trying to model the patterns of migration from birds, you're in you're making a model, where do you need priors on I don't know the velocity of birds. And I don't know anything about that. But then I'll just I'll just ask the question. So do you think birds go as fast as light? Right? No. Right. Well, that's prior right. Do you think birds go as slow as, you know, turtle? No. See? You have priors. It's just like they are so ingrained, correct? Actually don't don't know you have them. And that's

41:52

actually a really good point. So basically, if I asked you, you know, if you have a white sock prior, you would say, of course not or you have or you have a daylight prior, something like that. So it is actually a nice way to show that they are these priors aren't do exist, and the only manifest like voodoo effect and behavior or perception, you know, but you would never know how would you know, you know if you have

42:13

or not. Yeah, and actually from the I'm reading that book right now. From Robert Birkin. I think on being certain, okay, and a to neuroscientists, and it's super, super interesting in that part, the very first part of the book where he talks about priors and how they are basically ingrained and also, it makes sense because otherwise if you had to, you know, all your priors all the time, it would basically make you very, very inefficient, human being.

42:44

Exactly so as a cognitive scientist, I can say that thalassitis prior biases priors bias you for action and as you know, evolution being correct as a kind of a secondary consideration. But the first thing is you need to act to take action, and take action without getting crazy or being anxious. You have to be fairly certain more certain than today data justifies, but if your prior is good, that's fine. If you can feel good, that you'll be fine. You don't need to need to wait until you have all the evidence to get me to slow. Someone else is gonna

43:15

you know, take your food, eat you. Yeah,

43:18

you are the food. Yes, you are the food.

43:20

Yeah, exactly. Yeah. Yeah, most of the time. So those priors work really well. Sometimes you have to you have to think about them. And and yeah, exactly right. Basically in that Google so he goes into basically a spectrum of being some people being totally unconvinced most of the time because they have obsessive compulsive, compulsive, obsessive obsessive disorder or something like that. Yeah, yes, yes. Yes, exactly. You can see it like that and then you have the other end of the spectrum of people being totally convinced of something and then he takes the example of John Nash, for instance, who has like absolutely convinced against all evidence that he was going to be the next emperor over Antarctic. And and

44:03

that's a very good point that you raised. You know how I told you earlier that as they like, cognitive diversity in my lab, and that's exactly the kind of thing that we're looking at. So basically, there are people that look up into the sky, and they don't see constellations, they just see individual stars. Yes. So they don't they don't jump to conclusions. They don't connect the dots aren't people. They see all kinds of things. You know, just look up and they see whole stories play out mythology and all that. So there's a huge difference in the propensity of to connect the dots Yes, and to jump to conclusions. And as you just said, some people are very willing to take priors to jump to conclusions and others are not. And, and as a society probably needs, it needs a good mix of both, you know, some people who are cautious, and so we will very willing to take risks, and then we'll sort itself out. Yes.

44:48

Yeah, exactly. And indeed he goes, he goes into those topics in the book and that is fascinating for that because it's also like also the difference between knowing and the feeling of knowing and how I show you feeling of knowing in a lot of circumstances takes precedence and so you can feel like you know, something, when actually don't or it's kind of totally fine. Yeah, so basically that means that so the only thing is related to priors, which I found amazing. In Yeah, I mean, as a as a Bayesian modeler and also like, teacher, for me, it's really useful to have those concrete examples, because that helps me teach basically, about the Oh, yes, not only the importance of priors, but also the fact that it's just like they are they're this like, you know, try and elicit them. And then the problem is how do we elicit priors from people consistently. And we go through, you know, biasing too much that elicitation of priors, because ideally, you want people to really tell them their priors, and not priors that they think are acceptable, or something like that. And so that's really something important and hard to do is still

46:09

you might want to look into the slow motion prior. I'm gonna send you a paper that's very consistent. It's a very strong effect, and it's very consistent effect everybody has, as far as I can tell, because slowly moving things are just much so much more common and fast moving things.

46:24

Yeah, that makes sense. So yeah, definitely that that's super cool. And to me, does that mean that? Does that mean that the brain is essentially patient machinery?

46:37

Well, I'm gonna say, I don't know. And let me tell you why I'm hedging. I have a nuanced answer. Let me tell you why. The perceptual system Absolutely yes. I mean, it does ORTEC that the motor system Yes, and I will send you the papers, racial notes, but and I'm also gonna send you those papers all you probably already know about them. But you we just talked about this, and yet doctors ignored it prior you know, I'm saying it decision making, and yet people ignore the protein you know about the farmer. This is a classic classic example of Bob. Bob is a farmer. He has glasses, and he reads a lot of books. What is right, what is more likely that he's that he's so excited? It's unfortunate. Bob reads a lot of books and he has classes. What's more likely that he's a farmer or his librarian? And most people say, well, obviously librarian, but that's not true. He's more likely to be a farmer because there's 20 or 30 times more farmers than there's librarians, because people ignore the prior. So what's good what's going on? Part of your brain is definitely Bayesian, like perceptual system motor system. But another part the reasoning part, I guess, every time you show that and as you as you know, you can get an old Nobel Prize for showing this Kahneman and Tversky, right. Not so much. And GERD Gigerenzer voice doctors No, they don't think so. Which one is it? And I not sure is that is difference. Conscious was unconscious if that makes sense. In other words, the conscious brain is not Bayesian. But the unconscious what is maybe, maybe, maybe that's what the unconscious mind is priors. You know, you know, Freud said he knows what was in the unconscious in sex. Maybe we know what's the unconscious it's just priors.

48:36

Maybe make more sense to me.

48:38

Yeah, me too. You know, Freud, something's a fraud. But anyway, let's not go into that. But the point is that that's how he sold it though. He was like, I know what's in your unconscious. It's just sex, sexual urges. That's what that's what sold. That's what that's why he became popular, but I think I agree with you. I think it's probably priors.

48:58

Yeah, they were. Let's see making all of that.

49:00

Anyway. So that's my, that's my position on that. And I still haven't done the key experiment, right. So can you turn some reasoning more or less Bayesian depending on how much more conscious you make it, but that should that we represent? Yeah.

49:15

Yeah, that's definitely the thing I would like to do afterwards. Like, can you can you control that slider? Yes. Because I'm guessing that dark situations where it's evolutionary more interesting to be evasion and then not, you know, depends on the situations. And well, I'm guessing that the brain tried to try to do that slider, but we don't know how it does that. And so, if that is being How do you do?

49:40

That is my prediction that the brain will be adaptive, in other words, that when it's when it's when it benefits it, it will use it, obviously won't and that's actually by the way also what I told my students like that Bayesian reasoning is great. If you have a great prior, if you have a bad prior, then you're in trouble because I will not be able to change your mind. Yeah, I mean, let's say you believe some crazy stuff for whatever reason, right? And you are 100 Sure. 100 isn't sure. What am I what am I going to tell you? As you know, if you're if your probability of prior is one, it doesn't matter what the evidence is, I can never update that mathematically impossible, right? So the brain might want to be more adaptive about that. Well, sometimes I'm not I'm not going to be patient at all. And maybe that's what consciousness is. Remember? The idea from a neuroscience perspective, that conscious like this, this thing you need when you're in an unknown situation? So actually, what I would track is I'm saying that we've got your prior when you based on autopilot, yes. So basically, have you ever driven home not knowing even where you were going and what anything that's unconscious, right, and that's because of a good you've a good prior how to do that.

50:45

Kenny, definitely nit the other day and at least something really that made me laugh. I always take the stairs to to go up and down my flat and even on the fifth floor. Okay, and so the only day it was just like in my thoughts going up, and then I start I opened the door from the stairs to go to my client, and then in front of my door. I'm like, that's not my fight. But just you know, it was super fast and like, I'm not that I'm sure I'm not that the right floor. And I wasn't the fourth floor and not the fifth one. And after lessons like next week, because I don't know how did I know it was not my flight? You know, because I didn't see the I didn't see the number of the floor. It just and then it hit me like oh yeah, there were no 100 You see a road. There were no rug in front of the door. And it was completely unconscious. Like I'm

51:43

constantly realized that

51:44

Exactly, exactly. Yeah.

51:48

There you go. So that's the that'd be my prediction from that that the more conscious you make the processing, the less it's going to rely on priors.

51:56

Yeah, that would make sense, because that means that you cannot take the decision to magically so that probably is correlated with the fact that you have bad priors or Yes, priors that are irrelevant views. And then the system is like okay, we need input here because exactly not know what to. Exactly. Cool. So actually, you talked a bit about that. already. You said like in 1% of your test, people could see the two different combinations of the of the dress. I'm guessing you try to to replicate that also with the cracks. Oh, yes, basically. So can you just have like it's technically possible to be able, so it's technically possible to see that to see both combination, as I do, for instance, depending on the moment of the day do you so you said it's because the priors for both cases or two are too close basically. Right. And also so maybe if you have interested to dive on that. And also I have a related question, which is, is it technically possible to see the dress or the crocs and soaks in totally different combinations of colors? Dress For instance, we we suppose that it's going to be blue and black or white and gold, but can someone come back and be like, I see it green and pink. Is that also technical? Okay. So

53:23

there's a lot here So briefly, not green and pink. But what is relatively common is that someone is like, say an artist. And they say, and they say I see it as blue and gold, blue and gold. So the music combined the two. And actually, if you look at the actual pixels on on Photoshop, the color dropper, they are actually right. The pixels are blue and gold. So there are some people who are able to ignore that prior because they just good of colors because of their background of like, face can see through it, you know, I'm saying they're like their job is your job is just so good of colors. Because that's what you do all day in jobs artists or as a photographer, photographer photographers to actually that they can just tell and they will tell you it's actually blue and gold there, right. Hmm. And frankly, and frankly, that's what I see now because I've worked on this so much and I now see blue and gold.

54:17

Yeah, so to see that didn't you see that the trace is actually black and in blue.

54:23

None about the image is blue and gold.

54:26

How's that possible?

54:28

The black the dwell time from the lady from the from the from the it's appropriate lighting from the photo. So the track the dress, one second I have it here somewhere. I can get it over there. The dress is black and blue.

54:40

Under under normal way.

54:43

But exactly but the pixels in the image actually blue and gold. Okay. And that's what and that's that's what people who are very familiar with that. Photographers, artists, they see that but there was another question that you had no, it's about i Yes. So some people see it like about 1% If they switch Yes. And and so so here's the experiment I want to do. I haven't done these yet for purely logistical reasons. But I have the content information of several 100 of these people. And what I want to do you add me now I now use those a couple 100 plus one. And what I want to do but again this for for purely logistical reason I've not done this yet. I want to send them a incandescent light bulb versus a light bulb that is very like you know, the other end of the spectrum. And that'd be like, Okay, you use this light bulb for a couple months. And then we'll see if we can lock you into this or that pool. Because one thing that's not clear yet is one possibility is that the priors are just very close and we can bias you towards one or the other. Yes, that's our hypothesis, but not possibly building is there are some people that just don't rely on priors that much at all. And then it has nothing to do with that. So and we don't know which one it is yet. But that's one of my long list of experiments that I want to do. It's just I haven't done it yet. But yeah, that would that would that would be a tease apart that question is it that some people just have present too close? Or is it that they don't rely on present all that much?

56:16

Yeah. And it's probably maybe even something else.

56:19

I was gonna say or maybe even something else. I was just gonna say that yes. Or a little bit of both. But But if our story is true, and I think it is true, then it should be possible to bias those experimentally. Push them in disorder.

56:34

Yeah, yeah. All right. Can tweet, trumpet experiment. And so as we, as we said, basically, that, that experiments you did with the crocs and socks, yes. Basically, that that shows that we, we can elicit people's priors, we can identify their priors, even if they don't even know that they have those priors.

57:00

Yes. And actually, they will be ready to price. Go ahead.

57:04

Yeah. I mean, that's incredible. I love that.

57:08

Yeah, no, it's a wild wild story. I think people will be very surprised if we told them, Yo, you have a white sock prior. And by the way, this is another this is another experiment I want to do. But I haven't found a object yet. And maybe your listeners can help me. Here's, here's what I'm looking for. I'm looking for an object, a candle, some object that is one color in one culture, but another color in the culture and X ray so maybe Brazil vs. Argentina, or something like that, or France was Germany, that the same object looks different in different cultures. I didn't know. I haven't done

57:47

so do you mean when I think when you tell me that object in my brain is picturing something or when you show me that object? My brain? No,

57:54

no chest experience, like let's say, I don't know, let's say male trucks are red in France and blue. I don't think they are but I should say that's the case. Yeah. And then I'll show you one gets kind of ambiguous. With LiDAR, then you can you can maybe elicit that prior for people from that or that culture? I think so. But again, I haven't found a suitable object yet. Yes, but that if I'm right, that should work and then we can raise visualize that prior. Yeah. Or elicit a prior. D I know you tell me Is there like a prior for let's say soccer balls are yellow in Brazil and white in Argentina, but I'm not sure that's true. Then we could use we could use that.

58:38

Yeah, I see. I see what you mean. Okay. Nice. Well, definitely people you you heard Pascal so if you have any ideas? Let us know. And we'll get

58:50

going. Same object, but different colors, different colors that could be very valuable.

58:56

Yeah, for sure. Yeah. Like how is that it's a very important part of your experiments like you. You talk about that actually in the in David. My trainees podcast episode were like actually finding all your combination of Crocs and socks was very long like it took you a lot of work. That was fine. That

59:17

that was the contribution of my collaborator Michael column, which I had experience of Kraken so he he had experience with crocs I wanted to use like candles of different colors. Maybe that were would have worked too, but we settled on the crocs and socks.

59:30

Yeah. Yeah, honestly, it's, it's the kind of things that when I do a model, it's like a lot of things, you know, happen under the hood. And, and people don't see them but they actually take you a lot of time. That made me think about that. It's like it's just like the hidden work. A was a lot of that. Well, the

59:49

idea of expand was one thing, right, but implementing and making it work is not easy.

59:54

Yeah. Yeah, for sure. So, to get back to those prior elicitation. So basically, that's that's what he chose. You can people's have priors, people have priors, even though they don't know they had. So Yes. And does that mean that basically, we can change people's priors and be maybe that could bring us closer on some important issues like these, like here? It's reasonable to dress in some crops. But of course, those priors come up all the time. And so, yeah, my question is, okay, we know people have priors. Does that mean you can change those priors?

1:00:34

Well, how do I say, if I'm right, I'm not sure if I'm right, but if I'm right, then priors are set by experience. Yes. And so that would imply that by repeated exposure, you can change those priors. However, I am concerned about that. In other words, you said that it could unite the people. I'm very concerned that it's used right now actually, to divide the people. The idea being that if you if you consume a certain kind of news outlet, yes. And somebody else consumes another kind of outlet, you might stop agreeing about what even constitutes reality as and as far as I can tell, this is used by people who control these media outlets right now to actually control and divide the people or to divide the people to control the people. Yes.

1:01:26

Yes. Yeah, that's gonna be my next next question, which was actually worrying me when I was writing a question. Yes. Yeah. Which is like, but then can you change people's priors without them knowing? Because if they don't know they have priors, then probably you can change them without them even realizing.

1:01:43

There's no doubt in my mind that media companies social media companies are doing his right now. Now, are they doing that with that in mind, I don't know. But that's what's effectively happening that say you're on Twitter, you're in a certain group of people who share similar. You might think everybody's a Bayesian, because in your in your network, everybody is yes. I think every revision scientists, and so the idea is that historic is actually something that's very interesting and very scary, actually, historically, right? You would live in a small village and a small tribe like I'm talking about 10,000 years ago, yes. And you would all be more or less exposed to the same experiences? Yes. In the modern world, particularly online, that is not true at all. We all listen to different shows, listen, different music, watch different shows, listen to different read different newspapers. I'm very worried about that because it's now possible in the 21st century to take control of somebody's mind, if I'm right, without them even knowing it. And how would

1:02:49

you do that? Like, like, Yeah, can you can you just explain an example? Oh,

1:02:53

yeah, let's say you own the New York Times and The Washington Post and CNN and MSNBC and you just expose people to fake news all day long. And then they will start to believe that the world is not as it is, they will have all kinds of imaginary concerns. Manipulation

1:03:13

myth handles that relate to the Oh yeah, that trans to the prior speakers, He changes your experiment of something. You'll keep seeing if you keep seeing, for instance, the news that I know there are terrorist attacks all the time, then your brain is like Wait are way more terrorist attacks than I thought we definitely need to do something about that.

1:03:32

Correct. And so And there's lots of research on this already, actually not not with the Bayesian framework. But the idea is that people think something is issue, if the news mentioned is more Yes, even whether it's an issue or not, actually, and of course, they knew that their own incentives like they're being paid by clicks and you know, things like that. So they might have a vested interest to get create outrage when there's no reason for it. Because outrage sells, click sell, you know,

1:04:03

for sure, yeah. That's another part of the brain. And actually, let's act like because time is flown by better but I'd like to ask you a bit about the modeling part. Because you said actually, you're using Bayesian models. Yes, we model the brain which is it's right

1:04:26

or this this part of it that Roger this part of it but yes,

1:04:29

yeah. So yeah, like maybe take an example. How does that work? Like the surface models are really patient models? Yes.

1:04:38

Yes. Yes. So from our from a Bayesian perspective, yes. Innovation is and again, I will send you the evolution of this in a in the show and for the show notes. Yes. But But basically, I don't know. I'm trying to retrace our steps like in the history of like cognitive science, yes. But basically, by the early 2000s, it was clear that people use priors and perception. And I already gave you several examples. And there was highly well published and made a lot of impact and all of that. The next step, and that's where we come in is that different people have different priors. So basically, in a population modeling, you now have two priors two different priors. So one population has a long wavelength prior to light and one has a short wavelength of light and the UVA competitive model to protect who sees what so that's the direction so we go from a model of one prior to a model with like, two priors and they're competing with each other. So that's the from Bayesian modeling perspective to innovation and applying that to cognitive responses or perception.

1:05:48

Writing, okay. And so I guess the structure of your models is he's a Bayesian structure. So you have your data that comes and then the whole structure is your model, which is priors on your parameters. And, and then how do you run those models? Like what would you use?

1:06:08

Oh, you weeks? That's a good question. So we constrained it was just that's what that's why I said and I don't know, you know what your background is, but you said something about astronomy.

1:06:18

I didn't study that, but I'm fascinated by

1:06:20

Okay, okay. But anyway, so the point is, in one word, big data, like we use lots of data to fine tune the parameters and update the parameters of those models. You know, have a likelihood function, you know, me like yes,

1:06:37

but which I'm curious about the software. Oh, are using some open source open source tools like how how does that work

1:06:45

in your Oh, no, no, we use we use MATLAB and Python. We just write it ourselves.

1:06:50

Okay, thanks. Okay, if you're using Python you should give a try to play MC.

1:06:55

Oh, look, I'm sure that's true. It's just that yes, you're right. We should but it's just that

1:07:02

we're writing the center's for you. You're gonna it's gonna save you time.

1:07:06

I see. Yeah, that's that's definitely true. It's just that it's also not a big deal, right. I mean, I mean, are you right? We should not we should not reinvent the wheel. Yes.

1:07:16

Yes, I guess. Yeah. I guess Yeah. But he's like, basically your as the shell team is in the model. And like, basically, you're having the the nerdy people doing the algorithms and so on, and then you can just rely on that is something that I found super valuable. Because myself first please myself. I'm not I'm not a mathematician. So I actually having those frameworks empowers me because I don't have to do Heart Math. To do my models. I can do code, and then I execute the code and the code is doing the math. For me.

1:07:52

That's that's fair. That's fair to that's fair to that's also fine. Yes.

1:07:56

Okay, yeah. Well, if you definitely if you do that when they differently, let me know and if you need if you have some questions, but PMC Of course, I'll be happy to help.

1:08:05

Thank you. Thank you. Thank you for offering.

1:08:10

So maybe more of a more philosophical question. Okay to start closing of the show. I'm curious to you what does your research into these topics say about us as humans in in particular, how about how we literally and figuratively See the world differently?

1:08:35

Well, that's a good question. I think the reality is that an eye sees my students. Most people just assume that the way they see the world is the way others people see the world. And that's because in their mind, because that's how the world is yes. The reality is, and this is a little scary, which I understand people don't want to hear that. The reality is that you're seeing the world through a particular lens of a particular filter from a particular perspective. And that might and that's colored literally colored by your experience. So that might not agree with somebody else's perception of the world. Right. And I do think that we have to come to terms with that. Let's say, let's say I say something, and you say I disagree. My first instinct is to say, Well, you're wrong, or you're crazy, or you're trolling me or whatever. Yes, but I thought you're stupid. Oh, yes. Yes. Yes. All of that. Yes. So but I think we need to start evolving or developing router, a culture where we say, okay, we disagree. So that's interesting, right? Why is that? Is it perhaps possible that we have different priors and if that's true, were these priors come from? Is there a way to reconcile this somehow, you know, and right now, I don't see that culture at all. I see conflict. I see disagreement. I see polarization. And I think a lot of those unfortunate Yes. And there is love I would say needless conflict, because, you know, I think that we are not aware. I mean, not all of us, but some most people are not aware they have these priors of the way they are impacted their perception. And so I think a lot of conflict could be avoided by just acknowledging that that the way you see something is not necessary, because that's how it is. But because of your prior and someone else might have a different prior. And let's resolve that, you know?

1:10:39

Yes. Yeah, that's why I love this idea of priority station, right. It's basically priors seem to mainly be unconscious, your guardian. And so we are not aware of them. At least consciously and so like we think we're unbiased in we think we see the reality as it is, and and then if you're able to elicit the priors and show people that they do have priors even though they think they don't well, that can change the game because yes, that that makes you like anybody would be like in all attend, you know, it's like wait, what's

1:11:19

even know I was thinking that and so that's why it's so important. What you do, because making people aware of his Bayesian framework in general, right. And then on the importance of priors in particular, I think that could be a game changer, because as I said, I was not exposed to this whole Bayesian framework until I went to college right? And only because I went to college in this field, whatever had either not gone to college or not going to this field. I might have no idea that this whole framework even exists. And then we have what are you talking about? And so I do think we do spread awareness and then once that is a newer thing, then we can talk about specifically Okay, let's make these priors visible and then the ultimate goal will be to live happier together because as you said, or implied or what maybe I imply that a lot of this suffering is inflicted by other people on other people. There's no just no need for that. You know, I'm saying it's literally needless, we don't even disagree. We just think we just agree.

1:12:17

Yeah, yeah, I see. Damn. Cool. Well, perfect. Perfect time to own listing the show because the last two questions, asked every guest at the end of the show, maybe just before that. So if people are interested in following you and so on, everything will be in the shownotes. And maybe, which projects are you most excited about for the coming month?

1:12:44

Month? Well, so what we're doing what we're working on right now has nothing. Well, that's not true. But it's using different symbols material we use music for review. And but it goes similar similar ways. Basically, we look at like, how to, to make a long story short, the music you like you like for certain reasons, but not for the reasons that you think we can show in our study that has very little to do with the actual sway forms in the music more like with what you associate with that, but what you associate to music might be different with ICICI so it's a similar like thing, but with a totally different stems material. Okay. And I'm very excited about that because it says same story again, but if something on a surface level totally different listen to music, you know?

1:13:31

Yeah. But yeah, it sounds super interesting. Definitely let me know when that's out and it will be worth another episode.

1:13:38

Sure, happy to do it. But what I'm saying is, is that striking that how you feel about some music, and literally what emotions it evokes, has very little to do with like, the actual music but reflect what you associate with the music.

1:13:49

That's already freaking me out.

1:13:51

Right? Stop here.

1:13:54

And I can I can then I can also imagine that we food. Right, right is that the kind of food that you choose? And that you want to eat a lot or you know, Stress Disorder related to food can actually be related to stuff that's completely different than actually the food.

1:14:13

Correct. It's like it's not about the food. It's not about the dress. It's not about the socks. Kroc, it's about these deeper issues, Your Honor was right and it's a little scary to think about that.

1:14:22

Yeah. But it's like access accessing parts or yourself that you didn't even know were there.

1:14:28

And they're influencing you there. They influence your perception and behavior and behavior. Yes.

1:14:35

Yeah. So but at the same time, it's really awesome if we can get there because it's like okay, that's cool. Now we know and we can decide what we can do but that 100% Awesome. Well, okay, before letting you go to questions and ask every guest at the end of the show. I would have zillion questions. That's that's like. So first one, if you had a limited time and resources, which problem will you try to solve?

1:15:11

In your world or in science are both up to you? Okay, I will I have to be honest with you. I don't think I can just pick one because they all relate it. But But let me mention three that are pretty much equally important in my mind. One is the scaling problem. And that is the idea in a small society of 100 people which is what humanity spend 99.9% in that we evolved in 100 200 feet people. There are very few issues with like credit and like responsibility and all of that. Yes. In the modern world, you could make a bad decision but I pay the price for that. So society society societies that we scale, let's say you advocate for bad policies, that that rise crime, but I didn't get mugged. But you are you know, you can get the virtuous signal on the internet that you are a good person. That's unfair, right? I get mugged because of your bad choices. That's the problem society. So it's basically responsibility and incentives don't scale well inside and this as far as I can tell an unskilled unsolved problem. Related to that, in general, we have a lot of evolutionary heritage. Yes. That made sense in the evolution environment for instance, is realizing prior to act quickly so that you can eat something as opposed to being eaten, right? Yes, in the modern world, we should maybe have a higher priority on like accuracy and cream correct. So but evolution is not going to debug that. So we need to find another way of getting there. And the third and the third, which might seem unrelated, but it's very dear to my heart. Remember, I talked about cognitive diversity? Are there some people that don't have a sense of morality at all or conscience? You know, conscience, they don't feel they would hurt you didn't feel bad at all. They would feel good. Oh, yes. They won't they feel good. They will feel good that they got away with something. And that that's called psychopathy. In you know, someone like it's I don't like that term for many reasons. We can maybe do a different show about that. But basically, the bottom line is that there are people out there who don't respect value by as a merchant. They'll just see our faces game. And basically no matter what, what economic system you adopt, there will be hijacked immediately by people like that, as long as you have them running around, you know, yes. But that's a long story. But those are the three years that are all equally important. And I would like them all solved and solving them might be interrelated. You can resolve one of them while the other ones too. So maybe do a trifecta together.

1:17:50

Okay, nice. I love it. And also related to what you're already doing, so I'm going to write right. And second question, if you could have dinner with any great scientific mind that alive or fictional, who would it be?

1:18:08

Also, really, there's a tie. It's a tie. So I'm going to say several of them that I worked from a psychological relatively good stuff to to Fechner he was kind of the first second physicist. And if that's surprising, you might look want to look into that. So maybe he started his whole there's a physicist, he started his whole psychophysical investigations of the mind with clearly find some physical stimuli more broadly. Leonardo, right. Leonardo da Vinci. Yes, who can go wrong that like genius? Yes. Yeah. Other than that, John von Neumann definitely will be very interesting conversation, and underrated layer. CLR layers. He's kind of like the mind behind the mind. You know, the mind behind Bob. We had a chain reaction. If he hadn't written so much already, I would always add Richard Hamming. But I kind of know what he would say from that. Or Charles Munger who already came up but he already wrote so much. I kind of know what he would say so I will stick with those for Gustaf to to FECKNER Leonardo, John von Neumann, and Alyssa Lara, if those for you I will be happy to any of those four. We put them on an Android random I'll be happy to all for

1:19:25

anything. Yeah. Well, if you have the four four of them, it's C two parts and it's not the dinner anymore.

1:19:33

But I would have with either one of them. And I guess maybe we could add base but the problem is, is he wouldn't even know he did what he did you know, so he died. Yeah. So he'll be very surprised. He will be like, Why am I at this dinner? What's happening? Yeah.

1:19:49

I wish I didn't do anything. Yes,

1:19:52

yes. He will be very surprised. Yeah, Laplace is actually interesting. But again, he also published so much that I have a pretty good idea what he would say and I'm saying, Okay,

1:20:02

well, let's do a fake goodbye now and then stop the recording. And I'll tell you what to do with the audacity, correct. Well, thank you. Yeah. Well, thank you, Pascal, as my prior was that I would really enjoy that Ian would learn a lot. Yeah, the data didn't contradict my privacy. That's good, I guess you

1:20:25

have your prior. That's wonderful. i By the way, same here. I was looking forward to this and I'm very happy with it.

1:20:31

Awesome. Well, great to hear that. And, by the way, if you have any guests recommendations, like you understand what the podcast is about, like if you haven't guessed recommendations, definitely shoot me an email. I

1:20:44

will I will do that on the show notes. One is Alan Stocker. He did that motion prior and Conrad coordinated again. He did. He did the Moto prior if you had if you haven't had them on already. You should have them on your show. And tell them I tell them I said so. They will. They will know that you're very busy, but they will have a higher likelihood to say yes.

1:21:09

Yeah, definitely. So as usual, I resources in links to your websites, okay. And also the show notes for these excellent are going to be huge. So that's, that's cool. For anyone who wants to dig deeper. Right. Thank you again, for your time. And Lisa.

1:21:29

Thank you. Pleasure was mutual.

1:21:30

Thank you. So let's stop everything. Oh. And now if you go to a DCT

1:21:43

Yes, I click X. Stop. Yes. You stop. Yeah, you can keep the syskey Okay, no, I stopped and now what

1:21:53

and now export export as web

1:21:57

as as what is W AP is and I want to call it

1:22:06

whatever you want, but that that is that needs 24 beats PCM

1:22:14

is signed, signed. Yeah. Yes sign. Okay.

1:22:22

And then you can save it. We don't care about the meta data. So just leave it blank. And then you save that wherever you want. And when you have some time off, you can send me fine. So I'm gonna Drive or Dropbox or we transfer whatever you want.

1:22:40

Okay, tell you what, ah, Alex I have to go somewhere now. So after I come back from dat in like an hour or so I will send you those two papers for the show notes. The contact information of who did that I do think should have we should have done the show. And I will send you that. That expert.

1:23:00

Yeah, awesome. Thanks a lot. And for sure, like yeah, if you can make the introduction with those people that rate I will be happy to join

1:23:09

them. I called colleagues. I'm happy to do it. Yeah, so I hope I hope this is what you'd mind. It's perfect.

1:23:18

Right? Okay. That's even better than what they had in mind.

1:23:22

Even better, okay, great. Yeah. So, so yes, I have to go now. It's I have a 330 appointment. So let's talk soon okay. Or, at some point,

1:23:33

yeah, by email, send me it and we'll put him in. Okay.

1:23:37

But very nice meeting you, or saying thanks. For the free e meeting you Yes. Okay, are you thank you, bye bye.

Transcribed by https://otter.ai

Previous post
Next post