8. Paul Smaldino: Cubist chickens, formal models, and the psychology curriculum

Paul Smaldino is an Associate Professor at the University of California, Merced. His research focus is broad and includes cultural and social evolution, cooperation, and philosophy of science.

In this conversation, we focus on Paul's recent papers on modelling, which I have found very useful in my own attempts of getting started with creating formal models.

BJKS Podcast is a podcast about neuroscience, psychology, and anything vaguely related, hosted by Benjamin James Kuper-Smith. New conversations every other Friday. You can find the podcast on all podcasting platforms (e.g., Spotify, Apple/Google Podcasts, etc.).

Timestamps
0:00:20: The parable of the cubist chicken & the need for formal models in psychology
0:15:48: Why do psychologists not use formal models more?
0:26:23: Models specify the relationship between variables
0:40:02: What is the difference between a formal model and a theory?
0:50:46: If we add formal modelling to the curriculum, what should we take out?

Podcast links

Paul's links

Ben's links


References
Eisenberg, E. M. (1984). Ambiguity as strategy in organizational communication. Communication monographs.
Gigerenzer, G. (1977-present). Everything he ever wrote. Every Journal he ever published in.
Kauffman, S. A. (1976). Articulation of parts explanation in biology and the rational search for them. In Topics in the Philosophy of Biology (pp. 245-263). Springer, Dordrecht.
Muthukrishna, M., & Henrich, J. (2019). A problem in theory. Nature Human Behaviour.
Rabin, M. (2013). An approach to incorporating psychology into economics. American Economic Review.
Smaldino, P. E. (2017). Models are stupid, and we need more of them. Computational social psychology.
Smaldino, P. (2019). Better methods can't make up for mediocre theory. Nature.
Smaldino, P. (2020). How to translate a verbal theory into a formal model. Social Psychology.
Smaldino, P. (2020). How to Build A Strong Theoretical Foundation. PsyArXiv.
Wimsatt, W. C. (1972, January). Complexity and organization. In PSA: Proceedings of the biennial meeting of the Philosophy of Science Association (Vol. 1972, pp. 67-86). D. Reidel Publishing.

  • [This is an automated transcript with many errors]

    Benjamin James Kuper-Smith: [00:00:00] Actually, maybe can we start with, um, how does, if we want to understand how to do mathematical modeling, how does cubism and chickens help us get that conversation going? 

    Paul Smaldino: Uh, okay. Yeah. Um, well, I, I, the, so I, you're referring to in, in my, uh, 2017 paper, uh, models are stupid and we need more of them. I have this, uh, part in it that's called the Parable of the Cubist chicken, and it's not so much a, this is how you do modeling point.

    The point is, is about precision and, and therefore, you know, uh, kind of making the point that models are useful because words are imprecise. So this was a true story. Uh, a very silly story that happened when I was in college. Uh, so I was probably about, I don't know, 19 or 20. And, uh, a friend and I were [00:01:00] messing around and we were like, let's just say definitely not sober.

    And we went to collect a third friend of ours who was in a play, and we had already seen the play, so we didn't need to go watch his play again. We were just waiting for his play to end so we could take him and go off in an adventure, uh, which would probably And 

    Benjamin James Kuper-Smith: continue not being sober. 

    Paul Smaldino: Yes. And continue to be even less sober.

    And, uh. Um, there was a, this was at Wesleyan University in Connecticut, and in the, in the old 92 theater, there's a basement with couches and toys and shenanigans and all kinds of theatery things. And there was a big bucket of legos and just kind of absentmindedly while we were talking, waiting for our friend.

    We were, I was, you know, playing around with Legos, just, just putting Legos randomly together as like a thing. So is 

    Benjamin James Kuper-Smith: that a normal thing to have Legos in a theater waiting room? 

    Paul Smaldino: I mean, I, I I have no idea. Honestly. I, I hope so. I've 

    Benjamin James Kuper-Smith: never been to the, yes, I don't know, maybe in Europe [00:02:00] it's not really the normal, 

    Paul Smaldino: no, I don't, I don't think it's necessarily necess, I think it's more just like knickknacks and sort of, uh, things to play with or improvise with.

    So, uh, maybe Legos are useful in that regard. I haven't spent that much time in theaters, so at, at least in, you know, backstage. But, um. I'm, I, I learned very quickly that I'm a very bad actor. Uh, you tried 

    Benjamin James Kuper-Smith: or 

    Paul Smaldino: I've, I've, yeah. I tried. I tried in my youth. Uh, I'm, I'm not good at it. Um, I don't know what it is too self-conscious maybe.

    But, uh, but we were futzing around with the Legos and, uh, I, I honestly can't remember. And I, I've asked my friend, um, who's now a, a film editor in our, in LA if, uh, I said, you know, do you remember? He remembers the story. I said, do you remember which one of us first pointed out that it looked like a cubist chicken?

    And he said, no, I don't remember. Um, so one of us looked at this kind of smash of blocks and just [00:03:00]kind of were like, if you squint at it, we're like, look, it's like a cubist chicken. And the other one of us said. Oh, you're right. It is totally a cubist chicken. Like look at the Cubist chicken and being not that sober, we were, I think, more pleased with ourselves than we had any right to be in this case.

    But we like, and, and it, it would've ended there and just it been forgotten to the time, like, oh yeah, there was this time where for a couple minutes we talked about how one of us had made a cubist chicken out of Legos. Except that shortly after our, our friend that we'd been waiting for came down and said, Hey guys, what's up?

    And uh. One of us held up the, the mash of blocks and said, look, it's a cubist chicken. And he kind of, he was totally sober having Justin in a play and just gave us this kind of like raised eyebrow, like how is that a cubist chicken? 

    Benjamin James Kuper-Smith: Yeah. 

    Paul Smaldino: And I just said, well, look, here's the, the head and the, there's the, the body and here are the feet.[00:04:00] 

    And my friend who for the last 10 minutes had been agreeing with me that it was in fact a Cuba's chicken. Just got very upset and said, no, no, no. The whole thing is just the head and this is the beak and these are the eyes and this is the crest. 'cause it was a rooster in his mind. And, and we realized, and it that we had been.

    Using the same words and calling it a cubist chicken and agreeing for 10 minutes that it was in fact a cubist chicken. And as soon as our third friend had asked us to explain exactly how it was a cubist chicken, the whole thing fell apart. And we realized we were talking about completely different things.

    And, and the reason this story stuck with me is because I, I think of it as an analogy for something that happens all the time, is that we use words in ways that feel familiar, that, uh, allow us to feel like we agree with each other. [00:05:00] Because a lot of words are vague. Like cubism is vague, chicken is pretty clear, but like how something can be a cubist chicken.

    If you say, if I just told you, imagine a cubist chicken. Everybody listening to this is imagining something slightly different. But as soon as you say, explain to me exactly how the cubist chicken is made and what the parts are, all of a sudden we have to be very precise and. I thought about this a lot after, um, I be got more into, you know, becoming a scientist and spending time in psychology and social science departments and realizing that a lot of the times the words we use are, are very vague and we use words like identity or emotion, anger, fear, behavior.

    Right? Uh, all kinds of sort of psychological or social constructs, the economy, right? They're [00:06:00] vague words and there's ambiguity and it allows us to. Talk to each other using the same words and think that we're talking about the same thing. Whereas in fact, in our own minds, we're thinking about completely different things.

    And this leads to problems when we try to do science right. I think there's, you know, there's ambiguity in language for a very, you know, many reasons, some of which are adaptive, but 

    Benjamin James Kuper-Smith: by way, what, what might some of the adaptive reasons be? 

    Paul Smaldino: Well, um, I think it's a very efficient, for one, uh, in, on the one hand, if I can just sort of give you an impression of the sort of category of things that you need to deal with without having to specify exactly what I'm talking about.

    Um, also it allows. Um, sort of differentiated groups of people to coalesce around a common cause because they can all view the cause slightly differently, but use the same word for it and [00:07:00] therefore internalize the meaning. And I think it's actually really useful in coalition building. It's also useful, you know, for let's say slimy politicians to make vague promises and say, you know, we're gonna use, let's say, well, uh, I, I don't really nec need to go into any specific examples, but, um, you know, I'm gonna do this and use ambiguous language and, and sort of refuse to answer when they're asked.

    What exactly do you mean? So it allows lots of different, let's say, voters to say, oh, they're going to do something that I like because I'm going to interpret their language to mean this. Whereas someone else can also say that they're gonna do what I like because I'm gonna interpret their language this way.

    And both of them are right because there's ambiguity. In the language, but also both of them are wrong because neither of them have nailed exactly what is going to happen because there is no precision in the promise. Um, so [00:08:00] there's some really great literature on this, uh, at least starting in the 1980s, uh, from communication.

    There's this really wonderful paper by, um, Eisenberg, I think, um, call it something like ambiguity as a strategic, uh, communication strategy. It's, it's, I'm sure it doesn't use strategic and strategy. I'm sure it's a better title, but, um, 

    Benjamin James Kuper-Smith: yeah. 

    Paul Smaldino: Uh, yeah. Um, but when we're doing science right, we want to be say, this is what the world is like, and we're going to form hypotheses and theories that are based on com, uh, particular deconstructions of the world and say, well, if this is the case, then this will happen.

    And if this other thing is the case, then this will happen. And that's, that's kind of what a lot of science, that's at least the goal of a lot of science. And to do that, we have to say, I mean, some of the goal science is just description to say this is what this [00:09:00] is and this is what this is made of. But in either case, right, we need to be very precise about what those things are, what we are talking about, and not allow ourselves to be bogged by imprecision.

    But imprecision is rampant in say, a lot of psychology and the social sciences and probably other fields as well. But, uh, 

    Benjamin James Kuper-Smith: I, I feel like one, one assumption I've always had, or intuition is that in psychology it's much worse because we talk about concepts that we study, concepts that we also use as humans in everyday speech.

    I feel like in physics, maybe if you're talking about neutrinos or whatever. You know, it has a much more precisely defined meaning than if you talk about emotions or whatever that you might use every day in all sorts of contexts. 

    Paul Smaldino: Oh, absolutely. Um, and then there's a, there's a couple reasons, uh, for that.

    I mean, one is exactly what you're talking about, that everyone sort of feels like they have a claim to already some level of expertise, which in some, to some degree we do, right? [00:10:00] 'cause we are humans and we know what it's like to be a human and be around other humans and live in society. But you're right, both psychology, I would say the social sciences too.

    I mean, if you're, you know, a sociologist or an anthropologist, or an economist, you're using words like, you know, culture or, you know, I mean, just, we could stop there, right? There's nobody ask, ask a hundred people what culture means. You get 150 answers and. You know, psychologists. Absolutely. 'cause we're dealing with mental states, right?

    What it feels like to be something and what it appears like and what it seems like. And, um, it's not always clear that the way we sort of cognitively parse the world is in fact, you know, the best way to describe it. Precisely. Um, another issue, right? Is that, you know, there's this mismatch between measurement and theory in a lot of the sciences, basically all the social sciences and almost all psychology, except [00:11:00] for maybe, you know, uh, some of the more clear cut cases of like neurophysiology or behavioral science or psychophysics, where there's a mismatch, right?

    Or, sorry, not a mismatch, but a, an imprecision to. The mapping between the theory and the measurement. So this is what some philosophers call the, you know, the difference between the exact and the inexact sciences. And it's probably not quite such a dichotomy, but more of a continuum. But you know, if we're doing physics, let's say, a lot of physics, the theory is about the thing that is directly measurable.

    It's about that, right? So the theory is about how force and mass and acceleration are interlinked. The theory is about how charge and fields inter are, are affect one another, right? The theory is about how temperature and pressure are interlinked, right? [00:12:00] These are the, that's what the theory is about. And those are the things that we are measuring.

    And notice that, you know, any physics equation, all the components, all the, all the little letters, there's a unit of measurement associated with them all. That is not the case in psychology or the social sciences or even a lot of biology, right? It's, we're dealing with much more abstract concepts. There's an inex exactness to the, the, the mapping between what we want to talk about, like fear or consciousness.

    Well, not consciousness is its own, uh, pan of worms, but let's say, you know, memory, right? Or, uh, I don't know, even behaviors, certainly things like culture, cooperation, um, you know, disgust, whatever. Um, we have folk concepts, right? We have an idea in our head of what kind of that means to each of us as an individual.

    But then you say, well, how do you [00:13:00] measure fear? How do you measure generosity? How do you measure willpower or patience? Right? Well, you say, well. But I don't know, it's like this whole gestalt of different kinds of amalgams of behaviors and perceptions. So I'm going to operationalize it in, let's say an experimental context or some, you know, uh, some data collection where, okay, well I have this, you know, data that's been collected from some survey and it allows me to sort of, kind of get at the kind of thing that I'm kind of talking about.

    And it, it just makes it much harder to do this kind of science in general. And, and this isn't actually a, about models per se. This is like sort of a, a, a just a general limitation to trying to do science on, on complex emergent phenomena that involves, you know, a lot of organization [00:14:00] like, so. I was really influenced by, uh, the philosopher, bill Wimsatt, um, who was my sort of, I sort of hate this terminology, but my academic grandfather, right?

    He's my advisor's advisor. Um, uh, you know, he has this really wonderful paper from the seventies called Complexity and Organization, and he kind of talks about, you know, basically the difficulties in trying to do theory on anything complex. And he was really talking. He was, he wasn't at that time really interested in psychology per se, or, or behavior.

    He was, you know, interested in, in things like evolutionary biology and developmental biology. But even then he was saying, you know, compare a rock and a fly. You know, the parts of the rock are, are basically pretty straightforward and the way it's organized are, are, it's kind of uniform. There's a crystalline structure maybe, and maybe there's, you know, different elements that work together.

    But, uh. Organizationally, it, it's pretty easy to pin down. Whereas a fly has all these different systems that are, and you can describe it at [00:15:00] many different levels, right? The cellular level, the hormone level at the organ, like the, the level of the organs, the level of the, the, the, the body parts, what their function is, what their behavior is, how they interact.

    And there's both up, you know, sort of bottom up and top down interactions so that what the fly sees, let's say in the world interacts its physiology and then the physiology affects how it sees things. And all these things make it very, very difficult, um, to, to, to make sort of concrete theory. It doesn't mean it's, it's, we shouldn't try it.

    Just, I think just sort of like being aware of and paying attention to these kinds of complexities should make us a bit more careful as scientists. 

    Benjamin James Kuper-Smith: Yeah. I mean, one thing I was wondering also, when. Reading one of your, I can't remember which one it was, papers, is that at some point I wonder like, how did we get here in terms of psychology?

    Like why, like, why is this so bad? Is [00:16:00] it just because that it's so difficult that people say, well, we're just not gonna do that then, or I mean, there's of course many reasons why. Um, but yeah. Do you think it's just like at the, the complexity seems so high that people said, okay, let's just try a different approach almost.

    Paul Smaldino: I, I think there's many reasons why, as you say, I, I think like there's, I can, so two things immediately come to mind. One is that one just comes back to exactly what you were talking about before, which is that because we are humans, we sort of feel like we have a claim to understanding human behavior without needing to, so, you know, like.

    If someone says, I'm a physicist, and someone says, well, I don't, you know, uh, it's gonna be difficult for me to explain what I do. You, you, most people feel very comfortable. Like, yeah, I, I, I admit it that I, that I accept that you're a physicist. Physics is very complicated. I don't have that kind of expertise, [00:17:00] so I'm gonna sort of, you know, defer to you or understand that it would take a very long time and be very difficult to understand things at the level that you do.

    If you say, I'm a psychologist and I understand how memory works. People say, well, I have memory. Tell me how it works. I should be able to understand that right away. And, and I think that that sets up a, you know, uh, two sides to this system where, first of all. People sort of come into things without the expectation that it's going to be as hard as it needs to be.

    So people say, well, I, you know, uh, I should be able to, as a first year graduate student be doing cutting edge research and be coming up with really amazing research questions and getting published in top journals because I already, you know, have the ability to form good questions about memory or learning or behavior.

    Because I'm a person, whereas like 

    Benjamin James Kuper-Smith: no one, I'm, I'm still in that phase. 

    Paul Smaldino: Yeah, right. And, but, and this is it, this is a, [00:18:00] it's culturally indoctrinated in, in, in the pedagogy of psychology in other fields. Right. It's, it's very common for students to think that, and for professors, professors to encourage them to think that way, you know, uh, to, yeah.

    By the end of your first year you should be submitting papers to top research journals. Whereas if you go into like, you know, biochemistry or high energy physics or cosmology or whatever, nobody would think that. Nobody thinks that in your first year you should be doing and producing leading. Research, maybe, maybe you'll be on a project that's cutting edge, but nobody thinks you should be leading a project that is cutting edge in your first year because it, it's, it would be crazy.

    You have so much to learn 

    Benjamin James Kuper-Smith: somehow. Actually, I thought it was almost the other way around because I guess like, I dunno much about physics and not, like, I don't read much about popular physics, but I, the only, like, one of the few things I know is that like the quantum revolution was done by people in their twenties or something.

    Wasn't all of that, [00:19:00] like a lot of the, the breakthroughs in the early 20th century came from people who were younger than 30. Right? 

    Paul Smaldino: Uh, that is true. Um. A couple things. So 

    Benjamin James Kuper-Smith: somehow, yeah, my assumption was that 

    Paul Smaldino: I, I think there's a couple things about that. One is that it, it, it is sort of the outlier case.

    The other is that, you know, at that point there, people in their late twenties are often will have been doing this, you know, sort of consistently for a while. And the other is that we're talking about theory, right? And math. Yeah. 

    Benjamin James Kuper-Smith: Yeah. 

    Paul Smaldino: And so they're, they're coming in and, and basically looking at how mathematical functions work and, and a lot of really cutting edge work in mathematics is done by young people.

    And there's, there's a lot of sort of speculation about why that is. You know, it, it may just be that at, at, that's the age in which you can completely focus on some sort of crazy abstract thing, um, and not worry about anything else in the world. Um, but yeah, no, that, that's a fair point. [00:20:00] Um. So I, I think that the other side of, of, of, of that same issue with, with something like psychology is that people who are not scientists have, are interested in knowing about human behavior in ways that they're less interested in learning about physics or biochemistry or the behavior of insects or fruit flies or something.

    Um, and so you can write a popular, like how many popular psych books are there on the bookshelves now? How many came out this year? Right? Like re an absurd amount. Uh, and because there's money to be made and telling people things that are either both counterintuitive or totally intuitive and supports what they thought all along and helps them bolster their argument, um.

    And so people love these kinds of books, and so there's money to be made. And this, this, this also helps [00:21:00] elevate people who, let's say, aren't necessarily doing the most, um, rigorous work because by being kind of vague and ambiguous and imprecise, but you know, they can become salespeople and they can become very successful in this way.

    And because academic success and popular success, while there's not a one-to-one mapping, they're linked. Right? If universities love to hire people who have bestsellers. Um, so, uh, the other thing is, is historical. Um, I mean, only thing about like, you sort of related to that point, like thinking even about someone like Sigmund Freud.

    Um, and I, my, I have a, you know, I have a soft spot for Freud because my, my dad, it was a, uh, psycho psychoanalyst and 

    Benjamin James Kuper-Smith: Oh really? 

    Paul Smaldino: Uh, he used to teach Freud at the New York Institute for psychoanalytic training. Um, so I have a soft spot for Freud a bit, but Freud, you know, it was started as a neurologist and, uh, [00:22:00]wanted to, but he wanted to talk about consciousness and, and emotion and, and all that kind of stuff.

    And, and he sort of quickly realized that. There was no way to make a path between neuroscience, especially as it was at the end of the 19th century. And consciousness in memory, there was just no way to, to, to make that link at that time. So he just said, well, forget it. I'll just, I'll, I'll become much less precise and I'll start using metaphor and, and, and, you know, obviously become extremely successful and influential in this way.

    Um, there's other things, right, which is, you know, and related to all that, like a lot of the early psychologists were extremely rigorous and, and, and scientific. I mean, they were all crazy because all early scientists in any field are crazy because you have to be crazy to start a totally new field of science, especially at a time where science was much less institutionalized as it is now 

    Benjamin James Kuper-Smith: by, uh, Lisa coaches, what, what are you talking?

    Paul Smaldino: Well, I [00:23:00] mean, I'm thinking of, well, uh, a few people come to mind, but like, uh. Uh, Villa comes to mind. Mm-hmm. Um, I mean, that guy's, that's 

    Benjamin James Kuper-Smith: psychophysics. 

    Paul Smaldino: Yeah. The early psychophysics, but like, if you read about v he's insane, right? He, he was, you know, he got hired as a philosophy professor at Leipzig and he kept asking for a lab, and they did.

    They finally, after years of asking. Gave him like a closet and he purchased all his own equipment and was just doing this all in his spare time. And I, I found there was some, uh, retrospective, I found that, that published his teaching schedule for his first five years at Leipzig. And the, the, he taught, um, cosmology and philosophy and anthropology and psychology.

    Right. It's like, it's, it's insane. And, and he was just like, he was, you know, part of his research was psychophysics, but Fardo was just asking people like I. What'd you see? What did it feel like? [00:24:00]Right? And like volunteerism. Um, but you know, he was writing things down and trying to be very precise. 'cause he didn't, there was, that's how you, he had to convince other people that this mattered.

    Benjamin James Kuper-Smith: I mean, so was he trained as a doctor physicist? 

    Paul Smaldino: You know, that's a good question. And I don't know. Okay. '

    Benjamin James Kuper-Smith: cause I feel like those are probably the two. Oh, and philosophy, right? Or probably the three reasons from which people came to psychological. 

    Paul Smaldino: I think probably philosophy, but I'm not totally sure. Uh, possibly medicine.

    I don't know. Uh, yeah, you're right. There was a lot of, um, mismatch or, uh, not mismatch. I was gonna say mishmash, which is apparently a word I use often. Um, uh. 

    Benjamin James Kuper-Smith: It's a good German word. 

    Paul Smaldino: Um, the, you know, but then, you know, sort of American behaviorism, uh, like Watson and, and Skinner, um, you know, and then especially, uh, Tolman, [00:25:00] Edward Tolman, like those early like behavioral things, like co all the cognitive map stuffs, it's incredibly precise what they're doing.

    And, um, you know, a lot of the, the behaviorists were the, the first people. They, they, they were extremely rigorous. And some of the first mathematical models in psychology came out of behaviorism, right? Like where Gola and Wagner in the 1970s is, is great. And that leads you directly from risk Orland Wagner in psychology to people like Sutton and Bartow in computer science who.

    Formalized it into reinforcement learning, which is now a com key component of machine learning. People who learned that then became neuroscientists, like Wolf from Schultz, and then discovered, you know, the prediction error system in the dopamine system. And it's like, I think of there as being like a direct line from one to another.

    And you know, it's really sort of, I, I think sort of the power of formalism and precision is that you can take this thing like, here's how it works. Oh, well if it works that way, [00:26:00] then we can program that algorithm into something else. Well, if we're familiar with that algorithm and we know it works, then maybe the brain is actually doing it.

    Well, look, we found it in the brain. Um, 

    Benjamin James Kuper-Smith: yeah. 

    Paul Smaldino: So, yeah, but I don't, I, I don't know. That was a very long-winded answer to your question, so I We can move on. 

    Benjamin James Kuper-Smith: Yeah, exactly. I've always forgotten. Oh yeah. How we got that. Secondly, yeah. So maybe one thing I found interesting in reading the, I think this is in models are stupid and we need more of them.

    The paper that I found interesting that I somehow never thought about in that way is that you, you said like that the point of a formal model is to just specify the relationship between variables and like, I dunno why, but somehow when I, you know, when I read that I thought like, well, yeah, of course. But then I realized like somehow in the way I've been thinking, you know, as like someone who studied psychology somehow.

    I don't know in that, that precise way I never thought about it, which is really seems like a bizarre omission in my thinking. Yeah. Sense. 

    Paul Smaldino: I think, I think it's, I think it's very [00:27:00] common. I mean, I think that, you know, we, we come into things and especially like with something like psychology, but we sort of have this idea that we know how the world is made.

    We know how, what the world is made of and how it's made and how it works. And all we have to do is, is figure out, you know, what exactly the things that we know about are, like how, how exactly they're constructed and then how they affect each other. But I think a huge part of science is, is figuring out exactly what's in the world to begin with and, uh, and how those things relate to each other.

    And, and I think I was extremely influenced. And when I was a grad student, I read this paper from the early 1970s by, uh, Stewart Kaufman. And so Stuart Kaufman, I don't know if you, if you're familiar with his work, he's, 

    Benjamin James Kuper-Smith: I don't think so, 

    Paul Smaldino: kind of legendary in complexity science and complex systems. He was kind of an early person at the Santa Fe Institute and did really all this extremely influential work on, um, rugged fitness [00:28:00] landscapes and, uh, epistemic gene networks.

    So, you know, thinking about like the, the emergence of structure and evolution, um. He has this, uh, he, he started as a medical doctor and then sort of switched to sort of this, you know, computational complexity science. And, and on the way there, or very early in his career, he published this paper in, in, um, uh, in, in the conference proceedings that eventually became the journal Philosophy of Science.

    And, um, and it has some terrible title that I can never, it, it's called, I think, on the articulation of Parts in biology and the Rational, rational search for them. Uh, good title. Yeah, it's a terrible title, but it's an, i I, I think of it as like the, to me it's like the most important philosophy paper in my arsenal.

    Um, and Oh, really? 

    Benjamin James Kuper-Smith: Okay. 

    Paul Smaldino: Uh, what you 

    Benjamin James Kuper-Smith: ci it in here? 

    Paul Smaldino: Uh, yeah. I cited in, in almost everything sort of philoso philosophical that I 

    Benjamin James Kuper-Smith: read. Oh yeah. Articul jps explanation in biology. It's [00:29:00] even worse than you said here. I'm just gonna highlight that then. Okay. 

    Paul Smaldino: Yeah. So, I mean, what, what Kaufman. Points, the paper is really about hypotheses.

    And so I think the point that is, you know, that I take from him and I, you know, this is sort of Kaufman then filtered through me, um, is, you know, most of us are very familiar in our training with the idea of science as hypothesis testing. Like we, we come up with hypothesis. Like this is sort of the, you know, the grade school version of Francis Bacon.

    Like, oh, science. The scientific method is we come up with hypothesis and then we test the hypothesis and we, you know, the evidence either supports or doesn't support the hypothesis. Um, Kaufman's point is, well, where do these hypotheses come from? What, how do you, how do you make a good hypothesis? And, and it would be a very bad idea.

    And, and [00:30:00] in psychology it's a very common way to do things, but, uh, it would be a very bad idea to just say, I'm just gonna make it up. Right, based on nothing based on my sort of intuitions. Now obviously, I, I do believe in expertise and I think that over time experts do gain an intuition for things and, and should be able to trust their intuition to some extent.

    But, um, the point here is not about, you know, how rigorous or how much of an expert you need to be to, to, uh, make a hypothesis. The point is that what a hypothesis is, is the articulation of parts and what that by, by this he means. And this is also, uh, an idea that, uh, herb Simon, at the same time was talking a lot about, um, and also Bill Wimsatt, but sort of all these, there's all these great papers in the late sixties, early seventies on this kind of stuff.

    But, um, the basic idea is that. If you wanna form a [00:31:00] hypothesis about some system, you need to have a question that drives what you want to know about the system. And in pursuing that question, you have to decompose the system into parts. And I, and I, what I've added is, and the relationships between those parts, and there's no single correct decomposition.

    I mean, maybe this, the, I mean, probably the only single correct decomposition is to say, well, everything is corks or strings or in brains or whatever. You know, the, the physicists say we're all made of at the, at the fundamental level, but like beyond that, right? We're talking about emergence and we're talking about things in the world that are made up of other things.

    And so any particular theory abstracts away a lot of levels of organization. And says, well, I'm gonna focus on the [00:32:00] neurons, or I'm gonna focus on perception. I'm gonna focus on memories. I'm gonna focus on physiology. I'm gonna focus on behavior and these behaviors and not these other behaviors or these organs and not these other organs or these aspects of perception or context or whatever it is, and not other things.

    I'm going to decompose the system into the parts that I think are most relevant to my question and think about how those parts, what are those parts made of? What are the properties of those parts? How do they affect each other and relate to each other? And then that allows me to construct a hypothesis.

    The hypothesis, any hypothesis is about the parts, is a story about what the parts are, what they're made of, and how they relate to each other and like, and what they do or don't do. And so. Particular system. There are usually multiple ways to decompose it into different sets of parts and relationships.[00:33:00] 

    Exactly what is sort of the right decomposition depends on the question. And it's sort of a great decomposition is one that really shines light on the question and allows us to get more information out of it that allows us to make better predictions, let's say, or have deeper understanding. And, you know, I think often, sort of what I think Kon would call paradigm shifts, shifts, uh, is the introduction of a new decomposition, a new articulation of a system into parts that allows new kinds of questions or better questions to be asked about it, right?

    When we, when we went from, you know, uh, Newtonian mechanics to quantum mechanics, right? We're decomposing the world into different kinds of parts and we can ask new kinds of questions about it Now. The reason Newtonian physics is still useful is because that decomposition still allows us to answer questions in a useful way, [00:34:00] and depending on the, the questions we ask, right, decomposing the world differently, uh, gets us different places.

    So here's what I think was so cool about the Kaufman Paper. In addition to all about that is at the end he says, well, and once we can do this, once we can decompose our system into parts, and we've articulated what the parts are, what their, what their properties are, and how they work. And, and we are precise then about our hypotheses.

    We can encode our hypothesis in a, what he calls a cybernetic model. He uses cybernetics 'cause that was the big thing back then. But what he's basically talking about is a computational model. Um, and this is like 1970. He's writing this, um, before we, the personal computers at that time weren't even invented.

    Benjamin James Kuper-Smith: Yeah, yeah. The one part that I found really useful, just for me personally, [00:35:00] um, relates to, um, part of what you just said, and also the, basically the title of the models are stupid. Namely the part that you have to, you know, choose, you know, you can, you can decompose it in all these different kinds of ways.

    And, but you have to choose the specific ones that are most relevant for the question you want to ask. And again, this is very obvious once you state it like that, but like, I think one thing especially I, is I do some social interactions and corporation that kinda stuff in my main work. And I often find myself just adding stuff thinking like, well, this also has an effect.

    This has an effect. That has an effect. But, and then I found it just really, how should we say? It's almost freeing to know like, yeah, you don't have to deal with everything. Like just focus on the things that explain the thing you're trying to understand. 

    Paul Smaldino: Yeah. Uh, I think that's, I mean, I think, you know, the, the frustrating thing about doing this, this kind of work on, you know, especially with social behavior is that.

    It [00:36:00] does all have an effect, right? It does all matter to, to, to some, to some extent. But, you know, when we, when we try to build theories, you know, what we wanna kind of do is say, well, what, what is the, what is the thing that matters the most? What are the things that matter the most? What are, what is the sort of things that, you know, fundamentally structure the interactions and provide sort of the, the foundational structure?

    Um, or, and you can almost think about this as infrastructure. Like what are the things that constrain and provide the affordances to the behavior the most? And then you can start adding things. Well, but if this is also the case, then it changes things this way. Or this is a new affordance that people have, then they can do things this way.

    And, oh, there's, you know, let's, if they come from this culture, then, you know, perhaps then it would be more adaptive to do that. Or, you know, whatever it is, you can start. Adding things, you know, if they're sick or if they have a mental illness or whatever, you can start adding things and changing [00:37:00] things.

    But what you want at, you know, to start with is a, is a foundation of what are, what are the key, what, what is sort of the key aspects of, of this behavior, this scenario that, you know, what, what is the, what is the situation here demanding of people and, and, you know, what are the, what I, I think this is also a call to, or, uh, to, to think about things more instrumentally than a lot of psychologists do.

    Um. 

    Benjamin James Kuper-Smith: Which maybe more instrumentally. 

    Paul Smaldino: Uh, what I mean is, and I, and I, you're probably already doing this to some extent when you're thinking about things like cooperation, but there's a tendency I think I've seen, at least in psychology, uh, which is something that, that I think separates it and creates an artificial divide between psychology and the other social sciences to focus on the internal perception of something, the phenomenology of something and stop [00:38:00] there.

    I'm not saying these things aren't important, but when you're studying things like identity or emotion or whatever, people will study emotion by asking people how they feel or saying, well, how did you feel when, when that happened? Or, imagine a time when this is how you felt. Um, things, people will study identity by saying things like.

    Who do you identify with? What do you, how do you feel? Like what is, you know, what, what identity feels appropriate? And those things are, that research is, is, is potentially useful and can be great, but it needs to be tied to something that, uh, to the kinds of questions that ask, well, what is the function of these?

    You know, why would, what would, what would it mean for behavior to feel that way? What, what purpose does an emotion serve? Maybe it's not adaptive, but at least what happens when this [00:39:00] emotion is, is present. What, like the emotion is just, uh, or the, the phenomenology is just what we call it internally, right?

    But, but what behaviors are out in the world, what behaviors it produces, what classes of behaviors it does are those are, I think the things that really matter. In part because they are the ones that, you know, that actually matter to other people in terms, you know, what we really care about is what people do, not what they feel.

    Um, we, we care about how we, everyone cares about how they feel. Yeah. And what other people do. Right. Um, 

    Benjamin James Kuper-Smith: yeah. 

    Paul Smaldino: But you know, the other thing is that like it's, if we want know why people feel particular ways and, and other ways, right? We have to think about the, the, the sort of social forces and the evolutionary forces and the cultural forces that create minds that do things and not other things.

    And that requires understanding the, the consequences of those internal [00:40:00] psychological states. 

    Benjamin James Kuper-Smith: Except one question I had, which I wrote down, is, what exactly is the difference between a formal model and a theory?

    Paul Smaldino: Sorry, I paused to take a drink. Um, that's a great question and it's, it's, it's one that I, I, there's, there's some debate about, and, and I'm not sure that I'm gonna be a hundred percent convinced by my own answer here, but I'll do my best, right? I mean. The word, the word theory is to some extent fairly imprecise.

    Yeah. Uh, you know, and, and this is the issue, right? A formal model is something that is much more precise. So I'm, I'm more comfortable talking about, um, you know, a, a what a model is than what a theory is. But I think of a theory. I mean, a theory is an idea about how, what something is made of and how things work, how, uh, what things are made of and how they work.

    So you can have a theory that says, you [00:41:00] know, well, my theory is that when this, you know, if something hap something is the case, then you know, something will happen. That would be a theory. Um, more precise theories are better theories usually. Um, well, uh, another way to put it that seems, uh, oppositional, but I think it's complimentary, is more generalizable theories are better theories, and generalizable theories aren't necessarily more precise.

    Um. But you know, it's important there, there's this idea from sociology that has not migrated into psychology, but I, and I only recently discovered, but it, it's, uh, you know, the, the, the, the sociology papers on it are from the 1980s, uh, this idea of, of the importance of scope in what they call scope in, um, in dis in, in describing a theory.

    And the idea is basically, well, if you say, I have a theory about, um, I don't know, segregation or [00:42:00]polarization or whatever, uh, you have to say when your theory applies and when it doesn't apply, right. What are the conditions under which, you know, what are, what conditions need to be met for this theory to be applicable?

    And what, what doesn't? This was a, a, a big critique of, uh, in psychology of, let's say, the sort of condiment and diversity, uh, decision making. Um. Biases and heuristics literature that, you know, especially Ger Zer, uh, launch, right? I mean, this is, this was his, I think it's, it's his most compelling critique of that stuff, which was, you know, to say, well, you, okay, you say you have an availability heuristic or a confirmation bias or whatever.

    Well, there are cases in which these things are incompatible with each other, or, you know, having one heuristic would preclude having the other heuristic. So when do we expect one heuristic to be dominant, and when do we expect another heuristic to be dominant? And there's basically nothing [00:43:00] in the t risky literature that that tells you.

    The answer 

    Benjamin James Kuper-Smith: is that what is, is there like a single paper where gig answer questions? That, or almost, because I know that he's criticized them, but I don't, I've never actually read script. Probably. 

    Paul Smaldino: I don't know. I've read a lot of his papers. Um, and I can't, and he may have just, I, I, I attended a summer institute at his, uh, max Plunk, uh, in, in 2012.

    And he may have just said that out loud, honestly. Okay. Take it. But, uh, but I, it, I think it's in some, I don't know. He's written so many papers. 

    Benjamin James Kuper-Smith: Yeah. I'm also asking, because I always put the papers, we just, I always have references for these podcast that, 

    Paul Smaldino: okay. Yeah. Just, just read everything that Ger answered.

    Wrote everything that wrote, everything he wrote. Uh, 

    Benjamin James Kuper-Smith: by the way, I was just briefly talking about papers. Another thing I'll put in there is there's a good, uh, I thought a really interesting paper by Matthew Rabin on, it's a commentary or something. It's not, it's not an actual paper about, uh, I can't remember what it's called, but [00:44:00] it's.

    That that was also one of those, those, like, um, someone giving you a bit of a slap while you're reading it, uh, when he said like, you know, as you said, like a, a theory has to be broadly applicable. There's no point in having a theory about, like, I think I know his examples, like about a certain kind of farmer in a certain kind of year.

    You need a, you need a theory that applies to all humans kinda time. 

    Paul Smaldino: Yeah. And it's very, it's very, it's, it's difficult, right? Because what you want to do is to say, well, you know, I, I want my theory to be applicable to a particular situation. And any, and a theory that's applicable to a particular situation, especially when we're talking about human behavior is, is one that's probably not applicable to any other situation.

    You know, the more precise you can get it for one situation, the less it's gonna apply to anything else. Um. So there's this trade off, but also, I mean, I think the less you know about or the less formalized a, a system is, or a, you know, a, a type of interaction is that the more you need [00:45:00] sort of broad generalizable theory.

    Like if you're gonna like, so cooperation is a great example of this, right? You need to understand the, the simple prisoner's dilemma, uh, you know, two player game without any institutions or punishment or group structure and understand how that works. Before you wanna start building, you know? Okay, well now we understand how this works.

    Uh, how cooperation works when there's, uh, iterated interactions or when there's a population structure or kin structure. Oh, now we can add sort of contingent strategies like reciprocity. Um, what happens if we add, you know, a, a multiplayer game? What happens if we add the ability to punish non cooperators?

    And, you know, what happens if we add things like reputation? What happens if we add things like hierarchy or power dynamics? You know, those things are all important. You, you need to understand the baseline theory [00:46:00] before you can start making sense of the things that are more complicated. Yeah. 

    Benjamin James Kuper-Smith: Yeah. I guess that, yeah, that's true too.

    I mean, I think here is the thing that really like spoke to me when I read this is so, so I am, I am using the ic and in one of his footnotes, he wrote something like, um, and we certainly don't need another addition to the 40,000 explanations of how people behave in the DIC dictator game. Something like that.

    And then I thought like, oh, I am thinking way too much about the paradigm I'm using rather than the general thing and which this supposed to apply to. 

    Paul Smaldino: Yeah. Um, I, I wanna, I want to go, I want to, I have something to say about that, but I also want to go back to your question about theories versus models.

    Mm-hmm. Um, which is that, you know, the, uh, Michael Muha Krishna and Joe Henrik have this, this nice paper called A Problem in Theory, and they, they, they have this delineation, which I find useful between a theory and a theoretical framework. 

    Benjamin James Kuper-Smith: Okay. 

    Paul Smaldino: And, and I think, what 

    Benjamin James Kuper-Smith: is that? So, oh, I mean, yeah, I've read the paper, but I can't remember that distinction.

    Paul Smaldino: Yeah, I think [00:47:00] the, I think the idea is that, um, uh, a theory can be anything like what I was just talking about. Like, oh, uh, if, if X is the case, then Y will happen. And if Y is, you know, A is the case and Y will happen. If B is the case, then Z will happen. Um, but a theoretical framework is kind of a collection of theories.

    It is a, a, a sort of a, uh, and moreover, it's a sort of, uh, guiding set of assumptions and principles to say, well, we're going to, all of the theories that are built on this framework will have assumptions that are, that look like this, these are the assumptions that will go into all of our theories. These are the kinds of principles and uh, way things to guide, you know, the kinds of questions our theories will be about.

    And I think the theoretical framework could be almost, you know, I, I, I sometimes think of it as like a meta theory. It's a, it's, it's a, it's a set of parts from which to make theories. [00:48:00] Mm-hmm. If, if you will. And that's, that's a metaphor. And I don't know if, if Mike would agree with me, but, um, I, I bet he would agree with most of it and have some bones to pick.

    So, uh, 

    Benjamin James Kuper-Smith: that's pretty good. 

    Paul Smaldino: Yeah. So, you know, and a model is, well, I mean, a model is a formalization, a model is to say, well, you know, I have a system and I have questions about the system. So I've decomposed the system into parts and relationships between these parts and the properties of these parts. And I'm now gonna formalize to say, well, I'm gonna set it up.

    All, like all my assumptions and lay them out very clearly. You can do this, um, mathematically as a, as a set of equations that relate certain variables to other variables. You can do this computationally by saying, well, I have a, an object that has these parts and these, you know, are the things that it can come in contact with.

    And these are [00:49:00] the, it's affordances and the the ways that it can behave and make decisions and affect other parts of the system. And then you just kind of. Work it out forward and see what, what, what are the consequences of making all those assumptions? Um, I, I use this phrase, which I stole from the, the Jeremy Guna, who's a systems biologist.

    Um, which is that a a a model is a logical engine that turns assumptions into conclusions. And I, I really always love that line. And, you know, I think, I think that's all it is. And, and the reason why this is useful is that our brains are very bad at thinking through the, the consequences of our assumptions, especially about complex systems.

    And so, you know, and the model allows us to start messing around with the assumptions, well what if we assume this different thing? What, what if we, we change the situations here? What if we add this new constraint? What if we, you know, make it so that, uh, you know, this thing, whatever, some property is slightly [00:50:00] changed.

    And we can often find things that we wouldn't necessarily intuit, like, you know, critical transitions and phase changes. And we can find non-linear dynamics where something goes up and then it goes down. So I find them incredibly useful as sort of tools for thinking. I mean, I, I always like this Marie Mann, the physicist, and then, uh, uh, once said, uh, models are prosthesis for the imagination.

    Benjamin James Kuper-Smith: Yeah. Yeah. I think, yeah, I really like that too, because I think when you, I mean also just for like checking Yeah. Just seeing like what, what happens if you change this assumption? Like, does everything, like, does it really require a very precise parameter? What happens if you just mess around with it a bit?

    Um, so like, okay. I think it's fairly, I think many people can agree that formal walls seem like a great idea to specify what you're doing. At least in many contexts it's a good idea to do that and that kinda stuff. But for me, really the, the big question is. Like, what do we [00:51:00] not do then in terms of education?

    Like if you have a, you know, psychology undergraduate or something like, or graduate project or whatever, like, uh, program, not project, if we already have all the stuff. So if we want to add, I mean, I'm not, I guess you're not exactly saying that we should add this as part of the curriculum, but it sounds like it in some places.

    But then, you know, we have to do something less basically. So what is, 

    Paul Smaldino: yeah. Um, do you 

    Benjamin James Kuper-Smith: see what I mean? 

    Paul Smaldino: I do see what you mean and no, this is the, this is not the first time I've heard this. Um, I think we, I think we do have to change the curriculum. Absolutely. I mean, certainly psychology is a mess in the way it's taught is, is just not, is, is, is.

    I think, you know, it's the sign of a very young field that is maybe, um, I. You know, I, I'm very sympathetic to a lot of, uh, Paul Zen's, uh, arguments that he's made and sort of psychology, you know, got into hypothesis testing too quickly and, and should, uh, you know, do some more exploration. Um, but I [00:52:00] also think that, you know, toolkits are really important.

    So, you know, I, and I, I think of this really differently because, so I went from, I did my undergraduate in physics and I then went to a graduate program in psychology later, and I was, you know, as in most graduate programs in America, the, the graduate students are involved as teaching assistants for, uh, the, the undergraduate courses.

    And so I was very, you know, sort of seeing how psychology was taught and it's, the differences are very stark, right? It, your first year in physics, you, you know, we did mechanics. For the first half of the year. That was the main thing was mechanics and math, mechanics and calculus. And then the second half was maybe a little optics, and then it was, uh, electricity and magnetism and more math [00:53:00] and, and that's it, right?

    You, that's, that's what you learn their first year. You learn a lot of math and you learn the foundations of mechanics, and you learn the foundations of electricity and magnetism. You don't talk about quantum physics. You don't talk about relativity. You know, you don't talk about thermodynamics, you don't talk about materials.

    You don't talk about any of that stuff because you don't, you're not ready yet. In psychology. So basically 

    Benjamin James Kuper-Smith: be like pre 20th century physics. 

    Paul Smaldino: Yeah. You're doing pre 20th century, you're mostly doing pre 20th century physics. Absolutely. And I think that there's a really good, you know, there's a good argument to be made that this kind of thing is, is really valuable.

    That saying, we're gonna learn tools that are gonna help you. Then any other problem that we're gonna fa faced with these kinds of tools are going to be useful. So we're gonna focus on particular problems. And a lot of the problems are not just the [00:54:00] topics because they're the older topics, but they're also the topics on which a lot of the methods were developed.

    So the methods that were used to solve problems and mechanics and in electricity and magnetism are very similar to the, the computational, the mathematical, the methodological tools that were used to study things like thermodynamics or quantum physics or relativity. Um. Contrast that with psychology, whereas, you know, you, you do this kind of broad sweeping survey where you have to learn like the whole history of psychology and you, you, you, you have, you learn a lot of facts and kind of theoretical ways of thinking about things and metaphors and terminology and, and there's very little about how anything works.

    Right. 

    Benjamin James Kuper-Smith: You get experimental, well, maybe it's probably different between each country, [00:55:00] but I did mine in the uk and yeah, you had like how to do an experiment, basically how to run one. 

    Paul Smaldino: Right. 

    Benjamin James Kuper-Smith: How did and stats, that was basically it in terms 

    Paul Smaldino: of, yeah, so how to run an experiment, but again, where did the hypotheses come from?

    Right? You just make 'em up, right? Yeah. Um, yeah, no, I mean, I took, I took a, I did a master's in psychology before I went to do the PhD and you know, I had to take a, a research methods course and yeah, it was, it was just, oh, what's your hypothesis? Just make it up and, uh, then go test it and. I remember when I was, I, I got really into, um, when I was a grad student, I went off on a number of, sort of deep dives onto things that I don't work on anymore, but I, I was really interested in learning and memory.

    Um, my, during my master's degree, I worked in a psychophysics lab working on visual attention, and I, I became very interested in the mechanics of, of sort of spatial memory and, uh, and learning. And I remember I read a lot of [00:56:00] papers, um, on these topics. And I read papers in psych from, you know, written by cognitive psychologists and I read papers by computer scientists and roboticists, and the, those papers were better and they, they, what they conveyed was almost always a much.

    To me anyway, a much richer understanding of what these things were and how they worked, what memory is, how it works, what learning is, how it works. The reason is they had to build tools that used their ideas and those tools had to work, right? 

    Benjamin James Kuper-Smith: Yeah. 

    Paul Smaldino: Whereas if you're a psychologist and you just have to come up with an idea that sounds good to other people, it doesn't do any work, 

    Benjamin James Kuper-Smith: and you find a directional effect.

    Yeah. Those are 

    Paul Smaldino: the two things basically, right? Yeah. Find a directional effect, like, oh yeah, if this happens, then this happens more than the other kind. Um, 

    Benjamin James Kuper-Smith: yeah. 

    Paul Smaldino: So I mean, I, I think to me, I think [00:57:00] I would, I would lose a lot of topical information and say, you can learn, you can always learn the topics. Um.

    Benjamin James Kuper-Smith: But isn't there also value to that? Because I almost, so I'm kind of in two minds about this because I completely agree with what you said. Um, but I also really value that I was exposed to a lot of different ideas in my psycho, like in terms of like almost finding your topic and then learning how to do research.

    Because I guess most people who do psychology don't necessarily want to do research, right? It's, it's, 

    Paul Smaldino: yeah. Uh, I, yeah, I do agree with that. Um, so as soon as I finished saying it, uh, the, your exact objection popped into my head, and so I'm glad you said that. Um, yeah, and also, I mean, it's very clear the problems in, let's say physics are, are much usually, at least at the, the undergraduate level, much more clearly defined, right?

    Like, what happens when you throw something up and, and it comes back down and we wanna know how fast it comes down, right? [00:58:00] Uh, it's very easy to, to articulate that problem, whereas like, you know, how is a memory encoded? Like, that's a really hard problem. So, talking about, I mean, so I'm in a cognitive science department now, and one of the things I really love about the cognitive science approach is that it integrates cognitive psychology with computer science, but also with philosophy.

    So we take the, the, the side that you need to be able to build computational models of these phenomena to really understand how they work and, and interact with, uh, you know, all the, the behavioral stuff and the, and the neuroscience stuff. Like you should be able to model it, but also. You should be able to engage with philosophy of mind and, and, and consider the question of what are we talking about?

    What are these things made of? What is it? Does it make sense to think of these things in this way? What would be an alternative way of thinking about these things and engage with those questions in a [00:59:00]serious way? Because I think that. Unlike physics. Exactly. This is, seems to be exactly your point, is that the, the, the, the, the, the, the, the parts that we're dealing with, the systems we're dealing with are much less well-defined and are much less intuitive.

    So yeah, being exposed to ideas, whatever they are, you know, uh, it could stay on memory and talk about things like, you know, visual spatial sketch pads or, or, uh, you know, uh, connectionist models of, of semantic networks or, um, you know, the idea of like Mars, David Mars levels of, uh, of understanding, right?

    Because computational algorithmic and implementational and how these are different kinds of questions. Um, yeah, I think all that stuff is super important. So it's, it's not obvious to me. Where the lines should be drawn in training and what needs to [01:00:00] go and what needs to stay. And I think that, you know, a, a complimentary to that.

    I think that, obviously not everyone, I think that everyone should learn some modeling. Anyone who's a scientist, who's actually a researcher should learn some modeling and should learn to engage with the models and understand modeling papers and what models say and don't say that doesn't mean that everyone should become a modeler.

    Benjamin James Kuper-Smith: You should be able to understand it. 

    Paul Smaldino: Yeah. 

    Benjamin James Kuper-Smith: A rough idea of what they're doing. 

    Paul Smaldino: Yeah. And I think of this in, in, in a way that, like, I, I think behavioral ec, my impression is that like behavioral ecology is a field is very good at this. Whereas that in that most behavioral ecologists appreciate, or at least many do appreciate the value of the models and what the predictions are from the models and how the theories are formalized in models.

    And they do research that are, you know, driven by the, the lessons that models, um, have, you know, have provided and, you know, uh, predictions that models have made. But it doesn't mean everyone's doing modeling. In [01:01:00] fact, most people aren't. Right. Uh, and that's, I think there's a lot of reasons for that, right?

    Is that we, we wouldn't learn anything. You actually have to go and study the world sometimes. And then, you know, also there's just different skill sets and interests. Psychology would be a, a much poorer field if there weren't people who just were obsessed with experimental design or just loved spending hours and hours in the lab.

    Like, I want those people there. I don't want them working on models, but at the same time, I want people working on models. And I, I, I don't think it's a a, an issue in psychology in particular that has to be resolved if the field is gonna move forward, is, uh, a greater respect for, for theorists and to say we are going to support people who, who focus on theory.

    I mean, I, I was once talking to someone in, in a psychology department and I said, would you guys be willing to hire a theorist? And they said, well, not someone who only does theory. If they only did modeling in theory, that would be they, we, we can't hire them. You know, they would have to also do experiments.

    And I was like, well, that's crazy because you don't demand that your experimentalists also do [01:02:00]theory and also do modeling. So it's, it's viewed as this add-on this, this sort of thing that's kind of nice to do on the side, but I think of it as so fundamental to, to actually learning anything in, in science that if, if you're not gonna, you're not gonna invest in, in formalization and informal theory, then you're, you know, past a certain point wasting your time.

    Benjamin James Kuper-Smith: I have a, so I had a thought when I read that part in whatever paper feels it was where you mentioned this need for a theorist. Um, I'm not sure I agree with this thought, but I find it kind of interesting and that's, did, did we kind of, to some extent, um, uh, what's the word? Um, can't think of a proper verb.

    Uh, but did we basically just make those people part of econom uh, economics? 'cause I feel like a lot of economists who do theoretical work, it's the topic sound like psychology papers to me. Some of them, they talk about motivation, self-discipline, and all these kind of things that I, I wondered at some point, like, did [01:03:00] we just say, okay, if you want to do formal theoretical work, be part of that part of economics and then Well, the problem is then obviously people, they don't talk to each other, so you lose the, the purpose of it.

    But 

    Paul Smaldino: yeah, I mean, yeah, I think that's a great point. I, I, I, I was talking to somebody recently, um, who's a, a journal editors, you know, sort of lamenting that I think she said something like psychology has, has, seems to have outsourced some of the most important things. That's the word. Thank 

    Benjamin James Kuper-Smith: you. Yes. That's the word 

    Paul Smaldino: I was looking for.

    Yeah. Uh, to, to beha two fields like economics and behavioral ecology and. I think to some extent that's true and, and it's, it's also, you know, I mean, I think the whole idea of. S hard lines between disciplines of people studying humans is crazy. Um, you know, I I, I was just talking to a, a program officer at some funding agency, uh, about whether or not some, some project would be appropriate for the social psychology office.

    And they said, well, uh, it's, it's a little bit on the line between [01:04:00] what would be sociology and what would be social psychology. And, and if you're gonna, we're gonna find it, then it has to be more, this kind of perspective. You have to focus more on the individuals. And if you focus more on the social structures than it's sociology.

    And I'm like, that's crazy because 

    Benjamin James Kuper-Smith: that's what makes it interesting. 

    Paul Smaldino: The interaction is what makes it interesting. It's the interaction between the individual behavior and the social structure that makes it interesting. But if you have one. Community only focusing on the social structure and one community only focusing on the behavior, the individual behavior.

    Then you, you don't, you, you miss out on so much. Um, so yeah, I mean, you know, economics, I, there's economics has its sort of own special problems. 

    Benjamin James Kuper-Smith: Yeah. 

    Paul Smaldino: Um, but one of the things, you know, a problem that they don't have is, is a respect for, for formality and modeling. I mean, people have argued that they've taken it too far, although they seem to be kind of dialing it back over the last, uh, bit of time, you know, with sort of behavioral economics and the like.

    Um, but [01:05:00] yeah, um, you have, you have people. Anyway. And then, then there are sort of certain fields of, of mathematical psychology or cognitive science where people are, are very invested in doing formal models. And they, a lot of them I've talked to feel like pariahs in psychology that like their co you know, their psychologist colleagues don't respect them.

    They don't seem to get what they do or why it's important. Um, and that's, I mean, I'm not saying all, you know, certainly there's, there's bad theory and bad modeling. Like there's bad everything. So certainly some is not, you know, great and should be revered or, or respected just by virtue of the fact that they can build a model.

    But, 

    Benjamin James Kuper-Smith: you 

    Paul Smaldino: know, but that is an, I think it's an important tool and 

    Benjamin James Kuper-Smith: yeah. Yeah, I find it really difficult. Like, I remember when I was in my. The, the point I wanted to make is that it almost feels to me that part of maybe the problem is that at least in the UK the way it was taught is that you have like [01:06:00] one psychology bachelor for everyone who wants to do psychology, most of whom are clinic people want to do clinical stuff.

    And I just remember in one of the first year cognitive psychology, uh, lectures, they talked about the CONNECTIONIST model and just half of the people just almost panicked at the, at, like just seeing something that's vaguely mathematical logical and. I mean like when, uh, and because they had like no interest in doing anything mathematical or doing anything that wasn't clinical almost.

    Right. So I almost feel like it might be useful to separate maybe more cognitive science, like as a thing in general from people who want to do clinical. But then Yeah. 

    Paul Smaldino: Yeah. I, I, I, I think that's probably right, is that clinician, clinical psychology should probably be completely separated. Um, I mean, there's certainly certain information, there are probably certain classes that it would benefit everyone to take on both sides, but, but in terms of the, I did my master's degree in a program that was mostly clinical people who wanted to become clinical [01:07:00] psychology and psychologists.

    And it, 

    Benjamin James Kuper-Smith: how was that as a physicist? 

    Paul Smaldino: It was extremely frustrating. I think it, uh, I think it probably, you know, helped develop my sort of curmudgeonly attitude, uh, toward a lot of psychology. But, um, you know, yeah, it was very, I uh. I remember, uh, I would just, you know, in a social psychology class, be constantly raising my hand and just be like, but, but wait a second.

    Like, why would it be that way? There's no, it doesn't seem like there's any good reason to think that. And the, you know, the professor would be like, well, it's, you know, there's, it's a theory that's developed by, you know, it's based on the writings of this existentialist philosopher from the 1960s. And I'm like, well, how is that a basis for anything?

    But, um, and then, you know, s was talking to somebody else in the class later, they're like, yeah, we notice you, you ask a lot of questions. Yeah. And I, with the implication that it was too many questions. Yeah. Yeah. It was. They're, we just, we just wanna get through this and learn [01:08:00] the things, write down the things we need to know to get our degree and, and move on.

    Benjamin James Kuper-Smith: Yeah. Yeah. We were, yeah. Yeah, yeah. Yeah. So actually one thing, one another, one other question I had about, um. Uh, the, again, the paper models are stupid. I only need more of them. Um, I mean, I think, you know, in large part, you, you chose that, I mean, title and the way of explaining like why it's good to have stu stupid models, um, that over, not oversimplify, but strongly simplifies things to just show, like, this isn't a bad thing necessarily.

    Right. This common criticism. But one, I was wondering, like, when I read that, like what, when kind of is something oversimplified, like how do you know you've oversimplified a model or when kind of does the, you know, and quote stupidity of the model actually become a problem? 

    Paul Smaldino: Yeah, that's a fair question. Uh, I think it becomes a problem when you try [01:09:00] to, uh, make quantitative predictions, uh, too soon.

    Um, you know, there's a, there's a, there's often a rush. To, to, to, you know, validate everything by prediction. And I think that is sort of the ultimate goal is to be able to say, well, the model predicts that this should happen, and so we need to go test it. But, but we're also dealing with systems that are both so inexact, as I've said, and also so complicated that it, it may take some time to develop theory that is good enough to make predictions.

    And there's, I think one of the things that that's at, still, at present, very undervalued in psychology at least, is the idea of doing modeling for the sake of understanding a model system and understanding the consequences and implications of certain assumptions. That doesn't necessarily mean that those assumptions are, uh, are.

    Precise enough to then go out and test the theory in the world. It may be that we need to then, you know, keep, keep [01:10:00] pushing through on the theory to get to the point where we can start adding enough complexity or enough precision that it then becomes a reasonable, um, representation of what's going on in the world.

    Benjamin James Kuper-Smith: So, so what is the kind of, when you say you try and create a formal model, what exactly is the goal then? Um, you know, 

    Paul Smaldino: I mean, the goal, I think there's a lot of goals. There's not just one goal, but, you know, one goal is to say, well, uh, what kinds of structures emerge under certain kinds of constraints that are similar to the kinds of constraints that exist in the world?

    So, I'll give you an example. Um, one of the, one of the first modeling projects I worked on in, in psychology or as a, as a social scientist, uh, was, uh, on, uh, some of the human mate choice, uh, literature. This is, it's kind of a wacky literature and I don't work on it anymore. But, um, you know, it was, it was, you know, people [01:11:00] had been using these models to say, well, how do people make decisions about, um, who they're going to end up with, you know, as a romantic partner?

    And what criteria do they use? And there was this work done in the 1980s where people said, well, you know, there seems to be this, this paradox because, uh, there's this high correlation between, um, between the physical attractiveness of couples. That would indicate to, to us the psychologists that people are searching or, you know, look, you know, sort of have a good sense of their own attractiveness and are looking for a match.

    Uh, however, if you ask people who they prefer and who they would like to be with, everyone prefers to be with the most attractive person they could possibly get. So how do we, uh, how do we, you know, how do these things work together? Well, you know, they built a simple, uh, computational model where individuals [01:12:00] have some attractiveness level that they, that is, you know, everyone else can see and they know about themselves and they put in preferences either for the most possible attractive partner or the most similar possible partner.

    And what they found was if everyone's looking for similarity, you get these extremely high correlations much higher than the correlations you find in the real world. Uh, but if you put in, uh, preferences for, um, you know, the, everyone wants the most attractive, uh, you get correlations that are very similar to what you see in the real world, and they say, well, look at that.

    They, and that the reason is that all the attractive people reject, uh, you know, less attractive people and wait until they, they get paired with someone who's more attractive. And then they had put into, they had to make the model work. They had baked into this thing, which they called the prettier at closing time rule, which is that the longer you wait, the less the lower your standards get.

    And 

    Benjamin James Kuper-Smith: Okay. 

    Paul Smaldino: Um. And so then [01:13:00] another, you know, then the, this, the next level attractiveness people pair off leaving only the night and et cetera, et cetera, et cetera. Well, this makes a, a, you know, so that, that's fine and that that's a, a nice use of models, but it also makes all these like secondary predictions that the, you know, are completely ignored.

    One is that, uh, under this, if, if the assumption of this model is correct, then less attractive people should pair off, should get married much later like that. There should be a, a very strong correlation between age at marriage and physical attractiveness. Which is, which is not actually born out. Um, it, it, it, it also, uh, it turns out that there's a number of assumptions, like exactly the, the strength of this prettier closing time rule or exactly the strength of preferences.

    Like, you can keep the order of preferences exactly the same, but change the strength of how much, a little bit more attractiveness gets you. And it turns out you can make either rule by doing this, fit the data as [01:14:00]perfectly as you like. Um. Then there's all these other data, like, um, well, like I said, like, like age of marriage, et cetera, that are completely ignored there.

    Neither of these rules generate anything looking like realistic curves. Um, you can change the model to make it so that they're, they're more realistic. But then it turns out, if you change the network structure, so the models assume that people meet each other and meet potential mates completely at random, but that's not the case, right?

    There are network effects, right? You meet people who are in your social networks who are doing the same things you are, or gonna the same places as you are if you start, you know, messing around with that. You can also get either decision rule to fit the data as much as you like. So the point here is. The model is useful because it helps you understand how different kinds of decision rules at the individual level can lead to different kinds of patterns at the population level.

    That's, that's useful to know and it makes the model [01:15:00] valuable. But if you start using the model to then make predictions about what should happen or to, to use data in the world as a way of validating one particular version of the model versus another because there's a better fit to the data, well, there are a bunch of assumptions in the model that don't fit what's going on in the real world.

    And when you start messing around with those assumptions, they completely change the fit to the data. So you need to know information, not just about what the individuals are doing, but let's say what the strength of their preferences are and also what the structure of their social interactions are Before you can start saying anything about how good a fit to the data it is.

    And that's what I mean, like, the model is too stupid to fit to the data. 'cause it doesn't, uh, it, it hasn't, it hasn't become specified well enough yet. It doesn't mean it can't get there, but it means that, you know, it's, it's not ready yet. And you need, so that's 

    Benjamin James Kuper-Smith: then, then, 

    Paul Smaldino: yeah. And this is also like, just to, to make the same point I made earlier.

    This is why you need people who are dedicated theorists [01:16:00] who can spend time working with models and understanding how models work and when a model is or isn't ready. And you need someone who's not just tacking, tacking on a model to their experiment as a way of sort of, you know, uh, trying to, to strengthen their argument without really understanding how the, these kinds of models work, right?

    What a modeler does not do is prove your theory with math, right? The model is itself information that should change your theory or add to the theory. 

    Benjamin James Kuper-Smith: That was kind of what I was thinking of doing. I feel slightly caught out right now. 

    Paul Smaldino: No, I think 

    Benjamin James Kuper-Smith: not as explicitly, but yeah, that, that was kind of, 

    Paul Smaldino: I think it's a really common assumption.

    Um, 

    Benjamin James Kuper-Smith: yeah.

    Yeah. And then, yeah, and, but then it's the kind of iterative process, iter iterative process of Yeah. Seeing under which assumptions, what works. And then, 

    Paul Smaldino: yeah. And I, and I, you know, look, I, I've, I've sort of just decided [01:17:00] somewhat arbitrarily, but I, it's 'cause I like it that I was gonna become an expert in, in this kind of modeling and modeling social behavior.

    It's just my, I just really love it. And so I'm really good at it and I know a lot about it. I'm not perfect at it. I've made models that I, or, you know, published modeling papers and I'm like, yeah, I probably could have done things differently. Um, but 

    Benjamin James Kuper-Smith: yeah, 

    Paul Smaldino: you know, I, I know a lot about it and part of knowing that much about it, it, it.

    Highlights how much there is to know. There's so much to know. And that's, you know, it's like any, any expert, any researcher, any professional, right? You wanna become top in your field. You wanna know all the things that you need to know to do your, your job well. And in order to be a, you know, it takes a lot to be a very good modeler.

    And I'm not saying like. I certainly don't think I'm the best modeler, and I think there are a lot of great modelers out there. Um, but a lot of the best modelers are ones that, you know, they have invested a lot in learning about [01:18:00] models, and they're probably not the best experimentals, uh, experimenters.

    Right. Uh, I'm co you know, I'm collaborating with experimenters now and I realize how much I don't know about experimental design and analysis. It's hard, it's really hard. And like hats off to the people who do that, right? The point is like division of labor. The point is that I can do what I do because there are people who do what you do, right?

    Who people who do, who, who dedicate their lives to become really good experimenters or good field researchers or good data scientists or whatever. And, um. You know, science I think, works best in teams where there are people who can build on each other and, and, and models and theory can inform experiments, experiment, you know, inform data collection or analysis.

    Um, because we're all trying to figure out, you know, not just, uh, we're all trying to figure out like what the best questions to ask and the best way to get them answered are that's gonna like, increase our understanding of the world in the most meaningful [01:19:00] way. 

    Benjamin James Kuper-Smith: Well, thank you for sharing your expertise in these papers because the people like me who haven't known, they're very useful.

    Paul Smaldino: Well, that is a, it is a pleasure.