video
Topic Series
Andy Clark
- Robert Lawrence Kuhn
- Andy, I like you have been obsessed with the mind-body problem my whole life. I did a doctorate in neurophysiology, not that that helped very much. You have three big ideas that are in this whole area of mind-body problem: embodied cognition, extended mind, and predictive brain. So, describe to me each of them and then let’s figure out how they articulate.
- Andy Clark
- Okay. I think – I think what I’ll do there is put the embodied cognition one first and then do the other two. So, I think embodied cognition is a kind of big framing idea for most of my work and most of my interest in this stuff. And the general idea there is that if you want to understand the mind which is, I think, the goal, then you can’t afford to just look at the brain. What you’ve really got to understand is how brains, bodies, and worlds kind of get together to do the sort of stuff that we need to do to survive and to thrive in our worlds. So, there are lots of examples of this. You know, one of the ones I quite like is an example of gesturing as you talk. So you might think you’ve got the have the idea, you’ve got to have the idea in full and somehow the gesture is just the icing on the cake. But there’s some interesting work where if you impede gesture as people try to explain their ideas, they kind of break down a bit. You get children to sit on their hands – this isn’t good either. More delicate interventions too. And so people are beginning to think – Susan Goldin-Meadow and others, in particular – that the gesture is kind of part of the process of – of thinking things through. It’s kind of its change in the task that the brain has to do in a way which nonetheless is really contributing to the problem-solving.
- Robert Lawrence Kuhn
- Because all that’s represented in the brain. I mean, when I’m moving my arms now it’s not – they’re not doing them by themselves. I mean, I’m doing it from my motor cortex and all – it’s integrated. So, it all comes together in the brain.
- Andy Clark
- Well, the brain is helping it all hang together. I think that’s what I’d like to say there. So it’s not somehow that everything of interest happens in the brain. For example, if you think about working through a lot of mathematical problems using pen and paper. There’s a famous exchange with Richard Feynman where one of the historians that was interviewing him said something like well, you did all the work in your head and the record of it is in the paper. And he says – no, no, no, that was working. You’ve got to – you’ve got to have working and this how the working happens. No, I think it’s that sort of notion that these things are being done by subtle properties of the body and subtle properties of the world. They’re often the kind of thing where if it was happening inside the head by some sort of biological brain bit, we’d have absolutely no hesitation in saying look, that’s part of the apparatus of cognition. So I think that a main thrust in a lot of what I’ve been arguing over the years is that if you would have that thought if it happened inside the head, then you should have that thought if it happens reliably, robustly, fluidly outside the head. You shouldn’t be sort of…Susan Hurley used to say a magical membrane chauvinist. You shouldn’t think there’s something magical about that membrane.
- Robert Lawrence Kuhn
- Alright. Let’s go on to number two which is perhaps what you’re most well-known for, at least originally – extended mind.
- Andy Clark
- Yeah. So I think that sort of falls rather naturally out of a general embodied cognition viewpoint. So, embodied cognition folk are kind of thinking that properties of the body make a big difference. But in fact, of course one of the things that we humans have done is built complicated, symbol-infested environments around ourselves. You know, the inscriptions on bits of paper, stuff going on in your iPhone and so on. The kind of idea is that when we surround ourselves with environments like that, what the brain does can become so delicately dovetailed to what’s being provided by this reliable bio-external sources of – of kind of structure and support that in fact, the problem-solving is now so tightly interlocked that you’ve got one kind of machine. I think there’s a very, very natural flow from considerations about embodied cognition to considerations of extended cognition. Embodied cognition is making the most of all these properties of embodiment and action as – as a biological organism solves problems. Extended cognition, I think, is what you get when you create symbol-infested environments around yourself, environments that you can use to, if you like, turbo-charge the kind of stuff that we really recognize as being mental stuff. So a lot of what goes on in work in embodied cognition, people will say well, I’m not sure if that’s really fully mental. So you can think of, for example, what do they call them, passive dynamic walkers. These are walking systems that make the most of the sort of links between muscles and tendons and knees and so on. And that really simplifies the puzzle of walking. It’s wonderful work, you can build powered robots this way. There are lots of insights coming from that work. In particular, an insight I think is that what the brain should be doing is just tweaking and nudging this complex system that has its own dynamics. Apply that picture now to a creature that is in the middle of a symbol-invested environment. Now for it to try and solve its problems, it can make the most of those symbolic structures in the same way that a passive dynamic walker makes the most of the – the sort of joints and muscles and tendon linkages, just as a brain of the passive dynamic walker can do less to achieve more, so the brain of the – the biological agent surrounded by notepads and iPhones and software programs can learn to do systematically less while still achieving the same kinds of things. And it’s in that delicate dovetailing that has very fine-grained temporal dimensions that I think you end up making the case for something which is more deeply extended than you might think. Because after all, we all know that we use notepads.
- Robert Lawrence Kuhn
- Right. Okay. And so the deep insight there potentially is that it’s more than just utilizing something but it – it is some sort of – of combination where now I and the iPhone are – are some kind of unit, a world-defined unit.
- Andy Clark
- Yeah. I think that’s right. The notion of…
- Robert Lawrence Kuhn
- I don’t know that it’s right. (laughs) I mean, you know.
- Andy Clark
- Well, a lot of people wouldn’t agree it’s right but I mean it’s a – it’s a good description of the thing that I do happen to believe, that in fact, we can dovetail what’s going on in our biological brains and bodies with what’s going on in these bio-external sources of structure and order in ways that are so tight and so intimate, if you like, that you really can’t sort of unravel it and just say – this is what the brain’s doing, this is what the other bit is doing. Because these are systems that are… A nice phrase I think is – interaction dominated. So some systems are complemental in the sense that you make good of them by just thinking – here’s a component and it does X and here’s another component and it does Y. But sometimes when those bits are very tightly coupled together, exchanging energy and information at very, very fine-grained time scales, then it becomes much harder to do that sort of thing. It looks as if then it’s the interactions that are actually carrying most of the explanatory weight. So in these interaction-dominant cases, then you’ve got an argument for drawing your circle around the – the bigger system rather than the – the bits that to us look like components.
- Robert Lawrence Kuhn
- And that – and that’s very interesting, although everything has to be represented in the brain in order for us to work. So in embodied cognition, if my gestures are meaningful, the gestures are from in my brain, and if I’m working with an iPhone, the visual impact of that and all that are represented by whatever is in the brain, circuitry and synaptic potentiations and molecule changes and all that. So it’s – it’s all represented in the brain.
- Andy Clark
- It’s all… So I wouldn’t quite put it that way. If you could really persuade me to put it that way, then I wouldn’t be an extended mind theorist anymore. (Laughs)
- Robert Lawrence Kuhn
- Okay. That’s what I’m trying…that’s what I was trying to do. (Laughs)
- Andy Clark
- What I would want to say is the brain’s got to be able to cope with it all. The brain’s got to be able to help orchestrate it all. So if you think about even something like using – using a smartphone to recall phone numbers when you need them, it is a case that at the time of recall, that information has to be read in to the biological brain. But a lot of stuff about how you actually use the iPhone, get the information out of it at the right time– a lot of this stuff is sort of dealt with by very low-level routines so that we just kind of move the – move the right information into the sort of biologically dominated workspace at the right time. But the idea, and I think this is maybe another good way of coming at the extended mind – the idea is that where cognition is concerned, access is what really matters. And so if you think about all the stuff you know, well don’t even try and think about all the stuff you know, because you can’t. But I’m happy to say that you know, you probably know when the Battle of Hastings was. You probably know that was in 1066 but you don’t walk around going 1066, Battle of Hastings, 1066. I count – I count you as knowing that because under the right circumstances that information is easily retrievable, easily accessed, easily put to use.
- Robert Lawrence Kuhn
- But now if I would do the same thing, if you had asked me and I didn’t know that and I checked on my – my search engine and instead of in two seconds answering it, it was eight seconds, your argument is that fundamentally it’s not that different.
- Andy Clark
- That’s right. If this was well enough integrated, robustly enough available, and you had the right kind of – the overall system had the right sort of meta-knowledge. So it’s a relatively recent development in thinking about the extended mind to begin to realize how important meta-knowledge is. So it’s not enough just for the information to somehow be there in the big system, some – the right stuff has to be available at the right time, and that requires something, and it doesn’t have to be the biological bit of the system, but something that has a pretty good idea of what’s there and when it’s going to be needed. I think actually this is something that we’re outsourcing more and more to the networks that surround us. They’re much better if you like knowing what it is that the bio-bits of me are about to request.
- Robert Lawrence Kuhn
- Your third big idea is the predictive brain which seems more traditional neuroscience oriented. So, what is that and how is it then integrated with the progression from body cognition to extended mind?
- Andy Clark
- Yeah, you’re right. The picture of what the brain does that I’m currently most impressed by is that the brain is a multilevel prediction engine. I’ll say a little bit about what that is in a second. But – but in thinking about what the brain does, I’m not stepping back from the idea that brains are often part of extended cognitive circuits, or brains and bodies cooperate to solve a lot of problems. Instead it’s – it’s a question of understanding what it is about biological brains that makes them such potent players in this complicated dance. And I think that’s what the predictive brain story has given us. It’s given us the first glimmer really of a systematic understanding of what brains might be doing that lets us understand how they play this sort of – it’s kind of like dance coordinator, choreographer, orchestra conductor, something like that. They’re very, very good at positioning themselves within very complex flows of information, some of it out there in the world, some sort of somehow in the sort of linkages in my body so as to bring the right stuff forward at the right time.
- Robert Lawrence Kuhn
- So then how do you integrate the three ideas?
- Andy Clark
- Yeah. Okay. So this is a – this is a big question and it would be nice to write to a book about it. So let’s – let’s hope that’s somewhere in the – in the future. I think the first thing I’ve got to do is get the predictive brain story on the table properly and then say what bit of it is doing the work here. So, the predictive brain picture is the idea that – that brains are forever trying to predict the sensory flows that are hitting the body in all of the ways that they body can take in information. So, stuff coming into the eyes, the ears, tactile stuff. All the interoceptive stuff too. So, you know, I’m – I’m getting a lot of information about the state of my stomach and my viscera and so on. So, all of this stuff, if the predictive brain story is right, is stuff that the brain is busy trying to predict from the top down. So — what the predictive brain story has really done is turned the traditional picture of perception upside down. There’s a traditional picture where you might think that the brain kind of waits for stuff to come in from the world, and that incoming stuff triggers a sort of cascade of processing, getting more and more abstract as you go deeper and deeper into the brain, and then you get an understanding of what’s out there in the world. If the predictive brain story is right, it’s pretty much the reverse order of proceedings. Moment by moment your brain tries to predict how the sensory flux should be and, to the extent that it can use what it knows about the world to do that, then you actually get to perceive the world. To the extent that the prediction is failing to adequately accommodate the incoming information, you generate prediction errors. And prediction errors play the role of perceptual input now. So it’s sort of like, where you used to think there was a flow of perceptual information, what there really is, if these stories are right, is – is a flow of errors with respect to predictive perceptual information. That’s what moves forward through the system. And this is predictive coding in the good old-fashioned sense that we have from – from working compressed sort of files for movies and so on. So, it’s a very efficient way of coding information within the system.
- Robert Lawrence Kuhn
- So, integrating the three?
- Andy Clark
- Yeah. In a nutshell, the idea is that prediction error, once it’s been calculated by the brain, is a prime quantity to self-organize around. The idea is that if you have prediction errors calculated, then you self-organize around that by trying to minimize prediction error using whatever resources are available. Those resources include moving the body around, changing the inputs. So embodied cognition is – is right there on the surface. At the same time, as you move the body around and try to minimize prediction error that way, you grab stuff from the world. And so if the best way for me to minimize prediction error with regard to what I’m trying to do right now is to pull out my iPhone and do something with my iPhone, or to scribble something on a bit of paper and re-inspect it, then that should be the strategy that is selected by self-organization around prediction error signals. So I think that’s the basic idea is that you needed a property that – that brains could self-organize around that would bring body and world fully into the equation and for my money, prediction error is that property.