Friday 27 February 2015

Are we Infomation Processers? (A brief note)

I thought about it a lot and I kept thinking: OK, he’s right, I guess, the information is in the light, it has to be there, because where else are you going to get it. It has got to be there and if it’s there, there’s a sense in which you don’t have to process it at least not in the way that I used to say “process.” But if he’s right, what am I going to do about cognitive psychology? How can I reconcile cognitive psychology, as I knew it, with this theory of Jimmy Gibson’s?
Ulric Neisser, from Szolkolszky, 2013

Are we information processors? No. At least, not if Gibson was right.
Information is clearly important, though, because we talk about it a LOT according to this Wordle.
For James Gibson, information is external to the observer. Information is structure in energy arrays (e.g. the optic array for light) that is specific to the object or event in the world that caused the structure. This structures becomes information when we use it to coordinate and control our behaviour.

This information is not transmitted. At any given possible point of observation, there is a uniquely structured optic array that an observer can interact with by going to (or more likely through) that point of observation. That structure is there as soon as the lights have come on and the light is done filling up the space.

This information is also not processed, because it does not 'get into the system'. The nervous system doesn't take information onboard, it resonates to that information; it's dynamical behaviour is altered by detecting that information

What about all the steps that have to happen once the information is detected? Doesn't the information have to be transformed into a behaviour? No. Behaviour simply is the activity of the kind of embodied system that we are in the presence of that particular information. We can alter the kind of embodied system we are via learning, but all the way through learning, the behaviour you exhibit at any given moment simply is that activity.

Therefore, in the radical embodied, ecological approach, it makes no sense to say that cognition involves information processing. This is due to 'information' in REC referring to Gibson-information, not Shannon-information. Shannon-information is not something that exists. It is an abstract description of how to reduce uncertainty between a sender and a receiver. It's an amazing idea; it's the heart of the digital revolution we live in and it's a powerful analysis tool (read James Gleick's great book, The Information for the history). It's just not describing what biological organisms are interacting with. 

Is this merely a semantic issue? Are we cheating by just defining away the problem? No. It's about precision in terms. Information means something very specific in the REC framework, that meaning is not the same as it is in the information processing framework and this difference has consequences. Following up on those consequences is what radical embodied cognitive (neuro)science is up to. We will either be right or wrong, and the data will tell us which, but only if we stop making this basic confusion.

Thanks to Greg Hickok (Twitter, blog) for arguing with me a lot on Twitter about this which helped me clarify a few things, and to everyone else who got into it too. You guys are a big help with your obstinacy and your refusal to take what I say at face value :) I think this post might be the first of a series of 'Brief Notes' where I just try to lay out one thing as clearly as I can. In true Gibson style, I reserve the right to keep critiquing and modifying the details as new evidence comes in!

Szokolszky, A. (2013). Interview with Ulric Neisser. Ecological Psychology, 25, 182–199. Download


  1. Thanks for this clarification, A&S. Suppose I do the absurd and agree with you. We are not processing information, we are resonating. Information is not transformed into behavior, it's just what happens when a body/brain like ours comes into contact with the environment. How does that change the research program? We still need to know *how* the system resonates one way to give rise to behavior x and another way for behavior y. We still need to how how patterns of resonators (shall we call neurons resonators?) are wired up so that they resonate in just the right way. We still need to understand how resonators can re-resonate to give rise to what we cognitive psychologists call memory recall. We still need to understand how resonances can change when we "learn". And if we want to understand the physical processes involved in these resonances, we should probably look to the neural wetware and understand how molecules cause resonances in membrane protein channels and how this resonance leads to membrane potential resonance which leads to a nonlinear resonance (or not) down the axon such that it can resonate with a few thousand other resonators. Then we'll want to know how resonant patterns across a network of resonators resonates with other networks of resonators to ultimately cause musculoskeletal resonance.

    I get it that an ecological approach pushes the investigator to look harder at what the environment and the body can do to help constrain the resonance load in the brain/mind. That's a good thing. I'm on board. But I think we agree that the brain contributes a lot to the resonance (100 billion neurons don't make a simple spring). So once we are inside the brain/mind we still need to figure out how the system resonates. And that is what standard cognitive psychology has been working on since the 1950s.

    1. How does that change the research program?
      Quite significantly. You are no longer looking for systems that implement the computational steps in an information processing model of the task. So you will no longer need systems responsible for processing the information into something else, for example.

      Of course you look at all the same bits; REC doesn't deny brains are made of neurons! But the function of those neurons and networks of neurons will be very different, and you will interpret their activity very differently. Remember, fMRI and other imaging techniques do not show information processing; they are (indirect) measurements of brain activity which is then interpreted as infomation processing.

      And that is what standard cognitive psychology has been working on since the 1950s.
      Yes it has, but from a very specific point of view that REC argues is wrong. Just because cog sci has been busy doesn't mean it's been asking the right questions or interpreting the activity correctly.

      I think a key confusion here is that you keep missing that the information processing model is an explanatory framing device, not a fact of the matter. There is brain activity; information processing is then something that activity may or may not be. That's actually very important.

      And of course the dynamics of the (actually 80 billion) neurons will be more complex than mass-springs. Luckily no one is claiming that mass-springs are all we need. Getting to grips with the dynamic characteristics of the brain is a formidable challenge, but it's not one that I see being motivated by an information processing account.

  2. It seems to me that readers arriving at this blog can be grouped as follows: those who have no familiarity with either relevant technical sense of "information", those who are to some extent familiar with one sense but not the other, and those who are familiar with both. Those in the first group and those familiar only with "Gibson-information" (or alternatively, "ecological-information") can't confuse the two senses. The last group won't if they really understand the two senses. So, the only group that needs help is those who arrive familiar only with Shannon-information (I was in that group). So, it seems simple enough to deal with the possible confusion by being careful in the ecological context to use a differentiating label, say "G-info".

    So, for no group does it seem necessary or even useful to contrast Gibson-information and Shannon-information, rather only to define the former for those who don't know its meaning. In particular, it would not be necessary to make somewhat strange statements about the latter such as that it "is not something that exists" - whatever that is supposed to mean, at least other than in the context of philosophy of mathematics.

    However, I do question some aspects of the description of G-info. You claim that it is not "transmitted" but that "At any given possible point of observation, there is a uniquely structured optic array". So, how did the optic array get from an observed surface to the observer's location? "Transmission" in the comm sense amounts to modulating a medium. In the case of vision, one can view the ambient light as being the medium and reflection of ambient light from a surface as modulation of that medium. In that sense, the structure is transmitted from surface to observer. I can kind of see how one might not think of it that way when the surface-observer-light source geometry is static, but when it isn't, and hence the light structure is time-varying, it seems clear that something is being transmitted via modulation of the reflected light's structure, presumably G-info.

    Also, you seem to be using "processing" in a rather narrow sense. People brought up in the digital age may understand the word to be short for digital processing, but having started out in an analog world I think of the operation of any input-output system as being processing. And since resonant circuits are common in analog systems, saying that a system "resonates" to the input doesn't alter that view.

    Finally, while inclined to agree that G-info isn't "transformed" into behavior, I would want some other verbal phrase in its place. Behavior is the result of motor neuron excitation, and that excitation comes from somewhere. I use the wishy-washy phrase "neural activity patterns are associated with behavioral dispositions" because I have only the vaguest idea what the neurologically correct description might be. But to suggest that the behavior just arises from the mere presence of the G-info seems even more wishy-washy.

  3. Greg, If Gibson is right, then, at the least, today's cognitive research paradigms need to be supplemented by aggressive research into several other areas. The types of things people do would need to be understood in terms of being specific types of tasks, embedded in an environment. The taxonomy of things organisms (such as humans) do would thus be quite different (i.e., today's divisions such as studying short-term memory vs. studying decision making would fall apart), also, we would put significant effort into understanding the environmental support for such tasks, to better understand what the brain would actually be required to do, rather than simply assuming the brain must do everything.

    When all of that transformation was done, there would still certainly be some areas of psychology where people proceeded in research paradigms that look much like the way cognitive research is currently conducted. So, if THAT is your point, then you are correct that no amount of accepting Gibson would totally get rid of that style of doing research. However the questions would be subtly changed, and its place in the scope of the field would be understood quite differently.

    Was that coherent?

  4. it makes me remember with semantic subject which i got from my lecture. I just knew that if there are informations which we get through many signs. it also depends on our environment, thus if we want to talk with new people who have different culture with us, we have to learn a bit more their region. at least it will help us in understanding what information which they are trying to deliver

  5. Charles,
    Part of what is happening here is a war of metaphors. And it is important to have the right metaphor. It is especially important not to have metaphors that intend things which are not true. G-info is not something "transmitted" or "processed" because those terms come from metaphors that are being rejected.

    It is true that light bouncing off the surface of objects create the structure of an optic array, but the light does not "transmit" the surface. It does not "transmit" the structure of the surface. The old metaphors for vision assumed that happened, and they were wrong, and have led to much confusion as a result.

    Similarly, one does not "process" the light, in the manner suggested by input-output models, The body does "processes" food, in the manner suggested by such models. Food is moved from one place to the other, with different things done to at different places along the way, until some resultant is converted into... "output." Perception-action system do not work like that. Traditional models of vision were uni-directional, and included several assumptions based on an input-output model, and they were wrong, and have led to much confusion as a result.

    You said:
    "neural activity patterns are associated with behavioral dispositions" because I have only the vaguest idea what the neurologically correct description might be."

    I think the problem is that there are two different goals that might be at play: One is partitioning variance, the other is accounting for mechanism. James Tabery lays these alternatives out very well, and shows how conflation of these goals created much of the confusion in the nature-nurture-debates" in his recent book "Beyond Versus" (which I am reviewing at the moment). It IS certainly true that certain "neural activity patterns are associated with behavioral dispositions," but that might or might not be insightful if we are trying to create a causal account of behavior.

    Roger Barker's work, for example, demonstrated quite conclusively that fairly global macroscopic variables (such as the size of the school one is attending) also account for a sizable amount of behavioral variation.

    You added:
    "But to suggest that the behavior just arises from the mere presence of the G-info seems even more wishy-washy."

    Well... yeah, I agree with that. I would probably say it is "equally" wishy-washy, but I no need to split hairs. ;- )

  6. Hi, Eric -

    Your comment motivated a reread of Andrew's post, and I think I have a clearer idea as to why you folks object to "transmit" and "process". So, I now agree that those can be misleading in the REC context and are better avoided. However, I see the same as applying to "information".

    At a point in space there will be light structured by reflection off of surfaces. One can think of the structure as being "transmitted" to that point by the light medium, but doing so isn't necessary and may be confusing. So, better to think of the structure as just being there, available to any prospective user. Point to the REC team.

    The structure presumably can be thought of as being used to produce responsive action. Because of my background, "processing" doesn't imply a specific architecture, so I don't think I'm misled by the word. But others may be, and in any event, for reasons not immediately relevant I happen to like "resonate" a lot. So, avoiding "processing" works for me. Another point to the REC team.

    However, arguing that it is potentially misleading to describe a structure as being "transmitted" and to describe what a user does with the structure as "processing" is different from saying that nothing occurs that one might reasonably refer to using those words. The structure gets from where it's created to where it's used, and the user does something with it. Suggesting otherwise seems a foul. Point against the REC team.

    Notice that my description of the scenario doesn't use "information" at all - the structure itself is used to produce responsive action. In my concept of how that happens, there are intermediate steps: the light structure stimulates the visual sensors, that produces sensory neural activity, and that may result in learned responsive motor neuron activity. But I don't see an additional intermediate step of "detecting" (another comm word perhaps better avoided) anything in the structure that might be usefully called "information". I have only a vague idea of what defines a "dynamical system", but that vague idea doesn't include anything called "information". And neither does the wiki entry on dynamical sytems theory.