Brain Inspired
Brain Inspired
BI 145 James Woodward: Causation with a Human Face
/

Check out my free video series about what’s missing in AI and Neuroscience

Support the show to get full episodes and join the Discord community.

James Woodward is a recently retired Professor from the Department of History and Philosophy of Science at the University of Pittsburgh. Jim has tremendously influenced the field of causal explanation in the philosophy of science. His account of causation centers around intervention – intervening on a cause should alter its effect. From this minimal notion, Jim has described many facets and varieties of causal structures. In this episode, we discuss topics from his recent book, Causation with a Human Face: Normative Theory and Descriptive Psychology. In the book, Jim advocates that how we should think about causality – the normative – needs to be studied together with how we actually do think about causal relations in the world – the descriptive. We discuss many topics around this central notion, epistemology versus metaphysics, the the nature and varieties of causal structures.

0:00 – Intro
4:14 – Causation with a Human Face & Functionalist approach
6:16 – Interventionist causality; Epistemology and metaphysics
9:35 – Normative and descriptive
14:02 – Rationalist approach
20:24 – Normative vs. descriptive
28:00 – Varying notions of causation
33:18 – Invariance
41:05 – Causality in complex systems
47:09 – Downward causation
51:14 – Natural laws
56:38 – Proportionality
1:01:12 – Intuitions
1:10:59 – Normative and descriptive relation
1:17:33 – Causality across disciplines
1:21:26 – What would help our understanding of causation

Transcript

James    00:00:04    The basic intervention its notion of causation, I think is a very kind of weak or minimal notion. All that it really involves is the idea that there’s some way of wiggling the candidate cause such that the, uh, candidate effect will change. Well, obviously, um, you, in many, many contexts, you wanna know a lot more than that. Uh, I just wanna say if you’re, if you’re a philosopher or even someone or another discipline, this, this is a, this whole complex of questions and issues is a really, um, this it’s kind of fruitful and interesting place to be from my point of view, in my own case. Uh, the reason why I went into philosophy to begin with was I thought that it kind of gave me a license to stick my nose into all sorts of different things. And that’s what I’ve done, uh, in my career. And I found it unbalance very, you know, very, very satisfying.  

Speaker 0    00:01:08    This is brain inspired.  

Paul    00:01:22    The nature of causality and causal explanation has occupied philosophers since there were philosophers, what does it mean when some thing or event or process causes some effect? Hello everyone. I’m Paul and my guest today is James Woodward. Who’s a recently retired professor from the department of history and philosophy of science at the university of Pittsburgh. Jim has written extensively, not about what causation is per se, but about what causation does. He begins with a minimal notion of causality often called interventionism, which is simply the idea that if you were to intervene on a cause or wiggle the cause as Jim sometimes says, uh, then it would make a difference to the effect from that simple notion things quickly get more complicated with questions about how reliable a cause is, how many different ways we can explore a cause how to treat causes and effects at different levels of granularity.  

Paul    00:02:24    How much of some single cause we can attribute to an effect with multiple causes. And so on much of that work was done in Jim’s 2003 book, making things happen, a theory of causal explanation and in his most recent book, causation with a human face, he uses the intervention framework of causality to expand on those ideas. And in particular, to explore the relationship between how we should think about causality, making claims about causality under the interventionist framework and how we humans actually do think about causality, how we make judgements about and reason about causality in the world are causal cognition for Jim. These usually separate lines of research have a lot to teach each other. So we discuss those ideas, some of the other major and minor themes in the new book. And of course, much more in the show notes. I link to both books that I mentioned, find them at brain inspired.co/podcast/ 145 to my Patreon supporters and students in my neuro AI course, I salute you for your support and generosity. Uh, I recognize it’s not the default to take action to support a podcast and I feel a deep gratitude to you. Thank you. Here’s Jim  

Paul    00:03:45    Jim. Welcome back old friend.  

James    00:03:48    Well, thank you, sir.  

Paul    00:03:50    I, I say old friend of course, because this is actually take two of our attempt at recording. A podcast episode due to yours truly is um, oh, what’s a euphemism challenge. Uh, ongoing challenges as a, uh, podcast, uh, creator, perhaps. So I really appreciate you doing this again. <laugh> and we won’t repeat verbatim what we did last time, but we’ll see.  

James    00:04:12    Okay. Sounds great.  

Paul    00:04:14    So your new book, uh, causation with a human face follows, uh, almost a decade after your, what I would call now classic book, um, making things happen and in some sense, uh, the new book causation with a human face, um, is a continuation or an expansion of what you did in making things happen. So in making things happen, you introduced your interventionist, um, framework for approaching causation. Um, and you revisit that in causation with a human face, but then you go on to do, um, things that you alluded to at least, and, and did a little bit in making things happen, um, where you take this functional perspective, functional list perspective, uh, on causation. And in that regard, you focus on connecting an account of, um, causation itself. What causation quote unquote is with how we do causation, something called causal cognition, how we make judgments about causation, uh, and reason about causation in the world. Could you elaborate, um, maybe, you know, we’re gonna have to talk about what interventionist, uh, the interventionist, we’re gonna have to talk about what the interventionist approach is, but maybe before that, um, can you, you know, elaborate on this functionalist approach and, and why you chose it, why you think it’s an important way to proceed,  

James    00:05:37    I’d be happy to do that. Uh, so to begin with the functionalist approach, uh, it’s just the idea that we, we think causally, we engage in causal reasoning because it’s useful for us. In other words, causal thinking is functional. Uh, and I claim that that imposes various kinds of constraints on how, um, we as philosophers or inquiry inquiry in general, uh, should think about causation. So that’s the, the basic picture of the, uh, uh, the functional aspect of causation. Um, you, you alluded to, um, the idea that my view is, is an interventionist view and, and that’s right, it’s a view that consi, uh, connects, uh, causal claims with, uh, claims about what would happen if you were to, uh, get in there and manipulate things. So if C causes E then if you manipulate C in the right way, or you intervene on C in the right way, then he will change. And that’s at least an important part of my story about the functionality of causation causal knowledge is functional for us because it encodes information that we can use to manipulate and control.  

Paul    00:06:52    So I know that, uh, there are other people that have this kind of functionalist approach, like UEA, Pearl, and so on who who’ve been working on this, but, but the history of the philosophy of causation hasn’t really followed that so much, right. It it’s been much more concerned with sort of the metaphysical, what causation is part of the question, um, you know, going back to human, of course, you know, Aristotle and, uh, all those good old philosophers. Um, but in, you know, you allude to this a little bit in making things happen as well, but, uh, you spend more time on it in your recent book on the, the difference between epistemology and metaphysics and why you think that there shouldn’t be necessarily such a difference. And in my view, you almost, um, it could be interpreted as if you say, well, as if it’s you saying, well, I don’t really wanna talk about metaphysics cuz it’s not important, but at the same time, you’re kind of saying they should be talked about in the same vein. So what, why do you hate metaphysics? What’s that about?  

James    00:07:56    Um, I don’t know that I would say that I hate metaphysics, but what, what I do think is this, and I try to say this clearly in the book that the pursuit of, um, an inquiry into the so-called metaphysics of causation that completely neglects the epistemology or the methodology of how we find out about calls, relationships is not going to be very fruitful. And I think there is, uh, at least some tendency in, uh, people who work on the analytical metaphysics of causation to completely neglect the, uh, epistemological, uh, side of things. So in other words, the idea is that metaphysics is one thing. Uh, epistemology is something, uh, completely different. It’s a consequence of my functional way of thinking about causation, that those two things epistemology in metaphysics should not be, uh, taken as completely separate because according to me, um, we’re interested in causal relations because this is useful information for us. And that means, uh, that we have to be thinking about causation in such a way that, um, there are ways of finding out whether causal relationships, uh, in the world obtain or not. And so it, from the functionalist point of view, it’s a kind of constraint on any adequate theory of causation that it should link up with the procedures that we actually have for finding out which causal relationships obtain in the world.  

Paul    00:09:28    So metaphysics not useless, but perhaps less useful for the way we actually do causation in the world.  

James    00:09:36    I, well, I, what I do think is, uh, so, so from my point of view, there are kind of two projects associated with causation or calls or reasoning that I find interesting. One, as I say in the book is a normative project. How should we think or reason causally and you find, um, that this normative aspect of things certainly addressed in the philosophical literature, but also in, um, lit literature on machine learning, where there’s a lot of interest in, you know, how you can learn about causal relationships from, for example, observational data. Um, there’s a flourishing tradition and, uh, statistics and econometrics that is also normative. Um, so what I hope is that the book makes some contribution to that those kinds of normative concerns. I think that’s a completely valuable and interesting project. And then the other, uh, issue, which I find, uh, intellectually very interesting is how in fact do we do it, um, as a descriptive matter, uh, how is it that people think causally and how is it that when they think causally at least to some substantial amount of time, they’re, they’re, they’re, they’re successful in, um, uh, identifying causal relationships.  

James    00:10:58    So I guess it would be a fair thing to say about the book that it focuses on those two projects, what I call the normative and descriptive, and, um, shall we say is at least kind of quietest about, uh, uh, what the underlying metaphysical story is and that quiet is completely, um, uh, intentional it’s, uh, not that, uh, <laugh>, it’s not that I ha it doesn’t occur to me that some people out there think that, uh, one should provide a metaphysical story. Uh, I, I deliberately, uh, refrain from trying to provide one at least of any very, uh, elaborate sort.  

Paul    00:11:37    Well, I mean, being quiet can be pretty loud. So have, have you received pushback from the meta physicians out there that you’re, that you’re not, uh, paying enough attention or doing enough service toward metaphysics? We’re not gonna harp on metaphysics for long. I promise, but  

James    00:11:51    Okay. Well good. But, um, yeah, yeah, certainly. And even in connection with the first book, um, it was a, a kind of ongoing complaint that I really hadn’t provided, uh, an underlying metaphysics of causation. So for example, I talk about, um, what would happen, uh, to one variable if you intervene and change the value of another, but people have said to me, well, okay, but what’s, you know, what are the truth makers for those, uh, claims about what would happen if you were to intervene? Uh, you know, what’s the underlying, uh, story about, uh, uh, you know, what’s going on at a me metaphysical level? Are you committed to powers or dispositions as your account of causation or maybe, uh, relations of necessation between universals or maybe something else? And, uh, I’m, I have to say that I just don’t find, um, uh, the focus on those kinds of issues and, and there’s certainly a lot of it, uh, in the, uh, metaphysics and causation literature. I just don’t think it’s been fruitful, uh, in comparison with these other kinds of issues, both the normative ones and the descriptive ones about how, in fact we do it, where it seems to me there’s been a lot of, uh, really recognizable progress, uh, made, we’ve learned a lot of new things there. And, um, I don’t think that there’s been frankly, anything that is comparable in the, um, uh, the analytical metaphysics of, uh, uh, causation.  

Paul    00:13:31    Uh, you used the word fruitful there, uh, which I believe you use in the, in the book as well. And that is part of the functionalist approach. And I, I think that, you know, maybe that’s one reason why, at least from my perspective in, in the neuroscience world, and I guess in the biological world, uh, a lot of people have really embraced this intervention brand of causation because it lends itself well to actual experimentation. I know it doesn’t require an experiment because you can, um, infer causation from observable data as well. But the, so maybe, maybe just in terms of another, um, background to your approach, uh, you mentioned that earlier that, you know, people, uh, often are causally correct or fruitful in their causal judgements. This has to do with the, um, rational approach that you also take in thinking that there there’s a normative aspect there where people are rationally approaching judgements, judgements of, uh, causation. I don’t know if you have more to say about, about that. I’m not sure I have a question. I just wanted to make sure we, we got that out there because I, I guess on the other side of the coin, um, there are a lot of people who focus on our foibles I’m part of that group. I think I go both ways, but, but you are, you are more focused on, on the rational, um, fruitful, correct? Uh, judgements of causation, right?  

James    00:14:56    That’s true. That’s true. So, uh, just to say a little bit more about that. I, I said a moment ago that I was interested both in normative issues about how we ought to reason causally and descriptive issues about how we in fact do, but I see these as closely interconnected, uh, in just the way that you described that is I, I think it’s true and I realize this is controversial, but I think it’s true as a matter of empirical fact that at least a substantial amount of the time, people do reason in normatively appropriate ways. And so looking at normative models of a Coggle co cognition, uh, often, um, is useful in, um, uh, answering descriptive questions about how people, in fact, reason causally, um, this perspective is certainly not unique or original with me. I think it’s very widely adopt adopted in the, um, causal cognition, literature, where at least, uh, a whole lot of the people working in this area, psychologists and cognitive scientists, uh, do tend to assume some sort of rational model, uh, uh, picture of, uh, of, uh, co causal cognition.  

James    00:16:08    Now I realize that this stands in contrast to another tradition in psychology, which focuses much more on the mistakes we make, uh, our, our biases, um, uh, uh, et cetera, et cetera. Uh, I guess I just find it more, um, interesting somehow and more, more worthwhile, and maybe it’s some level kind of more respectful to, um, uh, human subjects that are in these experiments. Uh, if we, um, try to understand to what extent what they’re doing may, may turn out to be, uh, rational or norm normatively, uh, appropriate. And I mentioned in the book, um, uh, several cases in which, um, people, uh, engage in certain patterns of causal reasoning. Um, the initial reaction was, oh, this shows that people are being, uh, irrational or acting in normatively inappropriate ways. And then we’ve come to realize that actually, no, it was the normative analysis that was wrong. Um, if we kind of think in a, in, in a kind of deeper and more sophisticated way about what’s normatively appropriate, it turns out at least in some range of cases that, uh, their cognition is, uh, uh, rational after all.  

Paul    00:17:30    Can I put you on the spot? You, uh, and have you described one of those examples, maybe your favorite example of that?  

James    00:17:38    Okay. So, so there’s a measure of, um, what’s called causal strength. Um, uh, that is so, so a lot of, a lot of the psychological literature makes use of, uh, subject’s judgment of what’s called causal strength, where this course bound to some, the subject’s response to something like the question, how good of a cause would you say that C is for, uh, E so this is the, uh, experimental data, uh, uh, that, um, at least some, uh, psychologists try, try to try to fit a simple measure of, um, uh, that you might think is the normatively correct one, uh, when you have, uh, two, just a two binary events, uh, uh, a cause C that can take one or two values on or off in an effect E that can take one or two values on or off. Um, you might think that the normatively appropriate, um, judgment of causal strength to make in that case is just to take the conditional probability of the effect, uh, in the presence of the cause and subtract from that what the conditional probability of the effect would be in the absence of the cause.  

James    00:18:59    This is called Delta P and, um, people noticed, uh, when they did these, uh, they looked at this, uh, issue experimentally, uh, that subjects weren’t always judging in accordance with Delta P. And it turns out that there’s a very good, normative reason why they don’t do that. And I won’t go into a lot of details, but it’s laid out very clearly in, for example, uh, work that Patricia Chang at U C LA has done. And the basic problem is that, uh, if you’re just focusing on, on Delta Delta P you are not adequately or appropriately controlling for the possibility that there may be other causes of ere besides C. So Chang has a, a, a somewhat more, um, uh, which tracks at least in some respects, um, causal strength judgements that subjects make much better than the naive Delta PED does. So that’s just one example. I, I can give you, uh, uh, a number of others if you’re interested.  

Paul    00:20:05    Yeah. Well, you, you use a lot of examples in the book. I mean, obviously we’re not gonna go through all of them, but I just, uh, thought it’d be useful as we go along to maybe visit some of the examples, um, about the various topics in the book. Uh, so let’s see, where do we go from here? May, maybe let’s back up for a moment because, um, well, you know what, let’s not back up, uh, this is a terribly unfair question, but, you know, just in a, in a broad sense, and you’ve, you’ve been talking about this a little bit already, um, you know, what is the relationship between the normative and the, the, uh, the normative and the descriptive, you know, how do we broadly judge causal reasoning, I mean, is your rational perspective, uh, has it born out, uh, to, to a successful, um, <laugh> so that such that you think it’s a successful and, uh, oh, how do I put this has championed your rational approach have, have the, uh, results from cognitive sciences championed the rational approach. Like, how do you broad, you know, what’s the big conclusion because you go through a lot of examples in the book, what’s the big conclusion about how we judge and reason causality.  

James    00:21:15    So, so I, I think the question you’re asked, or one question you’re asking is how do we tell whether a, uh, uh, a normative proposal about, uh, reasoning is, um, is correct or appropriate? Is that that’s least question? So my answer to that is, is, uh, uh, a kind of means end, uh, uh, type answer that is we, we specify some goal, uh, what it is that we want to, uh, learn or figure out about. And typically in this sort of context, that this will be learning about some, uh, particular kind of causal concept. One of the themes of the book is that there are a whole variety of different.  

Paul    00:21:59    Yeah, well, I wanna come to that  

James    00:22:00    Al concept. So one of the things you need to do is to just specify clearly right at the beginning, what, what concept it is that you are, um, uh, interested in. I think of all of the causal concepts as connected in one way or another to issues about what would happen under interventions, but the connections are different in different cases. So suppose I, I say that, you know, what I wanna learn is whether if I were to perform an intervention on C, um, E would change. So that’s my goal. Then there are a variety of different ways in which I might try to answer that question depending upon the kind of data that I have available to me, if I can actually do an experiment in which I manipulate C and see what happens to me. Okay. That’s, uh, that’s really straightforward. There’s no complexity there, but suppose I can’t do an experiment suppose I just have so called observational data where that means that the data is observed, but it hasn’t come from, uh, any kind of deliberate experiment.  

James    00:23:07    Then the question I should be asking myself is can I somehow get information out of the observational data, perhaps in conjunction with other assumptions that I think are well supported that would allow me to answer this question about what would happen if I were to perform, uh, an intervention, which I change C and, uh, how can I get an answer to the question of whether E would change, um, under that kind of intervention and under some circumstances, um, you can actually produce mathematical proofs that if certain assumptions are satisfied and you have certain information, then you are gonna be able to answer that question about what will happen under an experiment that’s, uh, uh, under if, if the experimental manipulation were to be performed. Uh, so that’s, um, one way of getting at the normative question. Um, sometimes, uh, you can at least in get at the normative question, or at least suggest, uh, that you’re doing that.  

James    00:24:07    There’s something right about your normative proposal, uh, just by, um, investigating, uh, various kinds of inference strategies empirically. So, um, I don’t talk about this in the book, but, um, I do talk about it in a long, uh, subsequent paper that I published, uh, earlier this year in a, a journal called theoria. Um, there’s been a lot of interesting work in the machine learning literature, uh, concerning, uh, uh, causal inference on the problem of inferring causal direction. So I suppose you have two correlated variables, X and Y uh, you know, that, um, there’s no, uh, common cause. Um, so there isn’t confounding in that sense, but you don’t have any kind of temporal information that allow you to determine which is the cause in, which is the effect in the past. Um, people have thought if you’re just, and you can’t do an experiment, you’re, you’re just observationally seeing these a correlation of some kind between X and Y in the past.  

James    00:25:15    People have thought this was an insoluble problem. You can’t figure out whether X is causing Y or Y is causing X. It turns out that given certain assumptions, um, uh, there are techniques, uh, for solving this problem, that work pretty well. And there are mathematical reasons why they work pretty well. But in addition, the people who devised these techniques, they went out and, uh, they looked at cases where the causal direction was independently known and they applied their, uh, techniques to these cases. And it turned out that the, uh, procedures just as an empirical matter, uh, work fairly well. So one of the things that they did for example, was, uh, to look at the, um, correlations between altitude and rainfall in Germany. Well, of course you think that it’s the altitude that’s causing the rainfall, uh, uh, rather than vice versa and sure enough, um, the results are pretty good at, um, uh, uh, reproducing that result.  

James    00:26:17    So there, there sort of empirical calibration arguments, uh, uh, that you can make as, uh, uh, you can make as well. So my, my general story about, um, uh, rationality, it’s a kind of means end or hypothetical imperative picture are the means you are employing like a certain inference strategy or something like that, getting you to, uh, uh, uh, the end that you want achieve. Now, I think it’s true as I’ve been saying, uh, earlier that at least a fair amount of the time, it turns out that people, um, the methods that ordinary folk use, uh, in fact, do have a, uh, a kind of rational or normative justification of the sort that I just described, but I’m not claiming that those methods that people use are justified just because people use them. Uh, there has to be some kind of independent justification of the means and sort that I was, uh, uh, just describing a moment ago.  

James    00:27:16    So this is not there. There’s some philosophers who may think that we just take people’s ordinary judgements about things, and we kind of, we kind of systematize them and, uh, we describe them. And that’s the correct story about how we should think about, uh, causation or the normatively appropriate story about calls or reasoning? I don’t think I wouldn’t claim anything like that at all. I think you need some sort of, um, independent justification, um, uh, of this means end sort that I was describing. Uh, but I think in fact, given that standard, it does turn out that people often reason causily in norm normatively appropriate ways.  

Paul    00:28:00    So one of the lessons that I drew from the book, um, which was kind of a downer really is just how complicated causation is. Yes. And it just see, you know, every freaking subject is just turns out to be super complicated, the more you, um, the more you dig into it, uh, and that there are maybe, okay, so there’s a couple different ways we could go here being a, an ex neuroscientist I’m automatically, uh, reflexively wanting to talk about scaling up notions of complexity. But before we do that, maybe as even kind of a teaser, maybe we should talk about different notions of complexity because, you know, growing up, I, I guess in, in my naivete growing up scientifically, it seems like, okay, there’s one kind of, cause there’s one, not a thing can cause something or not. And I wasn’t, um, hip to the notions of proportionality of, in variance of different notions of causality. So maybe we can go through and just tease out why there are, you know, why there’s not just a single way to think about causation and some of the different notions of causality.  

James    00:29:08    Okay. Um, well, as you say, this is a, a somewhat complicated landscape, at least from my point of view, um, philosophers themselves have, uh, distinguished, um, at least several different, uh, causal notions. So one familiar, uh, contrast is the contrast between so-called type causation and token causation, uh, type causation has to do with, um, something like a, a repeatable, uh, causal relations. So if I say that, uh, smoking causes lung cancer, that’s a, a so-called type causal relation, a token causal relation, or sometimes nowadays it’s called an actual cause. Um, relation, uh, is a claim to the effect that for example, some particular person, uh, uh, Joe Jones, uh, his lung cancer, uh, was caused by smoking. So that’s one contrast and the way in which you find out about type level claims and the way in which you find out about token level claims are, is an important way.  

James    00:30:13    It’s pretty different. So any, um, treatment of causation has to be sensitive to that contrast, to begin with. Uh, there are other kinds of contrasts too, that I could, uh, that I could go into, and I’m not sure how, how useful they’ll be to the reader. But, um, I would, one thing, someone who I would particularly recommend, uh, in this connection is, uh, uh, Judah, Pearl mm-hmm <affirmative> particularly in recent work. And he’s been very, very sensitive to, um, different kinds of causal notions that really corresponded very different kinds of causal questions. We might want to ask about a system. I mean, for example, we might want to ask, does this cause in aggregate have a, a sort of total or net effect on some, uh, uh, effective interest. That’s one thing we might be interested in, but often, uh, particularly in, um, biological, but also in other sorts of context, we have the idea that a cause can influence an effect by many, many different roots along lots of different, um, uh, roots.  

James    00:31:25    So if you, you know, in a neurobiologic context, if you give a subject of stimulus, um, uh, there may be a number of different, uh, neuronal, uh, uh, roots that, um, lead to some overall response in the brain. You might be interested not in just in the overall effect, but what happens along those individual roots. And that’s, you know, what’s sometimes called the, a path specific, uh, notion of causation or, um, I, I talk about it in my 2003 book as, uh, involving a notion of contributing cause or causation along a route. So you wanna make that distinction too. There are a bunch of other distinctions now it’s, but that’s only part of the story, of course, cause in addition to different causal notions, I claim the following, uh, the basic intervention its notion of causation, I think is a very kind of weak or minimal motion.  

James    00:32:25    All that it really involves is the idea that there’s some way of wiggling the candidate cause such that the, uh, candidate effect will change. Well, obviously, um, you, in many, many contexts, you wanna know a lot more than that. You wanna know a lot more about the detailed character of the relation between C and E. And I have, uh, I don’t in any way claim, this is a complete framework or a complete classification, but I have, uh, several different categories that I think are, uh, important, uh, when we try to understand, um, what I call distinctions within causations. So in other words, we have a bunch of relations. They satisfy this minimal intervention condition for causation. I was talking about a moment ago, but they differ in other ways we can make distinctions among them. So for example, causal relations can, um, differ in how stable they are or as I use the language, uh, how inva they are mm-hmm <affirmative>.  

James    00:33:27    So it might be, um, that if you wiggle C, um, E will change, but only in very, very specific background conditions. Uh, so the CE relationship, it’s very sensitive. You change anything even a little bit. And, uh, the causal relationship between C and E breaks down another possibility is, is that that relationship is comparatively stable or maybe even extremely stable. Um, pretty much, no matter what else you do, uh, as long as C takes a certain value, E will take a, uh, uh, a certain value. And, um, of course other things being equal, we value the stable relationships. We, uh, we try to find them, um, um, we, uh, you know, if you find a stable relationship, then if you, um, if it holds in one context, you can, uh, exploit it or project it to another context.  

Paul    00:34:26    Well, wait, can lemme, just as an example, I mean, just to go, you know, give like the, the, uh, tried example of billiard balls, right? So that would be a highly inva relationship. Uh, the, the billard balls,  

James    00:34:37    What would be,  

Paul    00:34:38    Um, two like two billiard balls, one billiard ball hitting another, and then the projection thereafter, you could do it in water. You could do it in heat, uh, and in the calculations would hold over lots of inva, um, uh, situations. Right. Um, what would, what would be an example of one that is non inva that, that jumps to your mind?  

James    00:35:01    Okay. Um, think, think about, um, an experiment, uh, which might be done, say in, in a social science context, you are interested in whether, um, a certain, um, teaching protocol or, um, academic, uh, educational technique or regime, uh, works to, uh, improve, um, uh, academic performance. And you do as people in this area, do you, you do a, ran, you do a randomized, uh, control experiment. You have a, you know, you, you treat some of the, uh, uh, students, you expose some of the students to the educational regime and you have a controlled group that aren’t exposed. And you find in, in the particular experimental group that you are, um, uh, uh, making use of, uh, maybe it’s someplace in the middle of Indiana, uh, this educational regime, uh, uh, boost academic performance, you then take this, um, regime, uh, and you try it out in, um, San Francisco or in New York city, uh, with different populations.  

James    00:36:18    And it doesn’t work there. Uh, that would be an example of a non variant or relatively non variant relation. It might be that the relationship is genuinely causal, um, and your, your randomized experiment showed that mm-hmm <affirmative>, but it’s sensitive in all sorts of unknown ways to, you know, features of the particular environment, the particular population of students you’re working with and so on and so forth. And so it doesn’t generalize very well. And I think it’s unfortunately true that at least at present a lot of the causal relationships that are, uh, discovered in, uh, uh, the social sciences and to some extent in psychology too, uh, they tend to be relatively, uh, non variant. They don’t, they don’t export very well.  

Paul    00:37:06    I mean, the, it seems like, uh, the vast majority of causal, uh, uh, of causal situations that we’re interested in are gonna be of that type are the ones that matter to us the most seem to the, the biggest class that matter to us are non-BAR types. Is that accurate?  

James    00:37:27    Well, I think it, I mean, I, I think it really depends on the, um, particular case in hand. And this is, I hate to over generalize here, because this is gonna, I’m gonna get people in, uh, any, anyone in any one of a number of various scientific disciplines who may hear this is gonna get upset. <laugh> okay. I think in fact that the relationships that are discovered in physics and chemistry and, uh, the more, what shall I say, molecular or physiological parts of biology, uh, they tend to be more stable <affirmative> and, or in variant than the relationships that we discover in, um, say economics or psychology. Now often of course, what, I mean, the, of course we’re interested in the, you know, the relationships that are discovered in biology, chemistry of course, but we’re often what is really, um, a, a, a particular interest to us are these relationships in, um, uh, uh, social science and psychology, um, because, um, since causation has to do with what has, what will happen under, uh, interventions, we wanna discover relationships that we can, uh, make use of for, uh, social policy purposes. Mm-hmm <affirmative>, but it turns out that those relationships, even if causal are, uh, tend to be, they have very circumscribed, um, uh, conditions of application and exactly how one, uh, responds in, in these sorts of circumstances to that problem, I think is a really, uh, it’s really, uh, it’s really interesting and delicate question, you can always say, well, okay, we’ve, you know, the, the, the really stable relationships are there, even in the social sciences, and we just haven’t found them yet. <laugh>, uh, that’s when possible, how  

Paul    00:39:17    Likely is that?  

James    00:39:18    What,  

Paul    00:39:19    How likely is that?  

James    00:39:21    I, I guess at this point, I’m, I’m a bit skeptical than anything like that is true, but, uh, well,  

Paul    00:39:27    Is it, is it maybe could, could one generalize and say, you know, because there are different, um, I mean, maybe you have more to say about this, but, you know, the physics, the chemistry, these are all nano micro scale. I mean, I know there’s microphysics as well and macro chemistry, but is it accurate as a generalization to say that in variance decreases as scale increases or, and, or as I suppose complexity increases?  

James    00:39:56    Yeah, I would think probably complexity is maybe the, uh, the relevant consideration here because, um, of course, um, Newtonian gravitational theory and general relativity, uh, uh, works very well, uh, for all sorts of, uh, you know, very large scale, uh, uh, structures our universe. So I don’t think it’s just a matter of size, but it may very well, um, be, uh, partly a matter of complexity, but in, in a way, when you talk about a system being complex, that’s just sort of a, I’m not sure how helpful that is, just cuz it’s, it seems to me to be kind of a label for <laugh>, uh, for everything that we don’t understand. <laugh> um, so I think, you know, the, the general question of what makes a system complex and what is it that, um, uh, is going on in those systems that, that is kind of so makes them so refractory to, uh, uh, least simple forms of causal understanding. I think that’s an extremely, um, uh, uh, a deep and interesting question.  

Paul    00:41:05    Well, I’m gonna go ahead and jump in here. I know that I, I keep interrupting and interrupting your trains of thought, but I’m ask about, um, you know, neuroscience and, and brains then, but because a lot of, a lot of the causal cognition work is done and a lot of the, you know, armchair, philosophical thought experiments are done on relatively simple systems, right. Uh, but then, and that’s hard enough to grapple with, as you go through in, you know, in detail through the book with all of these different examples, you know, what, what is your outlook on how we will be able to scale these concepts? Mm-hmm <affirmative> I use the word scale, but in term now, in terms of applying this functionalist approach to causation, to, you know, complex systems in my case like brains, but also in psychology. And what I really wanna ask you, uh, as a follow up is what, um, you see, if you have thoughts on this, as, you know, the, the connection between a brain, a mind and behavior, do you, how do you see them causally connected? Do you, uh, are you brave enough to, uh, tread those waters?  

James    00:42:15    Well, let me, uh, uh, start with your, your first question, which as I recall, um, uh, uh, had to do with, uh, applying causal thinking to a, uh, a system that it’s as complicated as the brain, and it’s certainly true that there are features of that tho that system and indeed other similar systems that make that pose particular problems for causal analysis. And, um, let me just mention two, uh, one is, uh, the presence of causal cycles of various sorts. In other words where, you know, C influences E and E influences F and F influences G but then G comes back around, uh, and, uh, uh, yeah, influences C this is certainly, uh, I, I think something that you find in, uh, uh, uh, going on in neural processing, and then another thing that I think makes the analysis so difficult is that you, um, are involved in somehow trying to understand, uh, processing that’s going on at different levels, so to speak at, at the same time mm-hmm <affirmative>, um, and issues about whether there can be causal relations across levels, uh, et cetera, et cetera then. And, and, and if so, how you, you are even gonna conceptualize that, uh, that becomes, uh, uh, very, very, uh, uh, important in this context, in some of the physical systems that we were talking about earlier, not, not brain, but simple physical systems. Um, you often have, um, a relatively neat separation between levels. So if you wanna understand what’s going on at one level, you don’t necessarily have to pay attention to what’s going on at a, another so-called, uh, lower level. So there’s a, a certain amount of autonomy of levels, so to  

Paul    00:44:15    Speak well that exists in neuroscience versus psychology as well, right. That a lot of people think that that autonomy needs to maintain because, um, it’s dangerous or maybe, uh, reckless to start talking about between levels,  

James    00:44:31    Start talking about what the  

Paul    00:44:33    Cause about explanation between levels that everyone needs to stay on in, on their plane.  

James    00:44:39    But I think that in the case of the brain, <laugh> that sort of strategy isn’t gonna work there, there, there, there isn’t the clean, uh, uh, a separation of levels. And I think the, the prospects for a completely autonomous psychology that is, you know, autonomous of, uh, uh, um, with respect to, uh, what’s going on in the brain. I, I doubt that, uh, that’s gonna, um, that’s gonna pan out very well. So I, I think you have this, um, situation when you have sort of speak leakage between levels and <laugh>, uh, in, uh, uh, instructors like the brain. And that makes also the analysis, the causal analysis, very, um, uh, very difficult. And of course the, the other thing is we don’t really, at this point have all that good a handle or understanding on, um, exactly what the, the units so to speak are, uh, that we, um, uh, would expect in the case of the brain to be standing in, uh, causal relations. Um, one would like, uh, obviously given how many neurons there are that the, uh, that the, the level of analysis had better be something, uh, larger than the individual neuron. Yeah. But what is it circuits, um, uh, populations of neurons, uh, something else, uh, if you just talk about, you know, gross anatomical areas like, uh, V one or V two, or the hippocampus or something like that, that seems, you know, far too, uh, uh,  

Paul    00:46:19    Course,  

James    00:46:20    Course grain to, to, for these to be useful units of causal analysis. Yeah. So I’m, I mean, I don’t have, um, detailed or useful, uh, uh, uh, suggestions about how to proceed. There’s, there’s increasing interest. I, I know because I, you know, I see papers about this from time to time among, uh, uh, neuroscientists about what, you know, if we wanna have some sort of causal understanding of what’s going on, first of all, how do we even think about what causation should mean in this context? And then, as I was saying a moment ago, what, what are the, you know, what are the units, what’s the level, uh, at, at which we can, um, plausibly hope to find, uh, interesting causal generalizations,  

Paul    00:47:11    Uh, before, oh, well, no. Before we get off this topic. So I, I don’t know if you wanna talk about downward causation at all. Um, and I was hoping to kind of tease you out to talk about that when I asked you my second follow up question, which is what, what you see as the causal relation between a brain, a mind, however you define mind and, uh, overt behavior that we can measure.  

James    00:47:35    Yeah, well, so as you may know, for this, isn’t in my most recent book of the causation with the human Facebook, but in other recent papers, I’ve defended the idea that there is, um, such a thing as downward causation understood as, you know, causation from a, a so-called higher level where it’s usually the higher level is just a kind of course, graining of, uh, some lower level. There can be causation from a so-called higher level to a lower level. Uh, I’ve been influenced by, uh, a couple of, uh, recent discussions and books of this, uh, about this there’s a, uh, a long and quite interesting book by, uh, George Ellis, the, uh, uh, physicist and cosmologist among other things where he, I think, uh, advances a, uh, he, he, he, he, he provides a number of plausible physical examples of, uh, uh, donation that is examples from physics.  

James    00:48:34    Um, Dennis Noble, uh, in, uh, uh, biology, uh, works on heart modeling among other things, uh, provides plausible examples, uh, from, um, uh, biological context. So, yes, I see that, uh, I do think there’s such a thing as, uh, downward causation and I claim anyway, that it fits naturally within the interventionist framework, because all that has to happen is you wiggle some, you know, relatively upper level or course grain variable in something at a so-called lower level changes in response. That’s enough for causation. It’s not, uh, you don’t really require anything more than that. Now there are lots of people who think that there are, you know, metaphysical and other kinds of objections to downward causation, but partly because the, um, interventionist, uh, uh, framework for causation is, is so metaphysical thin, so to speak  

Paul    00:49:28    <laugh> yeah,  

James    00:49:28    Don’t have any problem. Uh, I, I don’t think with making sense of, uh, the downward causation,  

Paul    00:49:36    So does, uh, brain cause mind and mind cause behavior does brain cause behavior and mind?  

James    00:49:46    Uh, yeah. I, I, I’m happy with causal relations, uh, running in, uh,  

Paul    00:49:50    All directions does mind cause  

James    00:49:53    The directions. Now, of course, when I talk about the mind, I just, I assume that this is just something physical it’s, it’s just a, uh, uh, a kind of more, uh, quar grained and abstract way of, uh, uh, of describing physical, physical going. Okay. So I’m not here, um, you know, embracing any kind of dualism or anything like that. I don’t think the mind is something that’s, uh, separate from the, uh, uh, uh, from the brain. It’s just the, roughly the, uh, when, you know, when we describe, uh, psychological or mental goings on, we’re just talking about, you know, what’s going on in the, in the brain, or maybe even in the whole person, uh, just at a certain level of, you know, relatively course grained description,  

Paul    00:50:39    Doesn’t extend to the environment. You’re not a, um, inactive, uh, pro inactive approach.  

James    00:50:46    That’s something that I haven’t, uh, thought about a whole lot. Um, I’m not, I mean, I get, get why people wanna talk about, uh, you know, the, the role of the environment and, and, uh, embodied, uh, cognition and all that other kind stuff. I’m not sure at this point that any very interesting science has come out of that.  

Paul    00:51:06    Okay. I’m, I’m gonna have Michael Anderson on soon, so I’ll let I’ll pass that on to him and see what he has to say about that. Um, cuz he’s a pro inactive. Can, can we, um, talk about laws for a second? So I I’ll let you off the hook with the brain behavior, mind stuff here and go back to kind of the, um, in variance, uh, aspects of causation. I realized, uh, you know, again, growing up in science as I was growing up in science, especially physics, of course, where everything is laws, um, you know, the gravitational law, et cetera, I sort of just accepted that there are laws in the universe. And then recently I was in conversation with someone and I realized I had changed my mind so much on this, that I told this person that I don’t believe in laws. And then of course they made a lot of fun of me. And do you think that the earth is just gonna fly away from the sun? You know, that sort of stuff, do you believe in laws or how should we think of laws?  

James    00:52:06    Well, yes, I do believe that there, uh, that there are laws, although I may have a somewhat different, uh, picture of what okay. A law is than, than perhaps other philosophers do. So I just think of laws as highly inva, uh, generalizations,  

Paul    00:52:23    But that means a law doesn’t exist. Right. That means that you, you, your, you give the name law to this thing that doesn’t very much, but it’s not like a fundamental part of the universe or is it  

James    00:52:37    Well, so here’s the way I would think about it. I, I think there’s, uh, so causal claims and even things we might wanna describe as laws can vary in their degree of, in variance. Okay. So something can be in variant, uh, only under a very limited set of, uh, changes in other kinds of conditions mm-hmm <affirmative>, or it can be in variant under a hugely wide range of changes, um, uh, in other conditions. So take something like, uh, so-called law, uh, which says that the, um, you know, the force exerted by a spring is inversely pre proportional to the, uh, well, it’s the negative of, of the, uh, extension. It is in adversely proportional, sorry, it’s the negative of the extension of the, um, uh, of the spring of the extension. And that’s the general, that’s a kind of simpler F equals minus KX. That’s a kind of simple, uh, linear relationship.  

James    00:53:39    Um, it will describe the behavior of some Springs, uh, uh, okay. In a certain range mm-hmm <affirmative>, but of course, if you stretch a spring too much, uh, it’ll break and, uh, the law will no, no longer describe it all, uh, what the restoring force will be. So that’s a generalization, it expresses a causal relationship between the extension and the, uh, restoring force, but it holds only in a relatively, uh, limited range of circumstances, compared that with the Newtonian law of, uh, gravitation, um, holds under a very, very wide range of different circumstances, uh, holds near the surface of the earth holds, uh, for our solar system, uh, uh, holds elsewhere, uh, uh, uh, in the universe, uh, holds for, uh, uh, masses of, uh, varying sizes, et cetera, et cetera. But, um, the Etonian, um, uh, gravitational law itself will break down under some circumstances when, for example, um, uh, very massive objects are with strong gravitational fields are, um, are in play.  

James    00:54:53    Um, when we’re dealing with, um, uh, velocities that are very large in, uh, comparison with velocity of light, then Newtonian, gravitational theory no longer applies very well. And we have to go to general relativity. So that’s my, that’s my picture. We have the, you know, generalizations of more or less, uh, uh, uh, in variance, um, so far. Um, I think it’s a plausible claim, although some will dispute this, uh, that we haven’t discovered any generalizations that are absolutely, uh, exceptionalism. It would appear that, uh, in every case there are circumstances under which, um, uh, the generalization will break down. And this is true even for the highly inva generalizations that we give the name laws of nature to, uh, so case for general relativity, for example, thought that it will break down at, uh, you know, extremely small, uh, link scales, uh, the so-called plot link, that part of what the, uh, search of, for a theory of quantum gravity, uh, is all about. So kind of degrees of in variance. Uh, it’s not like there’s something magical about the notion of law. It’s just, it’s sort of like what’s going on with the, uh, uh, the spring and the restoring forest except much broader range of in variance. And I think that’s all there is to it. I, I don’t think that anything more elaborate than that needs to be said about, uh, what makes your generalization count as a law of nature.  

Paul    00:56:39    All right. So in the book you talk about, you spend, I think, three chapters talking about in variance and different causal cognition, um, empirical studies related to in variance, and then you finish the book with a chapter on proportionality. Um, what is proportionality and what, um, you know, maybe your favorite illustration, uh, of, of results related to proportionality.  

James    00:57:04    So I, I, I said earlier that that even, uh, if you, among those relationships, that count as causal in some minimal sense as captured by interventions, and we can make further distinctions, one distinction we can make has to do, as I’ve been saying a little while ago with how, uh, stable or in variance, the relationship is another distinction, uh, has to do with, to what extent the, the, um, relation between the cause and the effect are proportional. Uh, so that’s another distinction within causation. Now, what does proportionality mean? Well, the rough idea is that the cause and the effect have to fit together, uh, in an appropriate way. And I’ll give you an illustration in a minute, but I also, before I explain anymore, I wanted to flag that this is also by no means an idea that’s original with me. Um, it was, uh, introduced into the philosophical literature by Steve Yalo, or maybe now probably coming up on, uh, maybe close to 30 or 25 years ago. So here is, um, one of Yalo, um, uh, uh, illustrations of the, of the basic idea, have a, um, and, um, the pigeon has been trained, um, in such a way that it will Peck at red and only red targets.  

Paul    00:58:36    Okay.  

James    00:58:37    You present the pigeon with a target, that’s a particular shade of Scarlet, and sure enough, the pigeon pecks now compare the following two causal claims. The Scarlet color of the target caused the pitch into pick that’s causal claim one causal claim two, the red color of the target caused the pitch into P because after all the target in virtue of being Scarlet is, is red. And the intuition you’re supposed to have is that the second one, the second causal claim, the red color of the, uh, target, uh, caused the, um, uh, pigeon to pack. This is somehow it somehow seems better or more satisfactory. And the reason why it’s more satisfactory is that we, we set it in the very setup of the problem that the, um, the, the pigeon would pick whenever the target was red. So intuitively if you pick on, if, if you say that it’s the Scarlet color of the target that caused the pecking, you you’ve described the cause in, in two narrow ways,  

Paul    00:59:45    Okay.  

James    00:59:46    Is kind of overly, um,  

Paul    00:59:50    Cause it, it doesn’t account for all the other shades of red, that would also cause the yeah.  

James    00:59:54    That’s that’s right. Um, so it’s an interesting question. So this is just kind of intuitive idea that, that, that the cause and the effect should be, be at the same grain, so to speak. That’s another way of thinking about it. Okay. So the, the kind of graining that goes on when you talk about the cause of Scarlet is it seems to, it seems to narrow, it seems overly specific, something like that, given that the effect is, um, uh, pecking, you want the cause and the effect to be at the same brain, but several issues then arise. First of all, can you make this precise in any way? Uh, I’ve been talking about it in a sort of metaphorical, uh, way. And then secondly, to what extent do ordinary people in their causal judgements, uh, respect the idea that they, that they prefer causes that are proportional to their effects. And that’s one of the things I try to do in the chapter. I won’t, uh, bore you with the, uh, effort to make, uh, things more precise, but there is some, uh, experimental, uh, work that does seem to show that, uh, people, um, uh, do, uh, tend to prefer descriptions of causes that are, uh, proportional to effects,  

Paul    01:01:13    Uh, is this related to, uh, intuitions, but you spend some time in the book defending intuitions. Um, maybe you could say a, a word about that. I meant to ask you about it earlier, but we can tack it on here.  

James    01:01:26    Sure. So it’s very common practice, not intuition, just sort of, uh, kind of SP the one way of thinking about, about them is they’re fairly spontaneous judgements about particular cases and reasoning about philosophical matters in this way is very, very old, uh, tradition in philosophy. Uh, uh, if you go look at the platonic dialogues they’re full of these, uh, sorts of arguments about, you know, would you consider this a case of justice? Would you consider that a case of justice? Is this a case of knowledge, this so on and so forth? Um, and unsurprisingly, the philosophical literature on causation is also, uh, full of, uh, uh, examples that involve, um, appeals to intuition. That is what we would kind of the man in the street would ordinarily or person in the street, excuse  

Paul    01:02:32    Me. Oh, thank you. Yeah. Oh, you’d saved yourself. Nice job.  

James    01:02:36    The person in the street, uh, would, would, would kind of spontaneously, um, uh, say, and, uh, then P uh, philosophers, try to construct theories of causation, uh, uh, around this. So, you know,  

Paul    01:02:51    And, and this is, this got a bad name after a while, right. Because it’s not experi, it’s not, um, uh, it’s not experimentally robust. It’s, it’s, it seems, and especially given, and I know that, you know, that you find us all to be extremely rational and correct, uh, humans, but we do have human error and those, you know, there are errors regarding our intuitions as well, but  

James    01:03:14    Yeah, so the, there has been a huge amount of discussion recently about the, uh, role of these appeals to intuition, uh, in philosophy. Um, some, uh, philosophers have, have criticized, um, uh, any kind of appeal to intuition. Um, uh, others have, uh, worried that when philosophers appeal to, uh, intuitions they’re, you know, they’re basically among other things making claims about how people or so-called ordinary folk would, uh, judge or what they would say about various situations. Maybe those claims that the philosopher makes from the armchair are not, um, empirically very well supported. That’s been another, uh, another line of critique. Um, one of the reasons why philosophers, or at least some of them have been unhappy with appeals of intuition appeals to intuition, is that at least some philosophers who appeal to intuition have, um, assigned intuition, uh, uh, features that is very hard to see within any kind of, kind of naturalistic or scientific framework, how intuition can have those features.  

James    01:04:33    So, you know, people have thought that when you have an intuition of something, that’s involves some kind of rationalistic, uh, uh, grasp of the underlying metaphysical nature of things, something like that. Um, I do defend, um, appeals to intuition in connection with, uh, uh, reasoning or thinking about causation, uh to some limited extent. That is what I think is going on when a, uh, philosopher, uh, describes a case and then says, uh, you know, my intuition about this case is, is such and such is simply that the philosopher is reporting what he or she, uh, thinks or judges about the case and what the philosopher expects that other people, uh, will judge about the case. So if I give you a, this is a, a standard, uh, scenario, that’s very, very simple. And I think on controversial, if I describe a case in which two people, Billy and Susie are throwing rocks in a bottle, uh, and, um, the both rocks are headed right toward the bottle, but Susie’s rock gets there a little bit before Billy’s and the bottle shatters.  

James    01:05:44    And then if we ask, you know, who broke the bottle or which rock caused the bottle to break well then of course, it’s the one that, uh, got there. Uh, first, uh, that’s our so-called intuition about the example. And, uh, again, I don’t think there’s anything <laugh> anything very controversial about that just understood as a claim about, you know, how people would judge or think, uh, about one of the things I claim in the book is that at least to some extent, um, people in psychology who are doing experiments about, uh, the kinds of causal judgments that people make, they’re, they’re basically in the same business as the philosopher who, um, uh, describes intuitive reactions, uh, to judgments, uh, the, the psychologist is maybe being a little bit more, uh, systematic and, uh, maybe looking at a broader range of subjects. It isn’t just the philosopher and a couple of the philosophers friends who were the, the source of the information about, uh, how people judge, but I think it’s basically, uh, the same sort of, um, undertaking and to the extent that that’s.  

James    01:06:56    So, uh, in other words, to the extent that the philosopher is just making empirical claims about he or she, or most other people will judge, um, you might wonder whether it’s an empirical matter, the claims are correct or not. And of course you can go assess that in, in, in kind of obvious ways, but there’s nothing, um, there’s nothing problematic about that kind of appeal, uh, that kind of use of, um, uh, appeals appeals to intuitions. You don’t have to get into, um, you know, some elaborate, uh, or, uh, rationalistic story about insight into the nature of things or anything like that to defend that kind of use of intuition. Now, an important distinction though, that I do make in the book, uh, is the following. I think it’s one thing to say that, um, philosophers may have a pretty good sense about how, uh, other people would judge in, um, certain range of, uh, familiar cases about, uh, causal relationships.  

James    01:08:02    I think that philosophers and need people in general though, have very little insight into why they judge as they do. In other words, um, um, I might be able to tell you that in a certain situation, um, people are gonna judge that X caused Y and Z didn’t cause Y but if you ask me then what, well, how did you come up with that, uh, judgment or what were the underlying causal processes that led you to make that judgment? I don’t think that that is something that at least in most cases, we’re in a position to, uh, uh, uh, we, we can’t really answer questions, uh, about matters of that sort, uh, via, via appeal to intuitions. In other words, we don’t, we just don’t have the, uh, access my introspection to the causal processes that, uh, generate our intuitions. And I think one of the things that, uh, one of several things that can become problematic about appeals, uh, to intuitions and philosophy is that people sometimes, uh, think not just that they have had access to the judgments that they and other people are likely to make, but that they have some access to the underlying processes, uh, that are generating those judgements or the, uh, the underlying, uh, distinctions that they’re relying on when they make the judgements and so on and so forth.  

James    01:09:30    And I think that’s a good deal, more problematic to get at the underlying processes. You really need, uh, something like experimentation, I think, uh, of the sort that, uh, uh, psychologists do, or at least some much more sophisticated form of, um, of, uh, uh, causal analysis. You can’t really, uh, accomplish it just by, uh, just by inspection.  

Paul    01:09:53    Well, right. Cuz in my view, we, confabulate the reasons why the causal explanation of why we do almost anything and it’s, it’s embarrassing. Yeah. Yeah. But, but we do  

James    01:10:06    <laugh> yeah. There’s and there’s a huge amount of, um, psychological evidence that people, um, uh, at least often, um, uh, confabulate about why they have made the, uh, judgements and decisions yeah. Made. And I I’m, uh, I’m certainly not someone who’s, um, gonna gonna deny that that’s the case.  

Paul    01:10:29    All right, Jim, um, I have one more question for you before we take, uh, we go into sort of some extra time that, uh, I will release only to my, the Patreon subscribers, people who support the podcast. Before I ask you this one last question. Are, are there any, is there anything that we have missed that you want to highlight about, uh, causation with a human face that we haven’t gone over? I know we kind of skipped around a lot of stuff and I interrupted and we went back and forth, but did I miss anything or did we,  

James    01:10:59    Well, I just very quickly, uh, a couple of things. Great.  

James    01:11:04    One of the things I would like to emphasize, um, is, is the, what I see as the very fruitful interplay between the normative and descriptive, uh, in, um, thinking about causation and causal cognition, I would emphasize to philosophers, uh, that the causal cognition literature that’s been produced by psychologists and cognitive, uh, scientists is really a very, very rich source of information and examples, um, for, um, many, many of the problems that they’re interested in and they should really, uh, really pay attention to this, uh, going the other way. Um, as we were suggesting earlier, I think that paying attention to the ways in which, um, people actually make causal judgements can often be rather suggestive for, um, the right kind of normative thinking about, um, uh, causation. Um, if I may use an analogy there’s in, in, in trying to, uh, develop, um, um, uh, structures that, that do, um, uh, computer vision, uh, uh, uh, effectively, uh, has turned out to be, um, very, very useful to, uh, try to investigate how the human visual system does it.  

James    01:12:27    Yeah. Um, and I think that similar thing is gonna be true, uh, in connection with, uh, uh, causal cognition. If you wanna build a machine, uh, that’s good at, uh, causal inference, um, you should pay some attention to how it appears anyway, uh, human beings do it now that isn’t to say that the machine is gonna be successful by, uh, a copy of every respect, what human beings do. Uh, uh, the machine has certain kinds of abilities, like, you know, abilities that multiply enormous numbers together very quickly and so on, uh, that human beings, uh, don’t have. But I, I wanna say that the descriptive stuff can be very, very useful, uh, from the point of view of, uh, uh, uh, people are interested in normative theories. So I wanna just kind of get across this idea that there’s, that there’s this possibility of really, really fruitful interactions between the, uh, between the two I’ll.  

Paul    01:13:29    Hold on before you go on, I have a, just a follow up question about that because you alluded to the visual system and building these networks, that map onto our, you know, visual system and our brains, and can explain brain activity, um, in that case. And we’ve talked about that kind of, um, approach a lot on the podcast, in that case, the network maps onto, uh, like a well known one of the, um, streams, visual streams in the brain, well known architecture and hierarchical structures, right. Or the, the hierarchy of processing. And while I don’t, you know, I don’t even know what the, uh, current neuroscience of causality is, like what, um, you know, to use a phenology approach where in the brain causality happens, how causal reasoning happens in the brain. Do you have any idea? I, I really don’t have any idea what the current science is because you’re talking about these, you know, if you created a neural network that was great at causal inference, um, maybe we would learn something about how humans do it, but it’s not clear to me there’s not a ready known system in the brain that we could try to map that onto at least that I’m naively.  

James    01:14:36    No, I think that’s true. I think very little is known about the, uh, uh, the underlying neurobiology of causal inference will say the following though in, um, the normative, uh, work that’s been done on, uh, causal inference. Um, one of the things that, uh, people have, uh, one of the conclusions that people have come to, and at least this is my interpretation of, of things is that you basically need something like two levels of representation. You need a level of representation that sort of tracks correlations, but of course we know causation isn’t correlation, just correlation. And then, so the, the way in which this is you, you know, captured in normative, theorizing is there’s an additional structure, uh, um, over above the correlational structure that for example, might be represented by something like a directed graph or a set of structural equations, uh, or something, something like that.  

James    01:15:45    Now there’s also, um, a substantial amount of evidence that I go through in, in the book that human beings in their causal representations. However, it, these may be wherever these may be located in the brain, also have this kind of two level, um, uh, uh, kind of representation. There’s a, there’s a sort of more association or, um, correlational, uh, level of representation, but there seems to be something on top of that that is, um, uh, apparently, uh, essential to, uh, uh, successful causal cognition. And this isn’t just a matter of the representation of correlations and I myself think. And, um, this is a claim that Pearl and others have made, uh, that this has implications for, uh, designing a machine that will engage in a successful, uh, causal inference or learning it. It’s not gonna be a machine that just, uh, um, represents patterns of association. It’s gonna, it’s gonna have to have additional, uh, additional structure to it.  

Paul    01:16:56    Do, do you and Judeo Pearl hang out, do you guys, uh, do coffees and, uh, make causality jokes and stuff?  

James    01:17:03    No, not really. I mean, I do. I do know him. Yeah. Uh, uh, uh, and, and certainly have, uh, uh, talked to him on a number of number of occasions, but no, we don’t hang out together, particularly since, uh, right now he’s in, he’s in, uh, Los Angeles and I’m in Pittsburgh  

Paul    01:17:20    And you’re retired. I, you, you told me between these takes that, uh, you were retired. I didn’t know that congratulations on the retirement. And I hope that thank  

James    01:17:27    You.  

Paul    01:17:28    Yeah. Um, what else were we missing? You said you had a, a couple things, did you cover them all?  

James    01:17:33    Here’s the other thing, and, and I guess it’s implicit in, in what I’ve said already. Oh, one of the things that really appeals to me about causation and causal reasoning, just as a general topic is that it spans so many different disciplines. Oh. So there’s work in philosophy. Um, uh, that I think is, uh, at least sometimes, uh, <laugh>, uh, fruitful, uh, and interesting, but then there’s also work in, um, uh, machine learning and statistics, uh, uh, econom metrics and, and, and in psychology and, and, and all of these, you know, people in these different disciplines, talk to one another, at least to some extent. And so there’s very, very fruitfully eye, uh, uh, interaction. So if you’re a philosopher and you’re interested in cross disciplinary work, this is a, I think a very, um, a very interesting place to be. And it is an area which at least as I see it, um, identifiable progress, uh, is being made.  

James    01:18:39    We’re, we’re, we’re learning, uh, um, more and more about all sorts of, uh, uh, different, different things. And, you know, the issues about, um, causal analysis of complex systems, um, uh, thinking, uh, causally in a fruitful way when you have so-called big data, enormous amounts of data, um, building machines that might conceivably, um, uh, engage in, uh, uh, successful causal Ference. These are really, I, I think, uh, exciting and interesting issues. And so, uh, I just wanna say if you’re, if you’re a philosopher or even someone or another discipline, this, this is a, this whole complex of questions and issues, is it really, um, it, it’s kind of fruitful and interesting place to be from my point of view,  

Paul    01:19:35    If we built machines that, uh, did great causal inference, um, thinking about wanting to build machines that, uh, mimic our own general kinds of intelligence, would that mean that they would, confabulate the reasons why also  

James    01:19:49    <laugh> well, that’s a very interesting question. That’s a very interesting question. So, you know, our, our, uh, I guess to some extent our, our, uh, tendency to confabulate is you, you might think of it as a sort of byproduct of a, um, an impulse that we have that is actually quite useful. Yeah. Uh, is searching for explanations for things. Yeah. And I mean, we are the kinds of creatures that, uh, when something happens, uh, we wanna know why we seem to have an impulse to, to try to answer that question. And of course that sometimes leads us to, uh, yeah. Come up with, uh, answers to such questions that, uh, you know, have, uh, no empirical support,  

Paul    01:20:35    But right. It’s like we, we throw the D in, just are satisfied with wherever they land. And, and at least in this respect, maybe  

James    01:20:41    The, you know, I think the question, the relevant question we should be asking ourselves is not, not whether it’s true. Does this impulse sometimes lead us tos? Would it be a good thing if we didn’t have the, all that the trade off?  

Paul    01:21:07    Yeah.  

James    01:21:08    And, and it’s not at all. It seems to be the answer to that last question is probably clearly no. Uh, so it’s, it’s a, it’s an impulse we have, but it gets, um, <laugh>, it can go into overdrive in, in, in ways that is, that’s not, so, uh, yeah, not  

Paul    01:21:25    So there’s no way it’s an evolutionary mistake because evolution does not make mistakes, but <laugh>, that’s a, that’s a whole nother two hour discussion in there, but, okay. So, uh, I’m gonna close, uh, before I ask you just a few kind of one off questions, um, if you had a, so you mentioned progress and that were making discernible progress, um, in causation, in lots of different fields, if you had a magic wand that could give you the, the one thing you think that would help you make progress as a philosopher in causation, um, you know, be the experimental empirical results or some, um, philosophical kind of breakthrough or realization, do you have a sense of what, what you would wish for, with your magical wand?  

James    01:22:12    That’s a very good question. And I, and I don’t have a, um, uh, a single answer to that question. I think a lot of the progress that will occur is gonna have to be somewhat incremental. Um, there will be, um, you know, mathematical, uh, there, there will be, um, uh, additional, um, uh, uh, uh, experimental work that, uh, explains better, uh, what, what it is we’re able to do. I guess, one general thing that I would be very interested in, in understanding better is the following. We, we as humans, at least as adult humans seem to have abilities at causal cognition that are, um, in many ways, pretty superior to those of other animals, including, um, uh, other other primates. But I think the, the exact character of our, uh, advantage is not very well understood. That is, and, and, and I don’t mean to suggest it needs to be just one thing either, but, but what exactly is it that makes us so much better, uh, at, uh, causal inference? Yeah. Uh, uh, uh, than other animals. I mean, you know, other animals are great at, um, you know, kind of more association forms of learning. Uh, so that by itself doesn’t seem to be what the difference is. And so I think, I think that’s a really, uh, that’s a really interesting empirical question that I’d like to know the answer to.  

Paul    01:23:45    Yeah, me too. I meant to actually ask you about that difference. Um, uh, of course it’s not gonna be one thing as we’ve discovered with causality, it’s complicated. It’s always complicated. <laugh> all right, Jim, thank you so much for the book. Um, are we gonna see another one in 10 years?  

James    01:24:04    Sorry. Well, I produce another one  

Paul    01:24:06    In, yeah. In the next 10 years.  

James    01:24:08    Well, I, I don’t know. I mean, I have some ideas, um, that might possibly go into a book. Um, although I’m retired, I’m hoping that I, that I’ll continue working, uh, uh, as long as it’s as, as long as that’s possible. That’s certainly my plan. Um, so we’ll see,  

Paul    01:24:28    You don’t seem retired. I appreciate your time here with me today. And thanks again for recording it again. Uh, I had even more fun this time, actually speaking with you, which I, which was somewhat surprising given that we did it before. So thank you.  

James    01:24:41    Oh, you’re very, you’re very welcome.  

Paul    01:24:58    I alone produce brain inspired. If you value this podcast, consider supporting it through Patreon, to access full versions of all the episodes and to join our discord community. Or if you wanna learn more about the intersection of neuroscience and AI consider signing up for my online course, neuro AI, the quest to explain intelligence, go to brain inspired.co to learn more, to get in touch with me, emailPaul@braininspired.co you’re, hearing music by the new year. Find them@thenewyear.net. Thank you. Thank you for your support. See you next time.