Friday 12 August 2011

What's the difference between perception and conception?

As Andrew has been tackling a new job description for the brain (part 1 and part 2), several comments have been made that suggest that his approach (and the ecological stance in general) might be fine for perception/action, but not for other types of tasks/behaviours. Later on in this post I also think about how we might be able to distinguish between association and direct access to meaning, which is another idea that has been coming up repeatedly in the comments (see here).

In this post I want to think about what makes perception different from conception (Andrew reviews William James' views on this distinction here). I will argue that both occur as networks of evolving neural activity (with perception, this network extends to the environment and the body), but these networks have different properties because they are driven by signals of differential stability. I do not think it is accurate to think of perception as something that happens in V1, for example, and conception as something that happens in SFG. If parts of the brain reverberate in a system along with parts of the brain that are directly interfacing with an energy array in the environment, it seems correct to refer to that whole system as perception. In contrast, if a network of activity appears to be relatively encapsulated with respect to external energy arrays, it seems correct to refer to that system as an act of conception. For now I will leave aside the problem of how conceptual networks develop, but I hope to return to this later.



By this characterisation, a conceptual network is identical to a perceptual network except that it is not being driven and constrained by the detection of perceptual information. This avoids a dualism between perception and conception while providing a basis for predicting different properties of the two systems.

In principle, it should be possible to distinguish perception and conception by analysing the properties of networks of neural activity. In practice, of course, we lack the tools to do this (although mathematical tools do exist that can distinguish between properties of different types of networks). Still, it is worth considering what would be the expected characteristics of perceptual vs. conceptual networks if we could peer inside the brain to observe them.

Perception: Perception involves directly interfacing with energy arrays produced by the environment. This perceptual information is structured and continuous. As the information hits some type of receptor (e.g., in the retina) the energy interacts with specialised cells to generate structured neural signals. I have no idea what the fidelity is between perceptual information and the neural signals it produces, but I think it's safe to say that there is a tight coupling between perceptual information generated by the environment and the resulting neural signal (if signal conjures too linear a picture, think of a reverberating neural system). The tight coupling between environment and neural activity must, to some extent, be communicated to the body to permit us produce flexible, appropriate actions in a dynamic environment.

Perceptual networks should be driven and constrained by the conversion of perceptual information in an energy array into a structured neural signal. In an action task, there will be additional information arising from the perception of our bodies as they perform. The structure provided by this interface with the external environment (and due to the fact that our bodies also possess relatively stable properties) should make perceptual networks relatively coherent and stable and they should reflect the dynamics embodied by the perceptual information driving them. The particular characteristics of a perceptual network will also be influenced by the history of activating the network (i.e., long-term changes in the brain that support a particular type of network behaviour), whether the network has been activated recently (i.e., medium-term changes in proximate network activity such as calibration), and other networks of activity (perceptual or conceptual) that might impinge.

The neural activity resulting from the interface between brain and structured energy arrays in the environment will continuously evolve in response to changes in perceptual information. In this sense there are not discrete networks that activate identically each time a given type of information is present. Instead, the present network will be pulled into a different state to reflect changing environmental conditions.

To the extent that the brain is not entirely single minded, the perceptual neural system can be impinged on by other networks of activity. There will be threshold effects such that the system is resistant to perturbations, except those of sufficient intensity. For example, if I am very pre-occupied with my growing to-do list at work, I might not pay as much attention to avoiding other pedestrians when walking down the street. But, if I am only a little concerned with my to-do list, then my walking performance will be unaffected. There will also be degree effects such that the system can either be influenced (but retain its essential characteristics) or interrupted (you are no longer a device that can perform that particular perception/action task). Thus, there are two predicted hallmarks of a perceptual system of neural activity: 1) some direct interface with an energy array and 2) stability against influence from other networks.

It is worth noting that this characterisation applies to a trained system - one that has already learned to effectively detect the relevant perceptual information (reflected in the long-term changes mentioned above). During learning, a source of instability in a perceptual network will be the organism's ability to reliably perceive the information variable.

Conception: Dreaming is one of the purest examples of conception. The perceptual world intrudes on dreams from time to time but in dreams we are more cut off from perceptual information than at any other period of state of consciousness. Compared to waking life, dreams are highly unstable. Locations appear to change in an instant, events do not necessarily follow any sensible causal sequence, impossible things occur.

Leaving aside these qualitative observations, what would be the expected properties of a network that was not directly interfacing with a stable external environment? At a guess, it seems like the stability of such a network would depend on whether or not a person has a history of activating this network (again reflected in long-term neural changes as a consequence of learning), whether the network has been activated recently (medium-term changes reflecting residual activation of the same or similar networks), and whether there are other competing networks of activity that might impinge. Without structured neural activity driven by the perception of perceptual information, conceptual networks loose an important source of stability. Even in the highly practiced domain of language, consider the differential stability of talking to oneself (without sub-articulating) and speaking aloud. Even so, language confers some important stability to conceptual networks, in that we can use words to maintain an idea in the absence of external perceptual information about that idea. Generally speaking though, conceptual networks will be relatively susceptible to perturbation. Thus, the predicted hallmarks of a conceptual system of neural activity are: 1) no direct interface with an external energy array and 2) instability to interference from other networks.

If it were possible to peer into an active brain and observe its activity in great detail, it would be possible to distinguish between perception and conception on the basis of the above descriptions. Although this is not possible, I think it is worth considering this distinction in light of another issue - how can we tell whether we have direct access to meaning (DAtM) via perception or whether the detection of perceptual information is associated with a meaningful representation? This issue has cropped up several times in comments to previous posts. Andrew and I have both expressed some concern about what type of evidence could distinguish one claim from another. Are they simply different ways of saying the same thing?

I will attempt to outline how these accounts might differ in the language of perception and conception described above. At this point I am not completely confident that I am not just re-describing the problem. But, I think there might be an opportunity here to use network metrics to disambiguate DAtM from association.

Association implies two or more things paired together (a representation of a tone becomes linked to a representation of food), whereas DAtM implies one thing with a higher order interpretation (tone means food). On the basis of these crude descriptions it seems like one could sensibly predict association and DAtM to be distinguished by their instantiation in networks of neural activity. Take a case where a rat has been extensively conditioned to expect a food to follow presentation of a tone:

For association: During training, the rat would experience two consecutive, independent networks of perceptual neural activity - one for the tone and one for the food (I'm obviously oversimplifying by referring to a single network of activity about food, but I want to avoid details like that for now). The activation of the tone network reliably predicts the activation of the food network. Overtime, basic associative neural learning mechanisms begin to reflect the statistical reliability of the tone / food pairing. In a trained rat one would observe a coherent system of perceptual neural activation (perception of a tone) causing the activation of a separate system of conceptual neural activation (thinking about food). These two systems should be relatively independent, except to the extent that the perceptual system causes the conceptual one to fire.

For DAtM: During training the rat would experience the continuous evolution of a perceptual network that is first driven by auditory tone information and is subsequently driven by visual and taste information about food. In the interim between tone and food the network would begin to loose coherence, but stability would return with the presentation of visual and taste information from the food. As I argued earlier, I do not think it makes sense to talk about discrete perceptual networks that fire the same way regardless of what else is happening. I think it is more appropriate to focus on transformation from one type of network into another based on perceptual demands. By this account, the rat is experiencing a single compound event (food -> tone) with a predictable dynamic. A key idea of event perception is that experiencing part of the event can give you direct access to the rest of that event. So, for example, hearing a bar of your favourite song gives you access to the rest of that song, not through association but by the activation of a single event structure with a coherent spatio-temporal dynamic. If this logic is extended to the current example, for a trained rat, experiencing the tone could give him direct access to the entire tone -> food compound event. In contrast to the association case, one would not observe two independent networks, but the evolution of a single network.

Still, one might argue that these two cases are very similar and might be indistinguishable. Both start with a network of activation reverberating with the perceptual information of a tone. But, the DAtM case involves the activation of a single (albeit complex) event structure. Therefore, the observed activity should exhibit the characteristics of a coherent perceptual network. On the other hand, the association case involves the activation of a perceptual network and an independent conceptual network. Thus, the observed activity will differ from the DAtM case because a conceptual network about food will be relatively unstable, susceptible to perturbation, and disconnected from an interface with the environment.

This is a very early first swing at the problem and I am admittedly neither a neuroscientist nor an expert in complex networks. I have tried to make minimal assumptions about both subjects in the hope that what I've suggested is at least somewhat plausible. To a large extent, I agree with Andrew saying that he's tired of having the argument and would like to move onto thinking about new positive ways to approach the problem. But, I'm also a little bothered by the idea that we can't figure out what type of evidence distinguishes an ecological account from a representational one, or a DAtM account from an associative one. So, in this post I've tried to start at the bottom, and hopefully later I can work my way up to making sensible predictions about what behavioural differences we'd expect.

27 comments:

  1. Hi, Sabrina -

    Yes, notwithstanding that I never comment, I still read almost all of your and Andrew's posts. Your introducing me to the ecological approach has influenced my thinking about language processing (my major interest) and I continue to get insights - and encouragement - from posts like the recent ones that seem to support some of my own off-the-wall ideas. Eg, your "evolving networks", which is essentially the way I have come to model (in my mind) language processing, complete with an analog to your "one bar -> whole song" inference (more on that below). I am, however, a bit skeptical about "concepts", and wonder if you think the following makes any sense.

    Recall the problem we had in my introduction to your world about the shark's "edible" affordance (in an exchange at Ken' Aizawa's blog), Andrew's would-be wall-scaler's "climbable" affordance, and the would-be fly-ball catcher's "catchable" (and/or "trackable") affordance. At the time I suggested that perhaps the problem was misidentifying what constitutes the (by definition) immediately actualizable affordance: eg, ingestible rather than edible, approachable rather than climbable, and trackable rather than catchable. It occurred to me while reading your and Pam's recent exchange that this perhaps could be generalized to the statement that the actualization of any affordance necessarily results in another affordance, leading to a sequence (more accurately, I think, a continuum) of affordances ad infinitum (well, at least ad mortem).

    This appears to satisfy my desire to get rid of concepts (of which I take "food" to be an example, although your discussion of dreams makes me wonder if we have the same concept of "concept"). In my model, the rat is busy actualizing affordances (AKA, moving on a "scurry-about-able" surface) when the sensed tone combines with the sensed environment to create an affordance. From Pam's description I can't tell what this is, but I assume it's something like a lever affording pushable or a door affording "scurry-through-able". The rat has learned that actualizing the newly perceived affordance often results in access to an object-surface (AKA "food") which affords approachable, actualization of which creates the affordance ingestible, etc, etc. (Not, of course, the complete sequence, but you get the idea.)
    ============= cont'd ======================

    ReplyDelete
  2. Are you suggesting that your ecological event structures are entirely non-additive? So the structure for Tone and the structure for Food bear little relation to the structure for Tone+Food? Would that give you a way to distinguish the hypotheses?

    ReplyDelete
  3. =========== cont'd from above ========

    In order to eliminate your "conceptual" networks, I need to do (at least) two things: account for dreams (in general, "occurrent thoughts"??) and account for processing that results in a latent action that isn't immediately executed in order to actualize an affordance.

    The former is a direct consequence of expanding the "perceptual" network to include all stimuli whether from perceptions or from internal sources such as emotions and memories, This seems to me to be the natural way of viewing what I take to be our essential nature as not just perception-action systems but more generally stimulus-response systems (very complex systems involving multiple complex stimuli, context-dependent stored responses which are not necessarily immediately executed, various feedback loops, et al).

    The latter follows from the assumption that the network supports "stored" responses to familiar stimuli - responses that in any given context may or may not be executed. Completing a familiar song is one example of a stored response which is executed; in language processing, finishing someone's sentence for them is another. Such examples can be viewed as progressive pattern matching tasks that complete once a stimulus has progressed to the point where the number of stored candidate responses is reduced to one (not necessarily correct, of course - at least in the language example). Put into your language, the sensory neural activity patterns resulting from an aural stimulus combine in the network with stored motor neuron excitations to execute a response that actualizes the "completable" affordance resulting from the aural stimulus.

    I'm a bit unclear on your concept of "meaning", but I think we may be pretty closely aligned there as well. I've come to think of the "meaning" of an utterance as the responsive action the speaker intends to evoke in the hearer (AKA, actualization of the affordance the speaker's words are intended to create). The hearer extracts "information" from the perceived aural stimulus (et al) resulting in a certain responsive action (the "meaning" attached by the hearer to that information). If the hearer's action is that intended by the speaker, the latter has communicated the desired "meaning" to the former.

    Hope this isn't too far off. (It's clearly too long!!)

    ReplyDelete
  4. Hi Charles,
    I'm glad you're still reading the blog! You've raised a number of interesting points and it might take me a day or two to get together an adequate response, but I'm working on it...

    ReplyDelete
  5. Anonymous: Yes, I would predict that the event structures are non-additive, mostly because I don't think there is any such thing as context-free perception or conception. Coming to perceive food after the perception of a tone might lead to a different type of network than coming to perceive food following the perception of an electric shock, for instance. The categorisation literature is full of examples of how context biases what we notice about the world. The traditional cognitive view is that there is a core conceptual structure that is activated each time, which can be subsequently modified by additional context-specific processes. I would argue that instead of modifying a stable conceptual structure, context changes what we perceive in the first place. And yes, I hope that this might provide some basis for distinguishing between the different accounts discussed in the post.

    ReplyDelete
  6. 3 quick thoughts on original post:

    1) I really like the distinction between the different types of networks, but think you might want new terms. What you are talking about seems far from the original meanings of 'perceptual' and 'conceptual'. Note, James kept the original meanings, but asserted the contents of the categories were reversed from is typically thought.

    2) I'm not sure about the 'association' vs. 'DAtM' distinction. Behaviorist theories are (well... should be) first theories about the structure of the world. Things do not need to be associated in the head, because they ARE associated in the world. Thus, Pavlov's dog does not need a mental representation of the bell mentally-associated-with a mental representation of the food - the bell and the food have been associated by Pavlov, and the dog only needs to drool in response to the sound. In such a context, what Eco-Psych offers is a much better description of the thing responded to and the method of response.

    3) What distinguishes the Eco-Psych approach most obviously to me is the direction of investigation / the types of phenomenon predicted. I am increasingly suspicious that a crucial experiment of any kind can differentiate the Ecological approach from the Cognitive approach in a broad sense. My selective background in the history of science suggests that crucial experiments are only declared so after the fact, once everyone has already made up their mind anyway (the classic example being the Michaelson-Morley experiments which were claimed to disproved the existence of aether wind... but not until decades after publication). Thus, historically speaking, scientific wars are won and lost without crucial experiments.

    ReplyDelete
  7. We are (or at least I am) used to thinking about affordances in terms of visual stimuli so that it seems natural to talk of "seeing" an affordance offered by a surface. But presumably, the general case is any stimulus - or more to the point, combination of stimuli - that contain information sufficient for the subject to identify some feature of the environment as offering an affordance (which I see as an invitation to purposeful action).

    Looking at it this way resolves a problem I had with the tone-lever-food connection. In the absence of the tone, the lever doesn't necessarily invite purposeful action, ie, doesn't offer an affordance. But as Eric notes, the rat has learned to associate the tone with a purpose - getting access to something offering "ingestible" - so that the tone's sounding results in the lever's otherwise purposeless offer of "graspable" becoming purposeful, and hence an affordance.

    Viewing the scenario this way, I see no need for representations of either the tone source or a specific "ingestible" object - ie, for concepts.

    I think this view helps with the other problem scenarios mentioned above. Eg, the subject sees reaching a surface above a wall as a goal, achievable by actualizing a sequence (actually, continuum) of affordance actualizations, among which are (essentially) traversing a walkable surface to reach a ladder, which is then carry-able to where it becomes positionable so as to make the wall scalable, thereby rendering the surface above the wall step-onto-able.

    ReplyDelete
  8. Some comments I've made to Sabrina but that we thought might be worth making in public :)

    1. I like the idea of 'some kind of fidelity' between information and the resulting neural reverberation (a word I think we're going to try to use a lot, it seems useful!). I'd never really given it much thought but it is, of course, true. This fidelity is not the same is 'identical to', simply 'specific to', in the same way Gibsonian information doesn't have to be identical to the thing it specifies, just specific.

    2. A general sort of rule of thumb about perturbations in dynamical systems is that they have to be the right kind rather than the right amount. So 'stability in the face of perturbations' gets clarified as 'patterns of stability in the face of certain kinds of perturbations' - see this post about my perturbation experiments where I lay this out a bit.

    These minor additions aside, I'm liking this framework.

    ReplyDelete
  9. Eric: I don't think we're looking for the slam dunk. But we're both concerned that, from the outside, we often sound like we're simply describing well known effects in our own language, doing nothing but adding a layer of unnecessary confusion. We are actually exploring the consequences of an entirely different way of doing business: one way to try and show that is empirically, finding a result that makes more sense interpreted our way rather than the other way. It breaks the apparent equivalence of the two descriptions.

    So the plan is 'add this method to the broader attack', not 'rely on the slam dunk', if that makes sense. We're both scientists at heart and while theory is taking us far, we both feel the need for some data on the issue :)

    ReplyDelete
  10. Charles, I've got a treat for you (and hopefully Andrew and Sabrina will appreciate it too):

    http://fixingpsychology.blogspot.com/

    ReplyDelete
  11. Hi Charles,
    If you’re interested in language processing stay tuned, because I’m going to try to expand this framework to talk about language as well.
    I am also struggling a bit to come up with a defensible framework for concepts and, as I mentioned, this is only a first swing.

    I agree with you that it’s important to be specific about the language used to describe immediately actualisable affordances and that effecting one affordance changes the landscape of subsequently available affordances. So, applying this to the rat: Yes, absolutely, during training (when the food is physically present following the tone) the rat is actualising affordances and there is no reason to invoke concepts at all. In this post I was describing a simple conditioning example where the tone and food were merely paired (i.e., the rat didn’t have to push a lever or anything to get a reward, the food just appeared). After training presenting the rat with a tone leads it (on some level) to expect food – e.g., the rat salivates. In this case, the question is about how the rat has access to information about the food. The traditional learning account describes this in terms of the association between the perceptual tone stimulus and the conceptual representation of food – the tone activates the food representation. I argued that perhaps the tone and food were learned as a coherent event structure so that the presentation of the tone gave the rate access to the rest of the network, which also contained information about food.

    cont'd

    ReplyDelete
  12. The example you mention is a more complicated case because it involves operant conditioning – the rat has to learn the consequences of a particular behaviour (e.g., lever press usually leads to food). I agree that to accomplish this task the rat must learn (or already have learned) to perceive the relevant affordance property of, let’s say, the lever. The trick is that this affordance property also has to acquire a new meaning (leads to food). This type of meaning is tricky to deal with in conventional ecological psych. Typically, people talk about information as having meaning because of the specification relationship between the structured energy array and the physical world. This type of meaning exists in the rat example as the relationship between the visual information that means “pressability” and the physical properties of the lever (this is underdefined, but you get the idea). But the rat example invokes an additional level of meaning – one that does not have a specification relationship with the perceptual information. There is no necessary relationship between a lever press and food – the lever press could just as easily have led to an electric shock. Obviously, the rat somehow learns that the lever press means food. But, the question is, how is this meaning instantiated? If I understand you correctly, you are suggesting that the rat learns that pressing a lever opens up a possible sequence of affordances that lead to eating some tasty food. I think this is a perfectly reasonable way to describe the process, but I still want to know at what level is the rat thinking about the food when he presses the lever. If the rat only has access to the next possible affordance in the chain, then it’s hard to see what his motivation is – it’s the eating of food that is rewarding. So, it seems to me that the rat has to somehow have access to information about eating the food when he presses the lever. As Pam mentioned, there are also experiments showing specific satiety effects, so the rat also has to have information beyond just “pressing the lever leads to good things.”
    In this case, I think we’re on the same page about “expanding the perceptual network to include all stimuli whether from perceptions or from internal sources.” So, here I would agree that the system is working as an extended perceptual network. But, I would resist throwing out conceptual networks entirely for the moment and here’s why: The brain can be up to more than one thing at a time (i.e., multiple networks of activation) and these things can operate fairly independently. To give a perceptual example, I can be engaged in detecting the perceptual information that lets me catch a football and I can also be aware of the football’s colour. Colour makes no difference at all to my ability to perform the catching task, so my perception of colour must at some level be independent of my perception of the action relevant information. I can’t think of any reason to rule out similar types of independent networks that are not directly connected to a perceptual system. So, I’m calling these conceptual networks.

    cont'd

    ReplyDelete
  13. I agree less with the notion of “stored responses to familiar stimuli.” From a perception/action angle, a major reason that we get better at performing tasks is that we become better at detecting the relevant information. This isn’t really a stored response, it’s more that we learn to pay attention to the right things. Another thing that improves performance is that we become better at coordinating between the information and our actions. Again, this doesn’t seem to be a type of stored response – it’s more that we become better task specific devices (or, in the language of this post, the relevant perceptual network becomes more coherent and stable). I would also argue against a stored response idea for language processing. This is because it is often possible to finish someone’s sentence even though the sentence itself is novel. To the language user, it will seem inevitable that a given word or phrase will come next, but stimulus/response pairing are not an ideal way to describe this given the variability of the stimulus. One way out of this maze is to assume that there is a higher order similarity between the beginning parts of the sentence such that the language user learns that a certain response likely follows a given category of sentence beginnings. But, this implies that there is actually a detectible perceptual similarity between the sentence beginnings. And, I don’t know at the moment how to articulate what this perceptual similarity might be.
    In terms of the meaning of language, we have the same problem as with the operant conditioning rat example – we need a disciplined way to think about perceptual information having an arbitrary, conventional meaning (or multiple meanings). Thinking of the purpose of communication – to get some response from the hearer – is a good start. I don’t think the language issue is insurmountable but it is tricky, and, as I said at the beginning, I hope to attempt an explanation in the next post or two!

    ReplyDelete
  14. Hi Eric,
    1) Any suggestions for new terms? I agree that I’m cutting across lines in a way that breaks the current rules of the perception/conception distinction. One of my goals is to emphasise how much information is actually available via perception and to head off some of the misunderstandings that often crop up when people assume that anything that looks like a decision or a category effect, etc, must be conceptual rather than perceptual.
    2) I agree that behaviourist theories are primarily about structure in the world but as I understand it this is because the purpose of the theories was to accurately predict behaviour rather than provide a mechanism for how an organism came to exhibit that behaviour. Taking traditional learning theories at face value, the question still remains, at what level does a trained animal have access to information about the consequences of his actions? From my recent discussions with Pam on this blog it seems like modern behaviourists are happy to describe this process in terms of forming an association between an action and an internal representation of, e.g., a food reward. I also agree that ecological psych offers a much better explanation. The motivation for this post, though, was to deal with the problem of these different approaches seeming to describe exactly the same thing using different language (DAtM vs association) without any idea about how to empirically distinguish those.
    3) So, that said, I take on board what you’re saying about the need (or lack of) for crucial experiments. It may well be the case that the way forward is to continue doing quality research in ecological psych and, hopefully, gradually changing the culture. However, ecological types do need to branch out if this is going to be effective. And that means thinking about how to approach other areas of psychology (e.g., cognitive) ecologically. The challenge is getting this research past editors and reviewers who are understandably reluctant to have their basic approach to their science questioned. As a non-representational cognitive psychologist I am particularly sensitive to this issue!

    ReplyDelete
  15. I agree that to accomplish this task the rat must learn (or already have learned) to perceive the relevant affordance property of, let’s say, the lever. The trick is that this affordance property also has to acquire a new meaning (leads to food). This type of meaning is tricky to deal with in conventional ecological psych. Typically, people talk about information as having meaning because of the specification relationship between the structured energy array and the physical world. This type of meaning exists in the rat example as the relationship between the visual information that means “pressability” and the physical properties of the lever (this is underdefined, but you get the idea). But the rat example invokes an additional level of meaning – one that does not have a specification relationship with the perceptual information. There is no necessary relationship between a lever press and food – the lever press could just as easily have led to an electric shock. Obviously, the rat somehow learns that the lever press means food.
    The specification relation is indeed typically critical, because that is the thing that guarantees outside of the lab that the information you detected is about the same event as last time, and this is critical for learning. However, while I think that skilled action critically requires specification to be as good as it is, I'm coming on board with the idea that, in principle, if a relation is good enough, it will support behaviour.

    The reason I'm a little ok with this is that, in all cases, the organism can never 'peek behind the curtain' and thus can't tell the difference between a stimulus created by a specification relation and one created by, say, a situation. If the situation supports sufficient reliability, an organism can come to use it as genuine information.

    If something is sufficiently reliable, then you can spend time getting better at detecting it (perceptual learning) and coming to understand what it means by using it to control action. We typically talk about learning to detect time-to-contact information and learning that it means this by using it to control interceptive actions. You could easily, with the same mechanism, talk about learning to detect the information for the lever and coming to learn that it means food, by using that information to get food, rather than merely to press the lever.

    So I say 'if something is sufficiently reliable'. I maintain that this is a big if, out in the world; possibly too big for anything other than specification to ultimately work. In the lab, of course, it's less of a problem. So the rat can come to learn that the information about the lever means food, because it does; the higher order event structure is actually there to support this, in the same way time-to-contact means what it does.

    ReplyDelete
  16. Sabrina,
    I wish I had better terms to suggest.* One of the big lessons from 'pragmatism' is that there is no clear line between different cognitive phenomenon (e.g., perceiving and thinking). The problem is worse than the boundary being fuzzy. I am also very sensitive to the 'getting past editors' issue. In a recent past-life, I was a non-representational developmental psychologist. You can imagine how long it took to get my dissertation published (http://onlinelibrary.wiley.com/doi/10.1111/j.1467-7687.2009.00844.x/full) and I have two other developmental psychology articles that are still desperately looking for a home. Yes, ecological psychologists must 'branch out' and it is difficult. It is problematic that, after all this time, there are not more ecological psychologists on the editorial boards of major journals, at grant funding agencies, etc. In part I blame the overly adversarial attitude that (perhaps necessarily) characterized the early development of the field.

    I'm skipping the second point, I know, but the quest for a coherent behaviorism is a much longer discussion. Certainly, though, a good behaviorist should be as averse to metaphorical representation-talk as any ecological psychologist. (should)

    I like the direction the original post was going in. My comments were purely about the framing.

    *Occurred to me at the end of typing, that if the term "experience" didn't carry so much baggage, a good set of terms would be: Perceptually-grounded experience vs. Perception-independent experience.

    ReplyDelete
  17. There are still proper behaviourists out there - I work in the area of contemporary learning theories, not behaviourism - this is important to some people ;)


    now to a couple of the points raised above:

    2) I'm not sure about the 'association' vs. 'DAtM' distinction. Behaviorist theories are (well... should be) first theories about the structure of the world. Things do not need to be associated in the head, because they ARE associated in the world. Thus, Pavlov's dog does not need a mental representation of the bell mentally-associated-with a mental representation of the food - the bell and the food have been associated by Pavlov, and the dog only needs to drool in response to the sound. In such a context, what Eco-Psych offers is a much better description of the thing responded to and the method of response.

    As I understand it, Eco-Psych doesn't go beyond a simple SR approach to explain conditioning though, and there are several well established learning phenomena that don't fit that.
    e.g. sensory specific satiety - sate the dog, and he won't drool any more.
    e.g. sensory preconditioning. Pair a tone and a light. Then pair light and food. The animal will respond to the tone, as if it had been paired with food.

    --
    Animals are very sensitive to the information in the environment - they respond differently under ratio schedules under interval schedules, for example. They also are able to flexibly alter their behaviour, when they need to - and its this that I struggle to see how eco-psych explains.


    But as Eric notes, the rat has learned to associate the tone with a purpose - getting access to something offering "ingestible" - so that the tone's sounding results in the lever's otherwise purposeless offer of "graspable" becoming purposeful, and hence an affordance.

    Viewing the scenario this way, I see no need for representations of either the tone source or a specific "ingestible" object - ie, for concepts.


    By what process does this happen? It sounds a little like magic to me?

    One of the things I like about your approach is the parsimony - but when it becomes overly descriptive I struggle, because I can't see how this is a more parsimonious approach than one that invokes representations.

    Sorry this isn't quite as coherent as it might be...

    ReplyDelete
  18. Eric said:

    One of the big lessons from 'pragmatism' is that there is no clear line between different cognitive phenomena

    Which motivates my general posture - attempting to reduce behavior to pure responsive action, possibly along the lines of Rorty's reduction of phil of language to "pure Davidsonian semantics" (not to suggest that I am totally clear on what he means by that).

    Here, that leads to an attempt to reduce, if not eliminate, the role of the "food" concept. Note that I stop my description of the rat's behavior at "ingestible", by which I mean only "swallowable". I see the rat as having learned to respond to the tone by executing a composite action that often leads to that affordance. If an object offering that affordance is thereby reached, whether it is "edible" seems a separable issues. (In fact, I don't even consider that an affordance.) It's approaching noon, so I head for the kitchen - not because I'm salivating but because my watch is offering read-time-from-able. I'm just executing a "stored" procedure that may or may not result in access to any food. (My wife does the shopping, so I often have no idea which it will be. And she might find the analogy apt in other respects!)

    On to networks. Although "reverberation" seems to be on the right track, a better term might be "resonance". One can envision Sabrina's perceptual network as constituting an adaptive filter bank. The learning process would be to "grow" the filter bank so that a new member would pass a specific perceived input signal (ie, the neural activity consequent to sensory input). Then in a sense to be described, part of the filter bank would become "resonant with" that signal. The output from a specific filter would excite a part of the network that would activate motor neurons that in turn initiate an appropriate responsive action.

    There are several features of Sabrina's network view that make it a much better way of thinking about what I had in mind by "progressive pattern matching". It is inherently continuous; it provides a natural connection between sensory input and motor output; which in turn creates the feedback loop needed for adaptive behavior, eg, smooth tracking; and finally, it eliminates the need for precise "pattern matching". A so-called "matched filter" can be thought of as doing a correlation between the received signal and a stored version of a signal to be detected. But such a filter will respond - "resonate" - to any input signal that isn't "too far removed" from the signal to which it is matched. The response will be attenuated relative to the response of the matched signal, but can still activate a response that is "good enough".

    This view also arguably works with dreams. A matched filter can resonate in response to signals that have no meaningful relationship to the signal to which it is matched. Thus, dreams could result from outputs of the filter bank due to arbitrary inputs (AKA, "noise") spontaneously generated in the sensory input path. Usually, the responses would not be actualized as motion, although some can be - talking in one's sleep and, in extreme cases, sleepwalking.

    I agree that "stored response" is a poor way of describing the process and didn't mean to suggest that the system has no adaptive aspect. In any event, since Sabrina's filter view arguably responds even to inputs that "aren't quite right" and incorporates feedback, that issue more-or-less goes away.

    ReplyDelete
  19. Pam,
    As I read Gibson, he is agnostic about most of the core issues of behaviorism. He had opinions to be sure, as he was mentored by one of the greatest of the early behaviorists (or at least Holt was so known in his day), and his system assumed a great deal of what behaviorist have always arguing for, but he chose not to make that his fight. Gibson wanted to know how we perceived things, and thought perception was intimately related to action, but he offered no theory about why animals sometimes act one way and other times act differently. On the other hand, learning theory is primarily about why organisms act one way sometimes and other ways other times. Though I don't know quite how to pull off the magic, I am convinced that the approaches are complementary, rather than adversarial.

    You are right though, that some of the most influential ecological psychologists (e.g. Turvey & Shaw) have proffer something much closer to the old S-R behaviorism. Though they would never admit that this is what they have offered, they have, and you are also correct that it is problematic.

    ReplyDelete
  20. Thus, dreams could result from outputs of the filter bank due to arbitrary inputs (AKA, "noise") spontaneously generated in the sensory input path. Usually, the responses would not be actualized as motion, although some can be - talking in one's sleep and, in extreme cases, sleepwalking.
    As a side note to this from Charles, one of my favourite things that I know about dreaming is that the motor system from the neck down is very actively inhibited during dreaming, and failures of this inhibition are often disastrous because the behaviour isn't connected to the world (I don't think sleepwalking occurs during REM). Evolution clearly noticed it was important to utterly lock down a system, and it always makes me realise how important good access to information is.

    The other cool thing I know about sleep is that dolphins sleep one hemisphere at a time, so they can continue to move in the water safely under active control. I draw similar conclusions there too :)

    ReplyDelete
  21. Pam -

    Sorry, but I'm not familiar with the relevant lingo. Could you elaborate "Pair a tone and a light"? I assume "pair a light and food" means training the subject to take an action when a light event occurs, an action that will result in access to food. But it's not obvious to me how to adapt that description to "pair a tone and light".

    By what process does this happen?

    I'm back to being confused about how purposes or goals enter the picture, so I have no answer. Maybe it is by magic!

    ReplyDelete
  22. Charles,
    To "pair a light and food" means to present them at the same time (typically one presents the light immediately before the food). Pam is pointing out that her original claim was about "classical condition" (Pavlovian drooling-dog stuff) rather than "operant conditioning" (Skinnerian rat-pressing-lever stuff). To be more specific, if you are "pairing" then the behavior of the organism does NOT cause the outcome.

    As for purposes or goals, this is a problematic discussion. By some very reasonable reads, behaviorism / learning theory IS the modern science of purpose. A rat that works to press a lever is a rat that want's to press a lever. I don't know about Rorty, but Peirce would agree. The behaviorists built a science about how a rat that does not want to do certain thing becomes a rat that does want to do certain things. Alas, with a few notable exceptions, they eschewed the words 'want' or 'purpose'. (Note, I once again cheated by switching to an operant situation.) This has lead to all sorts of confusion, for example, you will not find learning theory mentioned in an introductory psychology textbook's chapter on motivation. Uhg!

    P.S. Pam, you made new blog as well, see the comments: http://fixingpsychology.blogspot.com

    ReplyDelete
  23. pairing = presenting one after the other.

    So put rat in skinner box. play tone for 10 s, when tone goes off, turn on light for 10s, then turn off. then intertrial interval, (eg 2 mins). repeat x 20.

    the animal has to make no responses during this training. no food is present.

    then in a separate session, present light for 10 s, followed by food, then ITI, repeat x 20. animal has to make no responses - there is no need for it to make any actions during the light to gain the reward.
    it tends to collect the food from the place it is delivered, and as it learns the predictive relationship between the light and the food, it will make anticipatory nose pokes into the place the food is delivered.

    then in a test session, the tone is presented with no consequence, and the animal will nose poke during the tone presentations more frequently than during presentation of a control stimulus (which it has had equal exposure to, but hasn't been presented follow by the light).

    ReplyDelete
  24. Eric and Pam -

    Thanks for the explanation. I of course have a top level idea about how conditioning works, but no idea at all what the detailed procedures are like. Those responses help a lot.

    So, what about the following. Suppose Sabrina's perceptual network is more or less accurately modeled by a filter bank. Then one can hypothesize that the learning process that pairs tone and light causes the simultaneous "growth" in the network of light and tone "matched filters" that are effectively in parallel but which have no output, ie, no connections to motor neurons, Then in the learning process that pairs tone and food, the resulting output connections might be shared by default by the two filters so that their sensory neuronal inputs are effectively logically "OR"ed with respect to producing a common motor neuronal output. That seems to produce the observed behavior (as I understand it).

    If I'm right in guessing that a relevant behavior included in "specific satiety effects" is the subject's not responding to the tone when not hungry, that seems straightforward as well (at least in the simplistic model). Being hungry is the kind of input included when you "expand[] the 'perceptual' network to include all stimuli whether from perceptions or from internal sources such as emotions and memories". Hunger would simply be an input to the network that could in principle be used to gate the parallel tone and light matched filters - or their inputs - on and off.

    ReplyDelete
  25. Hi Charles -
    not sure what you mean by a filter bank - but what you are describing sounds a little to me like an associative model!
    There are two types of satiety effect - one is the reduction in response when not hungry, and one is the reduction in response when sated on a specific reward.
    So if a tone predicts sucrose solution, and a light predicts a savory food pellet, when the rat has had access to the sucrose solution for an hour before a test session (which is conducted with no rewards present), it will continue to respond during the light, but will no longer respond during the tone.
    That seems to require something more sophisticated than a hunger 'gate'. It involves encoding the current value of the reward (is it still rewarding?).

    A similar experiment can look at learning about flavours. Animals are given a drinking tube of salty vanilla water, say 10 mls a day, for 4 days. They are also given a tube (at a different time of day) of 10ml of almond flavoured water. you can then give an injection that induces a salt appetite - so salty things are very valued. They will then choose to drink vanilla (no salt present) water more than almond water.

    I agree that our approaches can be complimentary - and I'm tremendously enjoying trying to get my head around your approach.

    Incidentally, I came across a lovely example similar to the steam governor - my son has built a maplin electronics kit to make a 'bug' that follows light. it has two LDRs, one on each side, that are connected (via transistors) to motors. So when the light shines, that turns on the motor on that side, making the bug move towards the light. No "representation" required, but if you stuck in an electrode, you might interpret the electric signal as a representation. Food for thought re. neural recordings...

    ReplyDelete
  26. Pam, those robot kits were probably inspired by some robotics work by Pfifer and Scheier (link to their book) done explicitly to show you could get complex behaviour emerging from simple systems embodied in particular ways. You should try adjusting the spacing between the sensors, you'll get interesting changes in behaviour with no change to the 'neural' control system!

    ReplyDelete
  27. So. assume the light tracker works as follows. In the absence of light, it just idles. When a light source appears, the wheels are enabled and the device tracks the light source. Ie, it has "light-light location" pairing (by design).

    Add a pair of "TDRs" that cause analogous behavior with respect to a tone source, and the device then has "tone-tone location" pairing as well (again, by design). In essence, the device's "brain" (or network) becomes a two element "filter bank" that responds differentially to sensory inputs. Its input signal can be either an aural frequency or a visual frequency and the output of the filter that "resonates" drives the wheel motors so as to track the active source.

    Now imagine that the device is "smart" enough to learn a drive pattern for the wheels that causes the device to move from a fixed location to the location of the light source (now assumed to be fixed as well). Then the device can be rewired so that the tone activates both the wheels and the learned drive pattern. That rewiring seems akin to "tone-light source" pairing.

    Something like "satiety gating" could be effected by measuring the total "amount" of sensory input (tone or light) over a time interval and inhibiting the wheel enabling if that amount exceeds a threshold. (Many common communication devices have an even more sophisticated version called "automatic gain control" - somewhat analogous to maintaining a constant level of hunger in the rat.)

    Not a perfect analogy, but perhaps it helps to relate electronics lingo and psych lingo.

    ReplyDelete