Embodied cognition – what is it?

I have posted 7 pieces on embodied cognition and it is time to ask whether I have come any closer to a description of what it is. Each post was looking at something very different from the others. Whether or not I’m closer to what embodiment is but I sure know what I think it isn’t. Here is L. Shapiro’s view of the other kind of cognition:

Cognition, on the traditional view, is the same kind of process one finds in a calculator. An organism’s sense organs serve as input devices, translating stimulation from the environment into a syntactic code that the nervous system can then manipulate according to various rules that are either innate or learned. This symbol manipulation is cognition, and its products are additional symbols, some of which might be translated into a form that causes bodily motions or other sorts of behavior. The nervous system, on this account, performs the same function that a CPU does in a computer. For this reason, traditional cognitive science has typically claimed that cognition is computation and that minds are programmes that run on brain hardware. Of particular note in this description of traditional cognitive science is the insular nature of thought. Cognition is cut off from the world in the sense that cognitive processes operate only on symbolic deliverances from the sense organs. Conceivably, were a psychologist able to create sensory code, she could deliver to the nervous system of an organism in her laboratory the same symbols it would have received were it roaming a jungle or a university. In such a case, the organism’s cognitive processes would not differ from those of the freely roaming organism. Because cognition begins and ends with inputs to and outputs from the nervous system, it has no need for interaction with the real world outside it.

… the greatest success stories emerging from traditional cognitive science involve analyses of symbol driven tasks that lend themselves to easy algorithmic description, like playing chess or solving the tower of Hanoi puzzle. In contrast, building a robot that can move about a cluttered environment, which seems to call for cognitive capacities far less impressive than those necessary for chess, is a terribly difficult engineering problem from the perspective of traditional cognitive science. Robots that depend on symbol manipulation to perform activities that would be easy for a cockroach might take days to calculate a course through a busy room and tend to be very slow and inflexible.

 

It was pointed out by D. Wolpert that treating cognition as a computer algorithm results in a computer that can play good chess but is rubbish at picking up a chess piece, moving it, and stopping the move clock in a reasonable length of time. What seems to us hard is easy, and what seems easy is hard.

 

To quite unfairly parody the brain-is-a-computer attitude, here is what the extreme sounds like to me. The brain calculates and therefore is a type of general digital computer, so further, it is a Turing machine (or if not can be stimulated by one). Further a machine will some day be created into which a real human brain state can be ‘uploaded’. This electronic copy of someone can operate in a stimulation of its body and environment and thus be immortal. There need be no biochemistry in the whole system. This is the ultimate ‘en-vatted’ brain. (Of course, I do agree that computer science makes large contributions to understanding brains.)

 

It is all very well to know what embodied cognition is not, but we really want to know what it is. Well, it is about how a biological organism deals with its environment. It is organisms that eat, move, chase, escape and whatever else they do; the brain is only part of the organism and it does not eat or move etc. As soon as we picture an multicellular animal that moves, we know we need some sort of system for the movement. The animal must answer the questions like where am I? Where do I want to be? How can I get there? “Look before you leap”, or according to the old joke, “I can’t yump if I got no place to stood.” What the animal needs is inter-neurons between the sensory neurons and the motor neurons to do some real time, real life type cognition, or in other words it needs a brain. Underneath all the cleverness, cognition is about the survival and flourishing of a real organism in a real world. Embodied cognition is about how the brain is fitted to its body and the environment the body exists in. We can see this embodiment in many ways that may seem disparate, but are all evolved biological adaptations to improve cognition.

 

In the previous posting I have taken 7 very different sorts of embodiment: posture, facial expression, space, the gut, morals, handedness, language. They are linked below and so are a number of earlier posts that touched this subject.

Links to embodied cognition series:

http://charbonniers.org2011/06/15/embodied-cognition-posture/

http://charbonniers.org2011/06/18/embodied-cognition-face/

http://charbonniers.org2011/06/27/embodied-cognition-space/

http://charbonniers.org2011/07/06/embodied-cognition-gut/

http://charbonniers.org2011/07/15/embodied-cognition-morality/

http://charbonniers.org2011/07/21/embodied-cognition-handedness/

http://charbonniers.org2011/07/30/embodied-cognition-language/

Other links to embodiment:

http://charbonniers.org2008/10/12/metaphor-to-embodiment/

http://charbonniers.org2008/11/30/metaphor/

http://charbonniers.org2009/01/11/epistemology-changes/

http://charbonniers.org2009/02/22/a-framework/

http://charbonniers.org2009/05/11/movement-as-the-foundation/

http://charbonniers.org2009/05/23/two-way-arrow/

http://charbonniers.org2009/05/29/embodiment/

http://charbonniers.org2009/06/28/unwilling-to-see-ourselves/

http://charbonniers.org2009/08/30/innate-categories/

http://charbonniers.org2009/09/23/a-pain-metaphor/

http://charbonniers.org2009/11/07/the-body-loop/

http://charbonniers.org2009/12/10/metaphor-2/

http://charbonniers.org2010/05/12/method-acting/

http://charbonniers.org2010/05/15/embodied-power/

http://charbonniers.org2010/06/11/motot-actions-connected-to-memory/

http://charbonniers.org2010/10/12/arriving-at-where-we-started/

http://charbonniers.org2010/12/08/embodied-metaphor/

http://charbonniers.org2011/03/05/embodied-robots/

http://charbonniers.org2011/03/14/space-and-time/

 

 

2 thoughts on “Embodied cognition – what is it?

  1. “To quite unfairly parody the brain-is-a-computer attitude, here is what the extreme sounds like to me. The brain calculates and therefore is a type of general digital computer, so further, it is a Turing machine (or if not can be stimulated by one)”

    Are you suggesting we can solve tasks that a Turning machine can’t? (Its the other way in my, and maths’, opinion).

    JK: I certainly would not want to be in competition solving problems against a Turing machine! Problems that can be formulated in symbols and solved by some form of logic can all be solved by the TM and only a somewhat simple subset can be solved by me (or you I assume). Two things here: My brain does not work like a TM because it is not digital, most thought is not algorithmic and processing is massively parallel. I am not saying it is better or worst – it is different. Now there is the idea that we can get around this difference by using the TM to simulate thought. I believe there are people who argue that is this possible, others that this is theoretically possible but not practically possible, and still others that say stimulating will not work. I don’t have a firm option on how well a stimulation would work but doubt that it is practical. It would depend on how much detail needs to be included to avoid growing errors. The second point is that some thought probably will not translate into some form of symbol/logic language – that is basic to the idea of embodied cognition.

  2. I doubt the existence of problems that can’t be formulated in terms of symbols – can you describe one? Also, there are problems which can be formally specified which can’t be solved by a Turing machine (some think that people *can* solve them, but I haven’t seen any evidence of this).

    “The second point is that some thought probably will not translate into some form of symbol/logic language – that is basic to the idea of embodied cognition.”

    I’ve heard people say this before, but I still don’t understand it. Whats special about interaction with ones environment which defies symbolic expression? I get the impression that because some problems and solutions can lend themselves to study in one way, say dynamical systems analysis, the idea of studying them in another is thought to be outright incorrect.

    JK:
    I stand corrected that there are problems that cannot be solved by a Turing machine.
    I think we may be talking about two different things. You seem to be thinking about types of problems and I seem to be thinking about mechanisms of solution. The same problem might be solved in two different ways that do not have a one-to-one mapping. There may be problems that cannot be stated in symbolic terms, I would not be surprised if there were, but I know of none. And we probably would choose to call them another name then problem. What I am saying is that there are ways of solving problems (or some problems perhaps) that do not use the sequential rule-based manipulation of symbols.
    Think of a feedback loop (normal example here is a op amp), you cannot put your finger down somewhere alone the loop and follow it making calculations at each step and end up understanding how the loop will function. You have to use a formula. Now if you have a few loops and they have a few common points (nested and/or overlapping loops), you cannot understand how they functions using the individual formulas, you need a single complex formula. Now suppose you have millions of such loops – you will have an unworkably complex single formula. It is a fact of neurobiology that if a signal path goes from A to B, there will also by a path from B to A. There may be exceptions but I have not heard of any. What we have is an almost unimaginable mass of nested and overlapping parallel feedback loops. But such a system can solve some types of problems in a flash. It will simply ‘fall’ (with a bit of ‘ringing’ for a fraction of a moment) into its most stable state (the best fit scenario – the lowest energy state – the hardest to perturb state – or the solution). Of course, its stable state may be a cyclic one – it need not be static. Change the value of some inputs or piece of the system and it will quickly change to a new most stable state. Although such an analog system can solve problems its method cannot be translated into a step-wise manipulate of symbols. With almost God-like knowledge, you could figure a overall equation in millions of variables and solve that used an algorithm. This does not seem practical.
    I do not think this stops us from obtaining a good understanding of the brain. The structure is not random; it was built, after all, using a developmental plan held in a single cell. It can be parceled into substructures with functions and so on. We can deal with just the loops between part of the thalamus and part of the visual cortex and see patterns of action. Or in more detail but smaller, Markham is building a model of a thousand or so cells in a small bit of visual cortex. But this is not the model of software working on hardware – the stimulation is just that, it is a simulation. Processes can be discerned in the brain but not ‘programs’. To the extent that the brain is a computer – it is a special purpose, physical, analog computer – not digital, not algorithmic, not general purpose. It is more like an analog patch board then a digital PC.

Leave a Reply

Your email address will not be published. Required fields are marked *