#1289 Wataru Toyokawa: Social Learning, Collective Decision-Making, and the Wisdom/Madness of Crowds

00:00 Ricardo Lopes: Hello everyone. Welcome to a new episode of The Dissenter. I'm your host, as always, Ricardo Lops, and today I'm joined by Dr. Wataruto Yokawa. He's unit leader at the Computational Group Dynamics Collaboration Unit at RCN CBS Tokyo. His research focuses on the computational underpinnings and eco evolution. Eco-evolutionary implications of human social learning and their relationships with group decision making and collective behavior. And today we're going to explore topics like social learning, collective decision making, the wisdom of crowds, and other related topics. So Dr. Toyokawa, welcome to the show. It's a pleasure to everyone.

00:43 Wataru Toyokawa: Hi Ricardo, thanks for having me. This is such an honor to talk to you.

00:47 Ricardo Lopes: Well, thank you. Thank you so much. I really appreciate that. So, uh, let me start with this. Um, WHAT is social learning, uh, which aspects of it are you most interested in and perhaps also, uh, tell us. If, uh, perhaps I can ask you uh it as a second, as a second question, but tell us also a little bit about the methodology that you, you use to study social learning.

01:19 Wataru Toyokawa: Right, right, yeah. Yeah, of course, yeah, so social learning, uh, well, literally it's, it's learning, um, but that is influenced by observing others or learning that is, uh, shaped by interacting with other individuals, usually the same species like con-specific individuals, but like, you know, it could be extended to other. Asians. But anyway, like this social learning is widely defined as this uh learning process that is somehow um related to some social uh situations. Um, SO that which means that covers a huge range of behaviors such as copying what someone does or just paying attention to, uh, the other's behavior or like paying attention to who succeeds or who fails at some tasks. Or following the, the majority's opinions or sometimes we defer, defer to experts advice or, you know, something like that. So, yeah, I, I find it, social learning is really exciting topic or exciting phenomenon because social learning is uh one of, one of the main drivers of, you know, collective level behavior and cultural evolution. So, yeah, in a nutshell, um, social learning. Create sort of feedback processes and biases in the information transmission in the population. Uh, SO which means some behaviors spread through a population while some other, uh, behaviors might be suppressed or disappear, uh, due to the, uh, you know, the social learning processes or copying processes. Um, SO yeah, this means social learning is not only about how individual acquire information, but for me, social learning is interesting because uh this also shape kind of how culture evolves and how social norms and technologies uh kind of fat fat and faster dynamics uh all emerges at the group level. So, yeah.

03:33 Ricardo Lopes: Uh, uh, AND I, and I mean, uh, yeah, just to, uh, ask you again the, or remind you of the, of my question, yeah, the, the second part, and there's, there's also the third part, but the second part, I mean, which aspects of social learning are you most interested in?

03:52 Wataru Toyokawa: Yeah, right, so, yeah, that, for me, um, you know, it's interesting because social learning, by definition is just a one way of learning, right? Like a learning by social social ways. But this kind of automatically means that. Once animals or human individuals employing some social learning processes, then it starts creating a feedback processes and collective level dynamics due to the interactive uh nature of uh social learning individuals. So, I, I'm very interested in this, uh. I, I'd like to disentangle uh between how individuals behave and how, as a result, how collective level pattern emerges. So, yeah, this kind of seemingly complicated and sometimes nonlinear relationship between these two layer is interesting and this is why I got interested in social learning process.

04:59 Ricardo Lopes: Uh, AND, uh, I mean, what kinds of methodologies do you use in your work to study social learning.

05:10 Wataru Toyokawa: Primarily, I'm then basically interested in the, the nature of uh this collective dynamics emerging from social learning strategies. Um, BUT I am particularly interested in human social learning because um because humans are not blind, kind of, uh, not random imitators. You know, we are known to be kind of surprisingly strategic about. Uh, ABOUT, uh, how, how we use social information. So sometimes, uh, you know, we, without even realizing it, we, we sometimes ask, like, is my own, is my own information good enough or is this person likely to know something that I do not know? This type of, uh, you know, then we would. Like we would choose the timing uh at which we want to copy others and also we, we usually select someone we want to copy from. So, this kind of strategic use of social learning is very uh remarkable in human behavior. Um, SO, I'd like to understand, uh, that patterns and Why we do that, how we do that, uh, by, uh, combining both, uh, modeling, like computational modeling, as well as, uh, running some behavior-based experiment. These days, I usually use online game experiments. Um, SO, kind of, yeah, I recruit, um. A bunch of people, uh, human participants, uh, joining the, the, my, my, our task online and, and then I, uh, just organize some kind of a simple multiplayer, uh, online game, multiplayer games, uh, where, where people hopefully enjoy some decision-making tasks there and while I'm getting some data from that.

07:12 Ricardo Lopes: Yeah, yeah, I mean, but that kind of approach that you use, I mean, what kind of information can we draw from it when it comes to Uh, how people learn socially, I mean, uh, what kinds of, for example, limitations would you say it has when comparing to how people would learn from others in a more sort of a natural environment or in, in an environment that would be, of course, I'm not. One thing to imply here that uh doing or having that kind of methodology is not ecologically valid, but you know, something that would be perhaps closer to how we learn socially in a more in a way that is closer to how we do it in the real world, let's say.

08:10 Wataru Toyokawa: Right, yeah, um. So for sure, there are various ways to uh to mimic uh real world complexity in, in sort of tasks, but um I'm, I'm particularly interested in the Kind of so-called exploration versus exploitation dilemma in decision making. So usually like uh this is like a very common feature of like learning and decision making in animals, including humans obviously. Uh, YOU know, because we have limited time and we have limited resource and lifetime is limited, right? So we are basically have to choose. Kind of timing or circumstances where where we should go for exploration to get more information versus uh we should uh sometimes switch to use that knowledge we have collected so far because otherwise, otherwise there's no point to, to learn things, right? Because when you learn something, you have to, you have to leverage that knowledge, you have to use it wisely, right? So then the question is, uh, you know, how we, how animals are we. Have a, a kind of, we, we usually balance this exploration and exploitation, and also, uh, in theoretically speaking, uh, we can ask. What kind of strategy should be adaptive, uh, and what kind of balance between exploration and exploitation should be optimal, uh, at the given circumstances, for example. So, uh, because I'm interested in this kind of nature of the general, uh, the common nature in, uh, animal decision making. So I'm therefore have been, I have been focusing on. Uh, THE so-called kind of reinforcement learning tasks, uh, like called multi-arm bandit program. So that is just a very toy model, that is very abstract, uh, gambling task kind of, but it's not necessarily gambling, but, um, it's, let's say you are the casino player. Right, or you know, then you have, uh, you have, you have multiple slot machine options in front of you, right, then your, your aim is to gather more coins from, from playing the piano, playing the slot machines. Right, so but you don't know which slot machine is optimal, you don't know which slot machine is good and which one is bad. At the at at the outset, you don't have any information. So you should therefore have to, you should, you have to play some slot. And then get some experience of like getting some reward or didn't get reward at all and this kind of repeated experience is eventually shaping your understanding about the world, understanding about which slot machine seems to be good and which slot machine seems to be bad. So this kind of information, active information gathering processes and decision making processes are kind of. This, like in this process itself is kind of called multi unbounded task in this uh kind of uh ndi uh theoretical world. And then of course that kind of task is very abstract, but thanks to that, thanks to this abstraction, uh, we can capture uh various range of real world situations ranging from, uh, animal foraging, for example. Uh, OR like, uh, which, uh, which stock, uh, you should invest in the long run, or, I don't know, uh, like a mating, uh, made choice decision making perhaps is also could be captured some partly by this sort of reinforcement learning process. So, yeah, so obviously there's a kind of a trade-off between abstraction and, uh, focusing on more realistic. Uh, CONCRETE programs in societies, but, um, I'd like to kind of, yeah, there's 11 extreme. I'm taking kind of one of, one side of the extreme, like, uh, going to the very abstract, uh, program in a way.

12:40 Ricardo Lopes: So I, I mean, this might seem like an obvious, uh, I, I mean the answer to this might seem obvious to many people, but why do we learn from others? I mean, why is it that we rely on others for information? I mean, what do we get from others that we couldn't just get from ourselves, let's say.

13:06 Wataru Toyokawa: Yeah, that's actually really excellent question. It's, I don't think it's kind of obvious, uh, to answer. Well, intuitively, yeah, it's obviously, uh, we, we learn from others because, you know, it's, it's a very easy way to learn something, right? We learn, uh, from others, uh, because the world environment is too large or, you know, the skill we need are too complex. And the environment is like too often very uncertain or very costly to explore if you are alone. So every individual, you know, this is due to the again, the fact, due to the fact that every individual having limited time, limited resource, a limited attention window, right? So then. So in this sense, we can see other people or other individuals as a distributed sensors er in this environment. Because they have, they might have already tried things that you haven't tried yet or that you, for example, uh, other, other people might have failed at things already that you don't want to fail at by yourself. Uh, AND also some, some might have discovered new opportunities, uh, that you might otherwise miss. So. And instead of yourself, if there are many other individuals around you, then you, someone else can, might have got some information and then if you can simply observe that or use that, then it's one of the cheapest way, like safest or cheapest way to obtain the information. Um, THAT'S the kind of intuitive, uh, one, I think, uh, that's obviously like, that's true. It's, it's one of the, the huge benefit from social learning. Um, BUT, um, social learning is Yeah, although this is partly about. Reducing the cost of exploration. Um, BUT I'd like to say like I, I, I think. It's, it's important not to describe social learning merely as. Information on free riding sort of thing, right? Uh, OF course this is cheap, uh, but. If this is just a cheap, cheap. You know, the being cheap is one of the, just the sole uh benefit from social learning. Then this is a, this is true for everyone else as well. So then people just start. Copying each other and then because it's cheap, or I'd like to go to the cheaper way and then. And then it end up everyone just being copying others, and then no one would explore new things and no one would uh go to the environment, taking the risk. So that's kind of dead end. Um, BUT we know that social learning, uh, have been generating, uh, like a culture evolution in human histories driven by social learning, and we also kind of intuitively think that, well, we kind of do, do very well thanks to social learning. So, yeah, and then this intuition is actually true, I guess. So because social learning is not just information free riding, I'd say, um. Because Uh, then, like, the interesting part of social learning is coming in. Um, SOCIAL learning is usually a, uh, more kind of strategic and flexible process, uh, and You know, When people use social learning strategically and interactively, um, as maybe we, we, we're gonna discuss this, uh, in the later, but, um, this can create kind of synergistic benefit in the collective level, um, because like intuitively like social learning can guide people, people's future exploration. And then, therefore, It's not just like it's not just free riding others finding, but it's actually. Uh, BY following, by copying others' behavior, this would guide you, this would tell you which direction you should explore, and then, which means your individual exploration behavior would be improved by the social, social effect, social learning processes. So this actually this benefit can accumulate uh by. Interactive processes between individual exploration and social learning. And, so, so this interaction I guess is key, and this is the key main driver of so-called collective intelligence.

18:19 Ricardo Lopes: Right, but I mean, of course, uh people, or at least I don't think people just er choose to randomly get information from anyone, right? I mean they have some criteria. I, I, I mean, what, what kind of, are there, for example, er biases, any kind of cognitive shortcuts like. Heuristics that are cues that people uh look into or that perceive when it comes to who they should acquire particular information from. I mean, uh, are there, for example, uh, social cues like uh the social identity of someone, whether it whether it is because the person is. Is older or the gender of the person or for example the person has high status or the person is has particularly high status when it comes to certain kinds of topics or particular kinds of information and so it's a reliable source of information. I mean what. Kinds of cues or or criteria do people use to decide, I mean, whether consciously or subconsciously that OK, if I want to learn more about this, these are the people I should trust. I mean how do people go about that?

19:54 Wataru Toyokawa: Mhm, YEAH, yeah, yeah, so I think. The, the like a various. Uh, OBVIOUSLY, there are multiple, uh, wide range of different ways in which we, uh, we rely on others, we, we copy others. Um, BUT, uh, the literature has, uh, kind of sorted because there are wide range of possible different ways of, uh, collect, uh, social learning strategies, social learning, uh. Like a use of social cues. Um, SO the literature has, uh, sorted out, uh, uh, these multiple strategies into basically roughly three categories. Uh, SO one is like, uh, when strategies that regulate, like when you should copy, when you copy. Uh, SECOND one is who strategy, like, uh, regulate from whom you should copy. Then the third one is what strategy that regulates, uh, what kind of information you, you should copy. And then. So there are two, yeah, the, I think it's important to differentiate between, uh, what human, what, what kind of strategy do humans use and, and, uh, and, and what kind of strategy would, strategy would be, uh, beneficial or optimal under the given circumstances. Right, because humans are not necessarily optimized to, to use. Uh, THE best way of social learning under the given circumstances. So. Um, FOR the As a, as a fact or or like as a, as a pattern of human behavior, uh, the, yeah, bunch of experimental empirical studies in uh human uh social, social psychologies and uh behavioral, behavioral sciences, um, including my, uh, my own previous work. Um, MY previous studies have, uh, basically focused on kind of conformity by social learning, that is, uh, kind of humans sometimes dis disproportionately copying the majority, like, you know, people sometimes follow the majority more likely, uh, uh, you know, more disproportionately likely, uh, uh, compared to the actual proportion of the majority size. And then this sort of conformityized social learnings are uh emerging more uh when, for example, like uh decision making tasks, tasks are difficult and people rely start relying more on those type of social learning when groups became larger. So, human social learning is, is really flexible. Uh, SO we, we change how, how to use social learning, how, how, how we follow others depending on, uh, the given environment or social circumstances. So, uh, so which means it's, it's really hard to say, uh, what kind of cues human usually use because it depends on the context. But, um, but other, uh, well, well studied. Patterns be, uh, yeah, it's the empirical bunch of empirical experiments have shown that people uh pay attention to, um, others' success, so which is called, um, success-based or payoff biased, uh, social learning, um, so which is, uh, also like again widely beneficial because It's, it's generally a good idea to follow successful individuals. So people would pay attention to those others, others, uh, success states. But um, but it. It can also be misleading sometimes, uh, when, when outcomes are, for example, noisy and when, and which is, uh, causing some success is basically purely due to the luck, you know, so, uh, you might like, oh, I'd like to, I'd like to follow this person uh because, you know, this one seems really successful, but this The, the fact that this person is very successful might be, might have been just by, by the luck and then in that case, copying what, what this person has done might not be so informative, right? So, so there's some like obviously this kind of cost of uh copying the pay-based social learning, but um. Yeah, or,

24:51 Ricardo Lopes: or also, I mean, I don't know if this is relevant to what you were saying there, but I was just thinking that maybe, I mean, sometimes celebrity, I mean we tend to associate celebrities to success in some way, whether it is financial success, social success, or whatever, but I mean if a celebrity is trying to sell us a product and they claim, OK, this is going to cure your skin disease, I mean maybe the fact that they're a celebrity should be, or I mean people should treat that as irrelevant to their sort of medical knowledge because they don't have any, I mean you. Right, so I, I mean how do people deal with that kind of information? I mean, do people tend to just blindly follow anything that a highly successful person tells them, or do they, or are they more careful and perhaps they think, oh, OK, maybe Cristiano Ronaldo. He's trying to sell me a health product, but he's not a medical doctor, so maybe I should be careful when it comes to what he tries to sell. I mean, how does it,

26:16 Wataru Toyokawa: so it might be good reason to listen to him, but, um, uh, yeah, in general, whenever we have Any sort of behavioral bias, not only like social learning biases, but also any, any decision making biases, then surely that uh the market or economic, economic dynamics uh might uh try, try to Relate, like utilize this bias uh in, in the uh kind of in the interest of some uh stakeholders. So, um, in generally speaking, we should be, uh, yeah, kind of aware, uh, maybe it's this is why, uh, I think education like economics and psychology is, uh, a good, good thing to teach because um. Yeah, we, maybe it's important to recognize uh that human, human behaviors are full of bias and and it's usually works very well and this is why we have, uh, the nature have evolved us, uh, this, uh, this bias in the first place. But obviously there's, it, it's usually comes with some cost or some come comes with some downside of this, having this bias and. Yeah, so it's in generally like it's good things to have some meta-cognition or some kind of think about why, why we are interested in this and why we are kind of tend to. Yeah, but, but it's if it's, I think it's very difficult usually because these biases might be just eliciting without even realizing it. So, um. But yeah, and But for the this prestige bias social learning case, I guess this is not uh well. These social learning strategies we, I have been talking about was usually has been focusing on the informational benefit of social learning in general. So, yeah, I was just talking about like because while environment is uncertain and decision making is like very difficult, so it's, it's a good idea to use socially obtained information uh for this information purpose. Um, BUT, uh, in this, especially, I guess, it, when it comes to the prestige bias, social learning, uh, there's another possible proximate, uh, motivation behind, uh, this is probably like kind of more like a creating, uh, Uh, social bonds like a, or like a peer pressure sometimes like to, to, to show that, uh, to show some kind of loyalty towards, uh, the person, for example, or to, to have like some, uh, to feel like a bonding with this fandom, for example, you know, if you are a big fan of, yeah, I'm a big fan of Rolling Stones and Rolling Stones, uh, yesterday announced a, like a new album. Yeah, and then I, I like to wear this, for example, like a Stones t-shirt. This is, well, this is not really like a kind of prestige biased copying, but you know, this sort of like a. Showing a feeling, the feeling, uh, this, uh, unified uh community is another, could be a possible, uh, driver, uh, where people copy something, people follow some, someone's behavior for sure, so yeah.

30:04 Ricardo Lopes: OK, so, but in our societies, I mean in every single society, of course, people are not carbon copies of each other. We have individual differences. So how, how is it that we learn from others? Despite our individual differences. I mean, how do, how do we do that without having, for example, identical goals and identical preferences? Right,

30:31 Wataru Toyokawa: right, yeah, yeah, this is, this is my, yeah, my favorite, yeah, the classic social learning research uh that I have been talking about often. Usually assume that we are individuals are all facing the same problem basically with, with the same uh having uh with the same reward function, uh maybe having reward the same utility function, the freefra functions. But in real life, as you said, that is not really, that is never true actually, right? Like for example, my, my, my favorite scotch whiskey may not be your favorite. Or no, perhaps, perhaps you don't drink whiskey at all,

31:21 Ricardo Lopes: do you? I mean in my personal case actually I don't, I don't like alcoholic

31:24 Wataru Toyokawa: beverages. OK, yeah, fair enough, fair enough, right? So we have like a totally different preferences and, and goals and stuff. Uh, SO yeah, so obviously my best risky option is not necessarily your best option, um, but that, that does not necessarily mean that my experience is completely useless to you, uh, perhaps, uh, but because we usually. May still share something in common. For example, uh, yeah, let's say, let's say you're, you're going to be visiting Tokyo, right? And then you want to find a good, good restaurant or good bar, you don't drink, drink, drink at all. So, yeah, you, you want, let's say you want to find a good restaurant and I may know some good, uh, good restaurants in Tokyo. Um, BUT yeah, then now my personal favorite, personal list of my favorite restaurant. Is not necessarily the, the perfect list for you, but you may have, you know, because you, you, you have different tastes, but, uh, but my list could still be useful, um, because it may tell you what kind of places. Uh, EXIST in the Tokyo and which neighborhoods have interesting options, which neighborhood might be uh good to visit, fun to visit, to, to explore restaurant in general. Uh, SO my, although the exact, my list is not, uh, fitting well with your taste, but still the information I give is somehow could be informative in some sense. So in other words, Socially obtained information often contains both useful part and useless components. So some parts are very very applicable to your own goals and circumstances, but other parts are just too specific to, to me, like to the person who originally had that information. So, yeah, so that in general, social learning should be beneficial if, if you can somehow extract the useful part from, from what you've just observed. So the, yeah, the question in the key question here is then like how, how could this, you know, how could we do this job that differentiate between useful part and uh unrelated parts.

33:59 Ricardo Lopes: And, and how do we do that? Do we already have a good understanding of that or not?

34:06 Wataru Toyokawa: Yeah, so this is like a kind of related to my uh previous collaboration work with uh with Alex Wit and Charlie Wu. Both were at the time was in uh University of Tubingen in Germany. Um, SO we, uh, there we put people, um, in a simple decision making task where rewards, kind of rewards differed across individuals, but rewards are still. Positively correlated. So that means, like let's say this is again the banded task, multi-banded task, but multiple options and the options are kind of uh aligned as a 2D map. So you can just imagine a uh kind of 2D uh 2D map landscape and, and then you, you don't know which corner, which. The point of the map is very beneficial for you. It's, it's matching well with your taste, right? But you have this unknown, uh, to an unknown restaurant list in the, in Tokyo, let's say, right? And then. Then it might then uh each participant has their, their own map, right? But in Tokyo, and then of course which restaurant is best for each participant, uh, it, it's different between participants, but um they are still somehow correlated. Um, SO that means, you know. So this creates the situation where, uh, you know. It's still somehow, uh, you know, the knowing others behavior might be still beneficial, like still informative, but you, you shouldn't follow exactly, uh, follow what others are doing, right? So, by, uh, you know, having this sort of simplified abstract version of, uh, uh, this social learning task, we, uh, To, to capture the, the uh social learning behavior of human participants, we've uh created a kind of computational, uh, cognitive models called social generalization model, uh, that captures that where we try to capture this idea. Um, SO in this model, um. Uh, THIS, this model uh captures, try to capture how people might arbitrate between their own direct experience and Uh, what others are doing. Social observations here are useful, but they are kind of, you know, in this model, this social observation is treated as a noisier cue than than one's own experiments. So, you know, this kind of Yeah, maybe I'm, I'm gonna, uh, dive into a little bit mathsy part later, but, um, in this social learning generalization model, uh, we Um, model, computationally model this kind of arbitration between Your own taste, your own experience, and others observations, and then social observation is just treated as a noisy cue. And then this noise capture the, this kind of distance between my taste and your taste. If, if you, if I treat your, if I, if you, if I treat social observation noisier, then which means I would think others are. Uh, NOT very, very well matching with my taste. So that's why I kind of give it more noise. And then, so by, by doing so, we kind of, we could capture the, uh, the human use of this social generalization processes. And then, yeah, it turns out like we kind of run online game experiment and then fit this model to human uh participants' behavior. And then we, we found that um this model actually uh fitting very well with um people's behavior. So, kind of, yeah, so, although this is like uh, again, very abstract toy, um. Uh, ABSTRACT decision making, uh, tasks, but, um, uh, at least we can say that humans are still very flexibly use social information, uh, adapt like even though we are not necessarily, uh, having sharing the identical taste.

38:48 Ricardo Lopes: Mhm. Yeah. Yeah, no, that, that's really fascinating, and let me just say that, I mean, even if I don't quite like alcoholic beverages, maybe sometimes I would also like to know about uh alcohol or, or about whiskey specifically because I might Have someone who I want to be friends with who likes whiskey and maybe then having that learning about what's the best whiskey out there to give it as a gift to that person and impress her. I mean, maybe it's useful. I don't know.

39:25 Wataru Toyokawa: Yeah, it's, it's good to explore new, uh, you know, new dimensions, uh, that you haven't explored before. Yeah,

39:34 Ricardo Lopes: yeah, yeah. So, uh, let me ask you now about, uh, collective decision making. So, uh, I mean, I, I, I don't know if this is, uh, or maybe it is a particularly Western thing. Uh, I don't know if in Japan because you, uh, live more, uh, in a more collectivistic society, if People would actually think about this in different ways, but we tend to, when, at least we in the West, when we think about decision making, we tend to focus a lot on individuals making decisions, on individual decision making. But what is collective decision making and how does it work?

40:18 Wataru Toyokawa: Yeah, I see, yeah, that's, thanks for asking this, uh, very interesting. I never thought, well, um, first of all, like you mentioned about collectivism versus individualism, like a cultural, uh, differences. Um, THIS, this sort of dimensions might not be necessarily like aligned with, uh, well with like uh collect patterns of social learnings or patterns of collective decision making. Uh, FOUND in, uh, in, for example, West and East societies. Um, YEAH, but, so I was, uh, yeah, so I'm gonna be talking about, uh, collective decision making, um. And then, yeah, actually, um, well, this is still like an ongoing work and it's not really uh published yet. Um, BUT we recently um had um kind of museum experiment conducted, uh, collaborating with, uh, Max Planck, uh, Institute in Berlin people. And we have like kind of very simple social learning tasks. And that is implemented in this simple kind of tablet PC. So it's quite easy PC like a social learning task. And we implemented this task in both Berlin Museum, uh, Science Museum in Berlin and also Science Museum in, in Tokyo. And yeah, we just kind of gather some data from, uh, science museum goers and. It's a fantastic thing is that there are a wide range of uh people participating in, in our task, and there are age ranges ranging from like uh toddlers or like uh uh younger than 10 years old till like uh more than 70 years young people, like the wide range of uh people participated. And then surprisingly, we found very interesting pattern like age. Uh, YOU know, related to social learning patterns. Like, uh, usually, like, roughly speaking, kids are like really, uh, using social learning a lot and people, uh, kids are copying others a lot. Whereas like teenagers, like uh like in a starting. Just focusing on themselves and then stop using social learning that much. And then this social learning patterns coming back slightly as, as like age is uh again going up. So this interesting kind of U shaped pattern. IS very robust and we found. Very similar pattern in both Berlin and Tokyo. So, if like, I, yeah, so this result itself might be a little bit um uh seemingly inconsistent uh from some of the previous arguments, previous uh suggestion that kind of collectivism, uh, individualism, uh, and then like. Predict, uh, people's use of, uh, social learning. Um, BUT, as, as long as at least our data, uh, Uh, are not really consistent with that argument. So, but yeah, obviously like there might be like we, we still don't have good theoretical understanding, uh, relationship between these two, collectivism, individualism versus social learning views. So perhaps we have to work on that. Uh, BUT, um, but yeah, sorry, sorry about that. Um, I, then let me then focus on, uh, what is collective decision making, um, and how does it work. So. Collective decision making is, so what happens when decisions of multiple individuals combine. Uh, EITHER explicitly or implicitly to produce an outcome at the group level. So, you know, that, you know, this general definition includes. Uh, KIND of familiar cases such as voting, discussion, deliberation, deliberative discussions, committee meetings, or like a jury decisions, um, and situations where the group needs to reach, uh, some sort of consensus. But, uh, on the other hand, collective decision making also includes, uh, many. Collective phenomena where no, no explicit consensus is required, uh, such as consumer behavior in the market or like non-human animals, uh, fish school, um, swarm behavior, for example, and, you know, or like social media trends, traffic jams, and people joining, joining a queue for like a lot of ramen shops and so on. But in all of these like. Cases, individual decisions. Interact Each other and then as a result, generating the uh the large scale uh group level pattern. So in a sense. Um, YEAH, I'd also like to mention that in this sense, cultural evolution can also be viewed as a, a form of collective decisions. Um, YEAH, although, although these two topics have traditionally been studied. Uh, SOMEHOW separate, uh, somewhat separate academic disciplines as academic communities. Um, BUT for me, like cultural evolution asks. How behaviors or how ideas, technologies or social norms spread and disappear over time among the population. And yeah, which falls, you know, which fits nicely with uh the definition of collective decision making. And although it's, it's time scale uh might be, you know, longer in culture compared to the some animal like a collective behavior and fat fat and fashion dynamics of collective behavior. Yeah.

46:45 Ricardo Lopes: 00, OK, so do you think that there would be an avenue or a potential avenue to explore in trying to combine the literature in cultural evolution or cultural evolution, cultural evolutionary theory and the work on collective decision making.

47:08 Wataru Toyokawa: Mhm, YEAH. I think that's um the key here would be um. I think again, like mathematical approach would be key. I'd say, uh, especially, um, I mean, like partly uh the collective, collective animal behavior people have been doing this job very well. Um. You know, in the collective, collective animal behavior literature, you know, there are, there are researchers studying ants, there are researchers studied honeybees, you know, there, there are other people studying cockroaches or then like a fish school and other mammal, uh, primate behavior and so there are various different systems, but, uh, they're like. Collective behavioral, uh, camp they they have been doing like kind of applying a very simple, uh, abs again, very abstract but the abstract level, uh, dynamical model into, uh. That, that can capture that wide range of different systems in, in a mathematical form and then uh try to uh make sense of seemingly totally different systems and different behavior, but we can still understand some essence underneath of that through the uh very simplified uh mathematical formula and I think we can still, we can then. Understand Uh, and then, of course, culture evolutions, culture evolution has also a similar tradition as well. The cultural evolutionists, uh, have Kind of developed a very nice formal models to capture a different uh range of Uh, different cultural evolutionary dynamics and systems. Um, SO, I think it's already like we can just compare these two mathematical forms and then usually, sometimes it's really similar. So sometimes like terminology might be different, but mathematically speaking, it's could be the mostly the same thing. So, in this sense, unifying uh these two systems is, you know, it's already like uh have been. Uh, HAVE been worked on by, uh, the, the researchers. Um, BUT still like, yeah, so, but. Even I guess maybe maybe due to the gap between the time scale, like usually cultural evolution, people are interested in Galunga uh time dynamics, time evolution, yeah, and, and then this would also involve some uh demographic. Transitions as well, like some member of the society is transitioning by like a demographic dynamics. I think this part and then then which then which drives longer term gene culture co-evolutionary processes as well. So, culture evolution could tend to uh have more like a genetic evolutionary. Uh, EFFECT as well. Yeah, compared to, uh, just the collective behavior, but yeah, this is also, uh, uh, probably, uh, many people might be, uh, disagree with me, but. Um, But like still, I, I guess these two systems are usually the same thing and it's just a matter of the time scales and yeah.

51:03 Ricardo Lopes: No, no, yeah, that, that's really a, a, a fascinating insight. I, I guess that perhaps the are many times the issue with implementing such, uh, I, I mean, I mean, trying to integrate information or knowledge from different kinds of, uh, uh, I mean, coming from different academic communities, sometimes the issue there is that. Uh, WE have this sort of, uh, departmental disciplinary approach to science and knowledge and Uh, many times it is really hard on an academic level for people to, uh, cross communicate and collaborate with people from, I, I mean, sometimes it's even considered another discipline or people are working in even in another department and sometimes people possibly are. NOT aware of the theoretical foundations and the methodological framework used by their colleagues that are working on other topics or in even what is considered another discipline, right? So there, there are, uh, it's unfortunate, but there are some constraints there, right.

52:28 Wataru Toyokawa: Yeah,

52:31 Ricardo Lopes: Uh, so, uh, let me ask you now about, uh, crowds, because I mean, there's these very pervasive, I, I mean, I, I'm, I, I'm not sure if it's pervasive necessarily among academics, but it's, uh, I, I mean in society in general, it's a very pervasive idea that Uh, the idea of the madness of crowds, that crowds, uh, go, go mad sometimes, and then there's these crazy ideas about, for example, a mass psychosis and all of that, but I mean, what do we know about how crowds tend to behave? I mean, are crowds usually wise or mad?

53:16 Wataru Toyokawa: Yeah, right. Yeah, like your, your view is exactly the, the view from like, uh, the view of the traditional social science, science view of the crowds. Like, yeah, like, yeah, it's, I think traditionally speaking, social science, the main agenda, one of the main goal of, uh, goals of social science, uh, uh, uh, from my view is, was, um, to understand. Uh, THE, the, the why, why crowds are so behave badly. And so this is why, yeah, and then exemplified by uh the famous, uh, the book written by like a journalist, Scottish journalist, uh, Charles McKay. Uh, HE, he wrote a, like a Collective, uh, madness of crowds and collective illusion, madras. And then in that, in that book, he kind of listed, uh, like a coup list of interesting, uh, case, case study where human societies went into kind of madness of crowds, such as like a very classic, yeah, tulip, uh, bubble markets uh in Netherlands. Uh, OR, you know, the, the, um, England, uh, like, uh. Stock market bubbles. Uh, SO this, um, yeah, this kind of seemingly, uh, malfunctioning of society, um, was the, has been the focus, main focus of the social science. I guess this is understandable, uh, because, you know, usually, uh, we don't pay attention to Uh, the system that is working very well. We, we don't, we don't really even realize this is working well. Uh, WE just, we just kind of, uh, feel that this is just, uh, just a normal state and this is, uh, you know, it should be. And then when it's got stuck, we realized, oh, what, what happened and why this isn't working well. So, yeah, uh, but which actually means that collective system works surprisingly very well. Uh, uh, IN a wider range of situations, of course, then sometimes it's. BEHAVING very badly. Um, YEAH, but having said that. Um, ACTUALLY, um, It's really, I'd say. That I, I'd not say crowds are usually wise. Uh, BECAUSE I, and then I'd say. I'd say crows, human crows or animal crows have the potential to be wise under the right conditions. Um, YEAH, the classic wisdom of crowds idea. Works very well when. When individuals have diverse. A diverse opinions and at least partly independent information so that individual level, uh, individual error errors can, can be canceled out each other by by merging these independent informational input. But, uh, real crowds are usually very interactive. People watch each other, you know, influence, influencing each other. And people copy others' behavior, so that kind of interactions are everywhere and which can improve. And then You know, this is social learning, so this interaction can in general improve the performance. But At the same time, uh, this social interaction would violate the, the, the, the requirement for the wisdom of crowds. So because this usually reduce the diversity of the opinions, independence of the opinions. So, yeah, therefore, like, it seems like uh social interaction is kind of a double-edged sword. So this can work very well or work very bad. And then I guess, so this is, has been one of my kind of key questions in my uh past uh research. Yeah, this like under what circumstances this social interaction uh due to social learning can can mostly promote collective intelligence and, and under what circumstances. Uh, THIS collective and social learning processes is, uh, kind of detrimental in terms of, uh, collective decision making performance.

58:04 Ricardo Lopes: Uh, uh, BUT I mean, in what circumstances can collectives demonstrate, uh, what is called maladaptive herding? I mean, uh, when does that happen?

58:19 Wataru Toyokawa: Yeah, so Then, yeah, I went to uh so this is another my kind of collaboration work with uh my um collaborator um in, in University of Saint Andrews um. So Again, so according to the wisdom of the crowd's idea, it might be, it seems to be a key to understand, uh, under what circumstances, uh, people, people rely more on others' opinion. People rely on uh copying and under what circumstances people kind of are still uh balanced, well balanced between individual learning and copying. Then, uh, yeah, then the previous research have suggested that people are, well, people are very flexible about, uh, this use of social learning. And previous theory suggests that people might rely more on social learning when the, the situation is, becomes uncertain and the group. The size of the, the number of other people you are interacting with increased. And so we simply like manipulate these two things, like, you know, then, then we observed what happened at the group level behavior. So we, again, this was an online game experiment. And where we, we changed the, we controlled the difficulty of the task as well as uh we controlled the the size of the group, each group size and then. And then we studied two things. So one is how individual people would react, react to these different environments, how, how people change their social learning behavior. In this changing, uh, uh, different, uh, uh, combination of environmental setup. And then the second one, we focused on was, as a result of that, what kind of group behavior emerged. So, and then we found that like surprisingly, um, When like group size is large, become large and uh task was difficult, people start like relying on stronger conformity bias. And then as a result, People get stuck uh in the more, more likely to get stuck in the suboptimal option due to this heavy reliance on conformist social learning. Whereas like when task was not so difficult and group size is moderate, people are still kind of uh flexibly, like a group level behavior, it's still uh flexible. So Yeah, so the Collective intelligence and madness, uh, is actually, uh, somehow again, within this simplified framework, uh, but we could, um, probably we could predict the, the, the How um we could predict the behavioral patterns of given collective uh group by, by estimating, uh, the, by estimating individual level decision making strategies. So, there's a like a very Uh, predictable linkage between individual level versus, uh, individual level and group level behavior. And then this linkage could be, uh, quantifiable by, uh, experimental method.

1:02:13 Ricardo Lopes: Mhm. Uh, SO, but, uh, uh, that flexibility that you mentioned there, is that one of the aspects that, uh, sort of regulate the wisdom of crowds? I mean, is the ability to remain flexible that allows for a collective to be wise,

1:02:34 Wataru Toyokawa: or? Yeah, I, yeah, this is, so there, there, there must be multiple different definitions of what is wise, but, uh, at that, in that study, we defined like a collective, we, we kind of focused on aspect of collective intelligence, uh, that is, uh, kind of how adaptive the groups are, like how adaptively. Uh, CHANGE their behavior, but, but to, to, to react to the environmental, changing environment. So, for, for, for, from, uh, for us, it, it wouldn't be so collective intelligence if Group might be performing very well at the fast, uh, at the initial, you know, state, but if they get stuck in this single behavior, even after the environment has changed, then, uh, we, we define this as a kind of inflexible, and then we define this as a harding effect, uh, you know, it's like a animal group, uh, gather, you know, behaving together as a hard. So Harding effect is easily like kind of emerging when when social lining is too strong. So this is like a byproduct of having this strong social lining.

1:04:03 Ricardo Lopes: Mhm. OK, great. So, I mean, let, let me just ask you one last question. Uh, THIS just came to mind because I guess that many people who will be listening to our interview, our conversation would, I, I, I, I mean, would want to ask you this. I know that this is probably not something that you've been working on directly, but do you think that this, uh, the literature that comes from Studying collective decision making and uh what we call the wisdom and madness of crowds could apply to understanding political phenomena. I mean, because I'm asking you that because, you know, many times that's one of the most immediate things that come to people's minds. I mean, like I I I don't know, sometimes people even apply it to the. COVID pandemic, other times they apply it to uh extreme political movements to uh the rise of the Nazis and the fascists and all of that to uh genocides that have been committed across history. So do you think that uh there could be insights coming from this literate literature that would help us understand these kinds of phenomena?

1:05:30 Wataru Toyokawa: Right, right. Yeah, yeah, this is like a big question. So I don't think I can, I can give a, an answer within this uh time. But um let me try. um. Yeah, I, I think so. Yeah, I, I, I think, yeah, we can provide some insights or, and which comes from. Like, so far, like what, what we have, what we have found in general so far from this uh project of social learning and collective decision making was that. Social learning strategies, social learning is Beneficial because social learning works as a tour guide, like a guide, exploring, like a guiding you to like a. Efficient exploration rather than, rather than that, uh, rather than social learning being giving you uh a right answer, immediate answer, right? So, you know, based on my, my past research or also based on other uh researchers from uh our communities, um, social learning, social learning. Tells you like. Which direction of, uh, which area of decision space is worth exploring? Or what kind of, uh, what kind of, uh, you know, the direction or area is, is worth checking in, in your, in your future exploration. And then, yeah, there's the one very, my, my favorite study coming from honeybee study, honeybees, uh, nest site selection. And yeah, so honey bees in the uh in I, I guess many of, many of you, uh, uh, you people already know familiar with honeybees are really good at choosing the nest when they are kind of swarming in the Aria spring, right? And then they are remarkably good at choosing the. Better new nest site and they can do that by like a scout bees, inspect possible sites and communicate through uh through this finding via famous wagu dances, right? Uh, SO at the first glance, now because they are interacting with each other via wagu dancing. So this seems to be like violating the, the requirement for wisdom of crowns, right? Because obviously they seem to be interactive, which means they should be, uh, the diversity of opinions might be reduced by this interaction or independency might be also reduced, right? So they are interactive. But key insight from the honeybees is that. Actually, Wagguri dances, even though wurri dances guide. Where other bees should inspect, right, but this vari dance does not impact. How each followers evaluate the the visiting site. So if you are, if you're following, following B, you know, you, you might go to the direction suggested. Once, once you visit the, the direction, basically it's, it's you who um evaluate the, the quality of the nest and then. It's mostly like independent, independent from the social impact. So, this combination of Uh, exploration being shaped by social interaction, whereas like, Investigation or like evaluation or belief updating itself is still independent. So this com this like a right combination uh is the kind of primal uh primary source of uh the driver of honeybee swarm uh collective intelligence. I guess this lesson should be applicable widely to uh non-honeybees, uh, societies as well. So I guess humans. Yeah, because social learning, if social learning, uh, is leading you to new exploration, uh, direction, but you can still, uh, learn things, decide things, uh, mostly independently from others' opinion. So, I think this, uh, any, any ways to, uh, Make, make sure, uh, any ways to, uh, if, if there's any way to, uh, promote this this type of com combination, this type of balance between independence and interdependence, then, uh, I guess, um. Yeah. That that this lesson should be, I guess, widely applicable and then hopefully including like a more political um dynamics or socio socioeconomic uh dimension as well.

1:10:42 Ricardo Lopes: Great, so just before we go, would you like to tell people where they can find you and your work on the internet?

1:10:52 Wataru Toyokawa: On the internet, yeah, you can find our work, uh, by, by visiting Tokyo and you can like, uh, just, uh, send me an email and you're more, more than welcome to visit us in Tokyo, but otherwise, you can visit, uh, yeah, maybe that website. Uh, WE have a website and we usually update, uh, what's going on. So, yeah.

1:11:15 Ricardo Lopes: OK, great. So Dr. Toyokawa, thank you so much for taking the time to come on the show. It's been a fascinating conversation.

1:11:24 Wataru Toyokawa: Thank you, God. It was fun.

1:11:26 Ricardo Lopes: Hi guys, thank you for watching this interview until the end. If you liked it, please share it, leave a like and hit the subscription button. The show is brought to you by Enlights Learning and Development done differently. Check their website at enlights.com and also please consider supporting the show on Patreon or PayPal. I would also like to give a huge thank you to my main patrons and PayPal supporters, Perergo Larsson, Jerry Mulleran, Frederick Sundo, Bernard Seyaz Olaf, Alex, Adam Cassel, Matthew Whittingbird, Arnaud Wolff, Tim Hollis, Eric Elena, John Connors, Philip Forrest Connolly. Then Dmitri Robert Windegerru Inasi Zu Mark Nevs, Colin Holbrookfield, Governor, Michel Stormir, Samuel Andrea, Francis Forti Agnun, Svergoro, and Hal Herzognon, Michel Jonathan Labrarinth, John Yardston, and Samuel Curric Hines, Mark Smith, John Lere, Tom Hammel, Sardusran, David Sloan Wilson, Yasilla Dezaraujo Romain Roach, Diego Londono Correa. Yannik Punteran Ruzmani, Charlotte Blis Nicole Barbaro, Adam Hunt, Pavlostazevski, Alec Baka Madison, Gary G. Alman, Semov, Zal Adrian Yei Poltontin, John Barboza, Julian Price, Edward Hall, Edin Bronner, Douglas Fry, Franco Bartolotti, Gabriel P Scortez or Suliliski, Scott Zachary Fish, Tim Duffy, Sony Smith, and Wiseman. Daniel Friedman, William Buckner, Paul Georg Jarno, Luke Lovai, Georgios Theophanous, Chris Williamson, Peter Wolozin, David Williams, Di Acosta, Anton Ericsson, Charles Murray, Alex Shaw, Marie Martinez, Coralli Chevalier, Bangalore atheists, Larry D. Lee Junior. Old Eringbon. Esterri, Michael Bailey, then Spurber, Robert Grassy, Zigoren, Jeff McMahon, Jake Zul, Barnabas Raddix, Mark Kempel, Thomas Dovner, Luke Neeson, Chris Story, Kimberly Johnson, Benjamin Galbert, Jessica Nowicki, Linda Brendan, Nicholas Carlson, Ismael Bensleyman. George Ekoriati, Valentine Steinmann, Per Crawley, Kate Van Goler, Alexander Ebert, Liam Dunaway, BR, Massoud Ali Mohammadi, Perpendicular, Jannes Hetner, Ursula Guinov, Gregory Hastings, David Pinsov, Sean Nelson, Mike Levin, and Jos Necht. A special thanks to my producers Iar Webb, Jim Frank, Lucas Stink, Tom Vanneden, Bernardine Curtis Dixon, Benedict Mueller, Thomas Trumbull, Catherine and Patrick Tobin, John Carlo Montenegro, Al Nick Cortiz, and Nick Golden, and to my executive producers, Matthew Lavender, Sergio Quadrian, Bogdan Kanis, and Rosie. Thank you for all.