Frank Keil on Causal Thinking

As a practical matter, how much effort do you put into pinning down the causes behind daily occurrences? To developmental psychologist Frank Keil, who studies causal thinking, that answer is likely along the lines of ‘not enough.’ A lack of causal thinking is both endemic, and, to an extent, hurtful these days, he argues, suggesting that lacking even simplified causal models makes things like the black box of artificial intelligence a potential problem.

In this Social Science Bites podcast, Keil, the Charles C. and Dorathea S. Dilley Professor of Psychology and Linguistics at Yale University, outlines for interviewer David Edmonds how causal thinking is a skill we seem to have at an early age, but which diminishes as we grow up. “[K]ids, by the time they approach elementary school, are asking up to 200 ‘why’ and ‘how’ questions a day,” he explains. “Within a year or two up to starting school, they’re down to two or three, often none.”

Furthermore, Keil sees this diminishment continuing in society today – and this comes as a cost. “I think it’s making kids today be pushed more towards surface understanding, being user interface understanders. I think it makes influences more influential. To just say ‘This is cool’ as opposed to ‘This is how it works.’ One of the negative consequences is that we can get fooled by misinformation more; one of the best ways to debunk an expert is to ask them to explain the mechanism.”

At Yale, Keil directs the Cognition and Development lab. He has written several books, from academe-oriented books like Developmental Psychology: The Growth of Mind and Behavior, to more general reader titles like Wonder: Childhood and the Lifelong Love of Science. His awards include the Boyd R. McCandless Award from the American Psychological Association (Developmental Psychology), the Distinguished Scientific Award for an Early Career Contribution to Psychology from the American Psychological Association, a Guggenheim Fellowship, a fellowship at the Center for Advanced Study in the Behavioral Sciences at Stanford University, a MERIT Award from the National Institutes of Health, and the Ann L. Brown Award for Excellence in Developmental Research.

To download an MP3 of this podcast, right-click this link and save. The transcript of the conversation appears below.

David Edmonds: The Scottish Enlightenment philosopher David Hume said, we could never see causation. We could only observe one object regularly followed by another. But the Yale psychologist Frank Keil says both children and adults naturally interpret the world in a causal way. One problem with the modern world, he says, is that as items in our lives have become ever more complex, causation has become much harder to detect.

Frank Keil, welcome to Social Science Bites.

Frank Keil: Thanks for having me. It’s great to be on the show. I really like your broadcasts.

David Edmonds: Thank you. The topic we’re talking about today is causal thinking. What do we mean by that?

Frank Keil: What we mean is thinking about patterns in a world where one entity or event happens and results in another occurring. People used to think it was beyond the capacities of young children and infants, and now it seems quite clear it is so they can do very easily.

David Edmonds: So the philosopher David Hume said that all we ever see is one event following another. How should we think about the difference between causation and correlation?

Frank Keil: That’s great. The initial views of Hume and others were largely associative or correlational. You picked up things that co-occurred, and that’s all cause was. But we don’t settle for that. We want to see the idea that something results in something else happening. In terms of kids and even infants, they’re very much more attentive to things where a billiard ball launch is not a billiard ball in motion, as opposed to when one just moves and stops and then one moves, but in a way, to say causal.

David Edmonds: So the kid is seeing one billiard ball hit another billiard ball, and is deducing that the second billiard ball moves as a result of being hit by the first billiard ball?

Frank Keil: Absolutely. And if you do the same event so that there’s a gap, they don’t think that causes it. It’s also true for human interactions, but they know they have a different tempo. So when I look at you, or I chase you, they know that that’s also causal, but they say it’s having a different kind of quality. So one consequence of that is the view that we come into this world, or very early on, we come prepared to see agents as one kind of causal system, and inanimate physical objects as another — physics and psychology, if you like.

Part of this relies on the idea there’s a causal pluralism. If you look around the world, there are different kind of causal regularities. Physics and psychology don’t work the same way.

David Edmonds: They don’t work the same way, but human beings are part of the physical universe. We might be caused in a very complicated way, not like billiard balls, but we’re still part of a causal world.

Frank Keil: Well, the same entity can be governed by several causal systems at the same time. We apprehend them differently. I can look at you and think it’s a physical object, because you’re sitting on my bench and I want you to get away. I’ve let you as an co-directed agent that I want you to move. I learned as a biological agent, that’s something that has tissue inside is capable of growth and reproduction. We can apply all these different causal stances to the same object. We know they’re distinct and they make different predictions.

David Edmonds: So give me an example of some of the experiments that you’ve done to show that children grasp causality at a very young age.

Frank Keil: Well, I guess some of the best work I like is done by Lisa Feigenson at Johns Hopkins where she takes infants, and rather than look at surprise (that’s one of the older methods looking at violation of expectation), she looks at the actions infants perform. So if an infant sees a ball come up and seem to go through another object, they’re very attentive. And then they go and try to figure out what’s going on. They pick up the object and they bang it against another surface to see if it’s solid. If, on the other hand, a ball hovers near, they drop it continuously.

Now the next step is she has a design where she reveals a hole in one of the blocks, and they really stop looking. They say, “Oh, that’s the answer!” So cause is often embedded in a search for explanations.

Why is this thing happening? In our own work, we’ve gone on to show that as children go through preschool years and go into elementary school, this quest for causal understanding gets deeper and deeper, and then it all stalls and often declines away.

David Edmonds: You draw a distinction between statistical and mechanical reasoning. Can you say something about that?

Frank Keil: Yeah, this is a great topic in the field now, in cognitive science. There’s been a flurry of research the last couple decades showing that kids and adults do track the statistics of our environment very impressively at multiple levels at the same time. I might correlate colors and shapes and movements of an object all at the same time. That’s very different from thinking in terms of symbols or language or thought, and that’s what we all think of when we think about causal arguments, almost a proposition in the head.

And that contrast between symbolic thought and what I call association of thought, which is more statistical, is one of the great controversies in the field. Are they hugely exclusive? Do they work together? How do we combine them? It’s a puzzle for an AI systems to puzzle all over the place.

David Edmonds: So I understand the statistical reasoning. That’s where we just see that time after time, X follows Y, or whatever. So we draw an influence, but explain again, what mechanical reasoning would be. Give me an example of that.

Frank Keil: So I’ve been very influenced by what’s called the new mechanism movement, philosophy of science. And this is the idea that we decompose events into components where one object seems to have an influence on another. Often there’s you can look at it and see that the effect of intervening has a result. So, if an invention on a system changes the system, that’s a causal event.

David Edmonds: So this might seem like a silly question, but why is it important that we learn to understand that the world is governed by causal laws?

Frank Keil: OK, I think it’s really a good question. By doing so, we can make more powerful inductions. And children know this. So we in our own world of research, we’ve shown that children as young as 3 or 4, if you tell them that someone knows a causal story, as opposed to simply a list of facts, they think that person owns a more valuable kind of knowledge. They think it generalizes to more objects.

Causal understanding allows you to intervene on systems, to improve them, to be an engineer, allows you to make models. I mean, one of the really fascinating things about a list is we don’t do it all correctly. Our causal models are oversimplified. They’re toy models. They’re sort of cognitive conveniences, but they work. That’s why technology has continued to advance. So we distort the world, but not so much as to make it useless. It is actually a very powerful purchase.

Now that’s a problem with community current AI systems. We don’t have causal models. They have millions and billions of variables that they integrate. And I think that’s a looming problem. I can talk more about that.

David Edmonds: The study that shows that children believe that you have much greater knowledge if you understand a causal story rather than if you just know facts. Can you flesh that out a bit?

Frank Keil: Sure. Suppose I tell you that John knows the following about car engines. He knows how old they are. He knows where the first ones were made. He knows what their name is in French versus John knows how when you put gas in, it causes it to have things happen inside that make it move and produce power, and then you expand that further by giving more details. Kids are very sensitive to interiors. That’s another thing that surprised people. They are what Susan Gelman calls essentialists. They think there’s something inside that tells the real story, and so they look for the insides, and they pursue it relentlessly as a way of unpacking what’s happening on the surface. It seems there’s a causal engine inside that they want to uncover.

David Edmonds: That’s very interesting, because that’s exactly what David Hume said you couldn’t do. You could look for the cause, but the cause wasn’t there.

Frank Keil: Even Bertrand Russell said there was no such thing as cause early on, but I think that’s been rejected more recent philosophy science, people like Nancy Cartwright … well, all the new mechanists have all said there’s always a causal architecture to the world, and our purpose is to discover and use it. And kids know this, and they sense it, and they want it. So one of the ironies — I want to get this out, because I think it’s so fascinating — kids by the time they approach elementary school, are asking up to 200 ‘why’ and ‘how’ questions a day. Within a year two up to starting school, they’re down to two or three, often none. And this is a huge and I think, devastating shift that happens to many cultures.

David Edmonds: Why is that? Is that cultural, or is that somehow biological?

Frank Keil: Neither. It’s universal cultural. It’s not biological. It has to do with a number of converging factors. One is that many school systems, the classes are too big, and they’re overburdened, and they have to teach the test. Extracting cause understanding is more complicated, it requires more teacher knowledge than simply asking, “Do you know the capital of Delaware?”

Then secondly, people have a misconstrued view of kids and how they think there’s an age-old misconception that children are concrete and adults are abstract, and so they think that they can’t understand cause. In fact, development often goes from the abstract to the concrete.

A third factor is they often misunderstand how to motivate kids, and they reward them for knowing facts, as opposed to sharing an exploration of causal structure.

David Edmonds: That makes me wonder whether there are demographic differences between our understanding of causation across cultures. But also maybe, I don’t know, between the sexes, between men and women.

Frank Keil: I don’t think there are as much as people think there are, I think there may be historical changes, though. One of the most amazing changes that’s largely ignored in the last 60 years is how the world is becoming causally impenetrable. Because when a child opened a music box in 1950 it was a little drum with little metal mechanism on it. You wound up. You could see how it made the music. Open a music box today. What is it? It’s a silicon chip, If you opened up an alarm clock in 1950, you could see how it worked. Open today and it’s the same-looking chip. It’s true for everything. The entire revealing nature of the causal world has disappeared. It’s an unappreciated change.

David Edmonds: That’s very interesting. So the fact that we can’t understand causation in the way that we once did, what’s the effect of that?

Frank Keil: I think it’s making kids today be pushed more towards surface understanding, being user interface understanders. I think it makes influencers more influential. To just say this is cool as opposed this is how it works. One of the negative consequences is that we can get fooled by misinformation more. One of the best ways to debunk an expert is to ask them to explain the mechanism. People who are anti-vaxxers almost have no idea the biology underlying it; if you probe them, they fall apart. It’s possible, but it’s very rare for a faux expert to be able to pull it off. In terms of what you asked, to unpack a mechanism. There’s a famous sociologist in Britain called Harry Collins. He pretended to be a gravitational wave physicist at a conference and pulled it off for quite a while. But for the most part, it’s very, very hard.

David Edmonds: I mean, if I think of the way I navigate around the world, I have a very surface understanding of almost everything. So I turn on the switch and the light comes on. But if you ask me how the turning on the switch brought about the light bulb turning on, I would have no idea. And much of my life, I think, is like that.

Frank Keil: Well, I would dispute that. I would say you probably know more than you realize. You drive a car?

David Edmonds: I do.

Frank Keil: You’re old enough that you used to drive a car with a clutch.

David Edmonds: I did.

Frank Keil: You know more about cars, then, I’m saying than today’s generation, because they have no idea of how engine power is transmitted. Automatic transmission is hard to understand, but clutch is very simple. You could feel when your car was talking off, you’d feel the engaging plates slip. You could feel the suspension. They don’t do anything. We even have data looking at how people talk; the same things that are in a Tesla that we’re in a ‘57 Chevy, in terms of suspension, they don’t talk about anymore. They ignore that. I think that we are giving up our cognitive agency. We’re outsourcing too much, because if you outsource this much, you can’t even identify the relevant experts. You get fooled by charlatans.

David Edmonds: It does suggest that experts become more and more important. But as you say, the problem is, then, how do we identify the fool from the expert?

Frank Keil: I have a new book around this that’s called The Disappearance of Insight, which is the idea that being able to see the inner nature of things is disappearing. We have to be very careful to try to inculcate an attitude of learning about it.

I don’t think that you have to know everything, but I think what people want to understand is a basic toolkit of basic causal pattern. They should understand things like, well, the second law of thermodynamics. It seems like an abstract thing, but with the world being inclined more towards randomness it’s a very basic principle that does a lot of work for us. Understanding things like levers and screws into the classical Greek components of how things work, how clock works, does have a lot of portability.

David Edmonds: You mentioned vaccines earlier, and you can take another example, like climate change. I accept that vaccines are very important because I’m told by the experts they are, and I accept that they’re only dangerous to a tiny, tiny, tiny minority of people. I accept that manmade climate change is happening because the experts tell me. But again, I have to rely on people who know how this stuff works, and I just don’t.

Frank Keil: But you know how to vet them, and you hopefully do more than prestige. Our interesting case was when Linus Pauling said Vitamin C was a cure for all sorts of stuff. People initially said, “This is crazy. I don’t know how it works, but he has two Nobel Prizes, not just one, but two. He must be right.” But that didn’t last very long. People kept probing and saying, there has to be more to it. Here. He was just wrong. He just became nuts over it.

Climate change is the hardest one. Vaccines are not so hard, because you can look into the molecular biology in an afternoon to figure out what’s going on. Climate change is these complicated models, and that’s a huge problem. There’s a movement in psychology, and across many other social sciences, to do away with causal language and do away with models and focus just on prediction based on these large LLM-kind of machines. The result of that is we can’t understand what’s going on. It’s called the Ex-AL problem, the explanation AI problem, and it’s a huge problem. I think it’s devastating if you don’t understand how things work. So in psychology, there is a movement to do away with models and focus just on enhancing our predictions, the better we can predict, the better our science is. Maybe that’s true in some sense, but it’s not a life I want to live, and I think it actually stifles creativity.

David Edmonds: I can see one way in which the prediction model would be less successful than a causation model, which we touched on earlier. Counterfactuals — because if all you know is the statistics, you don’t know which bit to move to change the result.

Frank Keil: Exactly. Counterfactuals are always hard. Even when you’re doing models, there’s still lots of things to change, but they help you constrain the space a lot.

David Edmonds: What do we not yet understand about causation? What’s the one question you would want to know the answer to that we don’t yet know the answer to?

Frank Keil: OK, this is one that’s controversial, and that is the idea that we are innately endowed with the notion of cause in the abstract sense, like a logical operator, like and or. Or, it starts off in perception as a perceptual primitive that’s bootstrapped up into, say, physics, and starts off independently in psychology, and eventually they coalesce into one very abstract notion of or that develops quite late, perhaps two, three later. I don’t think that’s right. I don’t see how it could emerge out. I don’t think you ever get there, but that’s a controversy in the field. So I’m leaning towards the view that there’s a primitive language or thought notion of cause. Don’t know how it’s represented, but something quasi symbolic that we attach to things. I think it can be shown eventually, but we’re not there yet. But I don’t know that.

Part of the fun of having causal explanatory focus is you should welcome the stuff you don’t know. One of the things I think that we have to inculcate in our kids and adults is not to be defensive when you find out you don’t know something, or your explanation is complete or wrong. It’s a chance to explore. It’s like when you’re in a park warning around a physical environment. You come around a corner, discover new terrain. You shouldn’t be afraid that it’s novel. That’s cool.

David Edmonds: You would think that the cultures that encourage that kind of ‘why’ questioning would be more successful in a whole variety of domains, but particularly scientific.

Frank Keil: You would and maybe I don’t want to be too Western-centric, some people would say the Enlightenment sort of did. I think ‘why’ questions have to be really unladen with agendas. You can have very politicized, “Why don’t you like oil and petroleum-based cars?” You can push people so it can be a weapon. And I think you have to be honest, and what I call free-range wondering. You have to really wonder in a way that’s not constrained by agendas or political things.

David Edmonds: Let me finish with a personal question. What is it about this topic that drew you in originally, and that’s kept your interest for so many decades?

Frank Keil: I’m not sure, but I was a Sputnik kid, so my entire childhood was flooding with science kits from all my neighbors and friends. Science was good. I went to MIT, and I’ve always been fascinated by discovering how things work, how is in a world of immense, almost indefinitely large, complexity, we’re able to zero in on just a few key elements, and we gain that light bulb moment, gain insight. It’s always partial understanding, but we get there. And so that, for me, has been and continues to be an exuberant, joyous feeling. I think, honestly, it’s the most fun feeling of life, except maybe having kids and having a marriage. Learning how things work, seeing the beauty that lies behind science doesn’t make things boring and quantitative. It makes it exciting and vivid and three dimensional. It’s like when you had my cataracts taken out recently, the world suddenly popped in Technicolor. That’s what explanations do.

David Edmonds: Frank Keil, thank you very much indeed.

Frank Keil: Yes, and my pleasure.