Sunday 12 September 2010

Is Cognition Extended?

Over the past week or two, we've been engaging philosopher of psychology Ken Aizawa on the topic of extended cognition. Ken is co-author of a book, Bounds of Cognition, in which he argues cognition is most definitely not extended in any way. We both think the other is wrong, which is always fun; I've been getting to grips with his argument and trying a few ideas out, and I wanted to take a moment to summarise where I think we're at.

Extended cognition is, roughly, the hypothesis that things other than the brain can count as parts of a cognitive system. An early Andy Clark example: you are trying to multiply two numbers together, and as part of the process you write the working out with pencil and paper. Clark wants to suggest that we should consider the pencil and paper to be part of an extended cognitive system, because the pencil and paper are doing some of the work generally thought to be part of the process (for example, relieving the memory burden on the user as they work their way through the computation).

Ken's argument is that extended cognition is essentially a fallacy: specifically, what he refers to as the coupling-constitution fallacy. He thinks Clark and others are simply making an error: while it is certainly the case that the pencil and paper are coupled to a cognitive system, this is no reason to think they constitute any part of the cognitive system. More specifically, they are not themselves cognitive processes. Non-cognitive processes can cause cognitive processes, and vice versa, via their coupling - after all, cognition needs to be about something and to be able to affect the world  - but nothing about that makes the external bits constitutive elements of cognition.

Personally, I think that this isn't a general fallacy: it's merely a risk you run if you aren't careful with your task analysis. The question, though, is one what basis does Ken separate the cognitive from the non-cognitive?

There are two basic requirements that they have for what they call 'the mark of the cognitive', although they don't have anything that fulfills their requirements themselves; their argument is that extended cognition examples don't fulfill these and are thus not cognitive. 
  1. Cognitive processes involve non-derived content, i.e. their meaning must not arise from mere social convention. The meaning of traffic signals is derived, for example; green means 'go' because we decided it does, not because green actually means 'go' in and of itself.
  2. Not just any process can be a cognitive process, and just because the outcome is the same doesn't make the process cognitive. A chess computer, for instance, can play a game of chess but selects it's next move via a process that is entirely unlike what chess masters seem to do. Only the end result is the same. 
These are not uncontroversial, and to be honest the hypothesis of extended cognition simply rejects #2 as far as I can tell.  But regardless, can these be defended?

The primary defence is, I don't think, that strong. Ken acknowledges that it is possible for a cognitive process as defined above to be 'transcranial', i.e. extended, but that it is a matter of 'contingent fact' that it isn't. Cognitive science has been looking at cognition now for 50 years, and has yet to really see anything that looks like extended cognition; worse, even the specific examples cited by people like Andy Clark and others fail to fulfill the criteria. For instance, the pencil and paper example fails because the symbols on the paper are derived content; their meaning is derived from mere convention.

There are two obvious rejoinders to this:
  1. Just because the philosophers you're arguing with have flawed examples doesn't kill the general idea; only the specific example. Killing a bad example is a service, no doubt: but it's not a slam-dunk to kill a single example; this is just healthy dialogue, and you may simply have identified a spherical cow error.
  2. More critically: this argument depends surprisingly strongly on the current state-of-the-art in cognitive science. But the science is based in a flawed premise (the assumption of mental representations, which reside in the brain), or, at least, this is what this blog has been working on demonstrating. Essentially, Ken is looking to a science that has explicitly ruled out looking for anything like extended cognition, and thinks discovering it has not found any examples of extended cognition is evidence against their existence. This is simply unconvincing.
Worse for Ken's argument, I think, is one simple fact: when Gibson abandoned the flawed premise of cognitive psychology after it consistently failed to help him explain his data, he took a simple step back and thought again about what cognition might possibly be about. He immediately found an endless stream of examples of what I think count as extended cognition. The current versions of the extended mind hypothesis are, I think, limited, because they haven't incorporated Gibson. His theories of affordances and information provides what I think is the only current show in town capable of getting extended cognition off the ground, but for some reason Gibson and ecological psychology have yet to be embraced. 

As for Ken's two necessary features for any 'mark of the cognitive' - I think affordances and information gives you (1) quite readily (meaning may be relative to the perceiver but it's not mere convention), although I'm still not sure how critical non-derived meaning is. As for (2), I can simply reject it as a necessary condition, because the (radical) hypothesis of Gibson (and extended cognition) is that this simply isn't true. It isn't evidence against a radical hypothesis when it doesn't fulfill something that hypothesis explicitly rejects; so because extended cognition is functionalist about what counts, (2) seems to not be a concern.

Essentially, Bounds of Cognition seems to simply be a restatement of the features of modern cognitive science that the extended mind hypothesis wishes to reject. That's not an argument that anyone not already convinced will believe (like Putnam's Ultimate Argument for Realism, which only a realist will buy into). But Ken is right that the extended mind people need a theory about the 'mark of the cognitive' to avoid what is charmingly referred to as 'cognitive bloat'; on what basis is something not cognitive? Where is the line, if not the brain? There must be one, or else you risk trivialisation. 

I've always believed that Gibson is the theory these people need: affordances and information provides precisely the structure for thinking about how cognition might possibly be extended, while ecological laws provide a mechanism to prevent that extension from being indefinite.
 
References
Adams, F., Aizawa, K. (2001). The bounds of cognition Philosophical Psychology, 14 (1), 43-64 DOI: 10.1080/09515080120033571

Adams, F & Aizawa, K (2010). The Bounds of Cognition. Boston, MA: Wiley-Blackwell.

15 comments:

  1. Andrew,
    Thanks for the extensive set of comments.

    Here are some preliminary remarks.

    One thing to bear in mind is that "extended cognition" is developed in different ways by different people. Mostly I've been addressing philosophers, who have been resisting me.

    So, for example, to our knowledge, even though "The Bounds of Cognition" briefly discusses Merlin's Donald's work, he has not, to our knowledge, written or said anything about the paper. Similarly, we seem to have provoked little response from the ecological psychologists. (Maybe you can change that. =) ) By contrast, Andy Clark has spent a lot of time beating up on us. So, we have spent a lot of time beating up on him.

    Part of what this means is that we have paid somewhat less attention to the Gibsonian strains of EC. There is, however, my paper on Alva Noe's Action in Perception and then there's chapter 9 of the book. And I have just had a second paper on Noe come out. That's not all that much attention to Gibsonian themes, but it's a target. And Nivedita Gangopadhyay has just come out with a reply to our criticisms of Noe (in this collection: http://theboundsofcognition.blogspot.com/2010/07/special-issue-of-cognitive-systems.html) and I am planning to write a reply (I hope within the year), so this could change. And, of course, I have been trying to learn something about ecological psychology.

    ReplyDelete
  2. "Ken's argument is that extended cognition is essentially a fallacy"

    Technically, we think it is a family of (very popular) *arguments* that are fallacious. In principle, one could drop this kind of argumentative strategy and still endorse extended cognition. To my knowledge, Mark Rowlands (in "Extended Cognition and the Mark of the Cognitive") and Mike Wheeler (in his book in progress) have adopted this strategy in trying to make a case for EC.

    "Personally, I think that this isn't a general fallacy: it's merely a risk you run if you aren't careful with your task analysis."
    Dave Chalmers was complaining at my blog that the C-C fallacy is billed as a general fallacy, when it is not. I can see that one has a ray of logical hope if one thinks that if something is coupled "in the right way" then one can get an extended constitution base. Still, I think Adams and I have written enough to challenge most versions that have appeared in print. (There is one prominent version out there that we have not discussed in print, but the paper is in the mail ...)

    ReplyDelete
  3. " 2) Not just any process can be a cognitive process, and just because the outcome is the same doesn't make the process cognitive. A chess computer, for instance, can play a game of chess but selects it's next move via a process that is entirely unlike what chess masters seem to do. Only the end result is the same.

    These are not uncontroversial, and to be honest the hypothesis of extended cognition simply rejects #2 as far as I can tell. "

    Yes, both of the conditions we have proposed have been challenged in the literature. No doubt. Regarding #2, the truth is that some advocates of EC accept it, where others reject it. You have to recall that there are many ways in which folks have argued for EC and there are many variants. Andy Clark, for example, appears to me to agree that not just any process can be a cognitive process. He thinks that the process has to be an information using process. (See his conditions for "trust and glue".) I've mentioned this a few times at my blog using a distinction between the use of a recipe and the use of an oven in baking a cake. According to Andy, cognition might extended into the recipe, but not the oven.

    ReplyDelete
  4. "Just because the philosophers you're arguing with have flawed examples doesn't kill the general idea; only the specific example. Killing a bad example is a service, no doubt: but it's not a slam-dunk to kill a single example;"

    One can score even without a slam dunk. Once one sees the problem with the pencil and paper example, say, it is a relatively straightforward move (I think) to see the problem with, say, gesture. This is why, I think, Rowlands and Wheeler have rejected the coupling arguments for EC. So, drawing attention to the C-C fallacy has driven the discussion in a new direction.

    ReplyDelete
  5. "More critically: this argument depends surprisingly strongly on the current state-of-the-art in cognitive science."

    It's fine by me if we draw a distinction between something like the hypothesis of extended cognitivist cognition and the hypothesis of extended Gibsonian cognition. In fact, pressing for such as distinction is part of the point about needed a "mark of the cognitive". Let's be explicit about what we are talking about here. Maybe this is not about giving definitions or doing conceptual analysis, but about being clear about what we are talking about.

    "Essentially, Ken is looking to a science that has explicitly ruled out looking for anything like extended cognition,"

    A) Well, folks like Edwin Hutchins apparently advocates EC on the basis of a putatively simple extension of cognitivism. Rob Wilson seems to do this at times as well.

    B) As I've noted at my blog, cognitivism does not beg the question against EC. (In fact, we explain this in the book.)

    ReplyDelete
  6. "The current versions of the extended mind hypothesis are, I think, limited, because they haven't incorporated Gibson."

    Does this comment apply to Noe? Is he insufficiently Gibsonian?

    ReplyDelete
  7. Ken - thanks for the comments! Taking them in order:

    1. Yes, point taken about the numerous flavours of EC out there. I just found that special issue of Cognitive Systems the other day so I have those papers to read when I have some time.

    Actually, at one level, the number of flavours of EC is a little depressing. Clark & Chalmers was only 12 years ago and the field has already fractured this much! This is what we see in psychology right now; no guiding principles, no real theory (in the way physics has a theory or two) to structure the conversation. It's exciting at one level (I think EC should be thought of as a properly radical hypothesis so that does mean it's open season a bit) but frustrating at another - you end up not knowing who's arguing against what. I'm hoping that special issue will catch me up on where things are at.

    2. If all you're doing is picking off the various specific examples, on what grounds are you so generally skeptical about the EC hypothesis? Not seeing any valid examples is indeed something to make you concerned; I guess it occurs to me that one constructive path is to talk about how to fix them. Fair's fair, though, it's not your job :) But I guess I don't see picking off examples as a general defence of bounded cognition, more as pointing out the holes in the current flavours of EC - worthwhile, but not that worrying to the idea of EC.

    This is related to the spherical cow thing a little: picking specific fights means everyone runs the risk of picking the wrong one, and even you sensibly pointing out the problem means you're then all just talking about spherical cows. It's a tricky problem, and it's not easy to dodge.

    3. How can you do EC without functionalism? The claim is that something that isn't that brain is doing what would be cognitive work if the brain was doing it. I guess there's also the angle of doing 'extra but still cognitive work', although the idea of extending cognition always makes me think of the former.

    4. Like I say, killing off bad examples is work well done. But it's not the most convincing evidence that the general approach is flawed; I'm willing to bet that EC, like evolution, is smarter than I am :) But that's about all I'm saying there.

    5. I personally want to draw such a line (between extending cognitivist cognition and ecological psychology), but mostly because I think the current cognitivist approach is so badly flawed. I also think that the hypothesis of EC should be treated as quite radical: doing it right will require a shake-up of what counts as cognition (along, I think, Gibsonian lines,at least as a first pass). Hence I'm not convinced by any defence premised on the usefulness of the state-of-the-art :)

    6. I haven't actually read Noe, so I don't know. Part of my goal with this blog has been a place to write about books like this that I'm reading, but the list is long :(

    ReplyDelete
  8. Well, it is true that I don't have anything like a proof that begins with the assumption that cognition is extended, then leads to a logical contradiction. So, instead, I have to proceed inductively.

    Why am I as confident as I am in my induction? Well, to me, there are stark contrasts between the idiosyncratic features of human cognition and, say, Otto's notebook. It's not just one or two things, it's a vast number of things. Compare the things described in research on memory and Otto's notebook and you'll find them.

    Moreover, the folks I've talked to about EC just don't have much in the way of a successor theory to cognitivism. They mostly make "philosophical" sorts of moves about how we should not be too tied to the current picture of the human brain and we need a more general understanding of cognition. That's just evasion.

    So, what about ecological psychology? Well, turn about is fair play. You don't think that the static viewing case for the Ames Room reveals much about cognition. Stuff happens there, but not much that illuminates Real Perception. Well, I think that stuff happens with optical flow in birds and catching fly balls, but these sorts of "stimulus driven" tasks do not much illuminate Real Cognition. Cases in which action at t follows more or less in lock step with what is perceived at t dos not capture, most notably, what seems to me to be going on in normal language perception.

    I'm fine with the optical flow and catching sorts of cases as perfectly fine within their own sphere, but it's the extrapolation that's the worry. You know exactly what I'm talking about.

    So, Skinner was probably right about rats in a box and Gibson was probably right about some bird flight, but it seems that neither scales up to language processing.

    In disanalogy, perhaps between you and me, is that I don't dismiss bird flight or ball catching in the way you dismiss (or used to dismiss) static viewing of the Ames Room.

    But, all of that is largely to lay my cards on the table, rather than make an argument.

    ReplyDelete
  9. Ooops. Couple of typos:

    "You don't think that the static viewing case for the Ames Room reveals much about cognition"

    Last word should be "perception".

    ReplyDelete
  10. I'm thinking that my "broad strokes" resistance is likely to be a red cape to the bull for you.

    But, military analogy is, first, to deny the EC folks their beach heads in examples such as the pencil and paper addition or Inga-Otto. Then, to deny them inroads.

    So, I'm thinking the optic flow and catching flies are probably reasonable beach heads, so there I have to deny the inroads. And, I try to do this by moving over to only slightly more difficult perceptual problems, e.g. the task of identifying objects in a cloth sack using probing spread out over time.

    I also try to deny Gibsonians the eye movements stuff as beach head by arguing that it's not eye movements per se that matter, but variation in retinal stimulation.

    I think the perception under neuromuscular blockade is pretty useful too.

    ReplyDelete
  11. Or, to return to the case of Runeson's analysis of static viewing of the Ames Room, today it does seem to me that he has a viable analysis of that case. So, now I am thinking about how it will break down upon minor extensions. I don't think he has a robust solution.

    But, I'm trying to work this through a careful look at extant empirical work, rather than by having to generate my own experiments.

    ReplyDelete
  12. Ken, these are all pretty good points, and I even agree with some of it! I don't feel too bullish, I like a good argument on clear terms :)

    I've had the feeling for years that EC was indeed lacking a replacement mechanism; I, like Chemero, think Gibson's the place to start looking, so I'm pleased his book is in the mix on that front.

    Well, turn about is fair play. You don't think that the static viewing case for the Ames Room reveals much about cognition. Stuff happens there, but not much that illuminates Real Perception. Well, I think that stuff happens with optical flow in birds and catching fly balls, but these sorts of "stimulus driven" tasks do not much illuminate Real Cognition. Cases in which action at t follows more or less in lock step with what is perceived at t dos not capture, most notably, what seems to me to be going on in normal language perception.
    Fair (and pretty common) response! Two things:

    1. I actually agree that there's something whacky about language. I don't know what that is, and I'm not planning on studying it, but it is indeed clear that language is up to something pretty interesting that ecological psych may not currently have any tools for. I won't rule it out, but I don't see any obvious path just now.

    2. That said, ecological psychology isn't actually stimulus-response, 'time-locked' psychology; it's less Skinner and more Tolman, for one thing, but it also got dynamics in it. Dynamics has become crucial, allowing for more genuinely useful flexibility to be built into the system. For example, Bill Warren talks about 'behavioural dynamics' in which the essential idea is you can control behaviour by being coupled to prospective information, which helps solve feedback delay issues and helps break the tight S-R chain. So I just want to flag up that this is indeed a critical issue but that there are ways of addressing it in the literature (that will either succeed or not, as these things do). I'll work on posts about this stuff.

    Your moves are all perfectly legitimate, ie they do need to be addressed. There are ways, though, and I'll start working some of these ideas into the blog posts on the relevant literature (although it'll take time to do it justice :)

    ReplyDelete
  13. Re: 2, I think you are probably right to draw these distinctions between Skinner and Gibson. The similarities are only superficial. Both reject mental representations. Both have a "driving role" for the environment. But, Skinner seems to have this discretized conception versus a more continuous conception in Gibson.

    But, personally, I don't think that there is much profit in exploring the analogy in any serious way. Too many people find it rude to be compared to Skinner. It shuts down discussion. (I know this from experience.)

    ReplyDelete
  14. Oh, I know; behaviourism is a dirty word in psychology and Skinner was the exemplar. I'm actually a long time fan :)

    But I meant the distinction quite literally. Harry Heft's book Ecological Psychology in Context lays out Gibson's intellectual heritage and his behaviourism is from EB Holt and Tolman, not Watson or Skinner. People forget there were flavours of behaviourism and Skinner was only one.

    ReplyDelete
  15. Heft's book is on my Amazon wish list, so I have at least at one time had the idea that I should read it. It's pretty expensive, though, and our library, of course, does not have it.

    But, near the top of my list on Gibsonian themes is the Fodor and Pylyshyn, then the Turvey reply, then this Shockley, Carello, & Turvey paper on "Metamers in the haptic perception ..." I have this sense that metamers should create problems for Gibson, since they are a kind of ambiguity, when there are at least at times some hints that there is no such thing as ambiguity of Gibsonian information. I think this will be in my October reading.

    ReplyDelete