Friday 23 April 2010

Reading Group - Gibson (1979) Chapter 4

In this chapter Gibson really steps up the pace, so hold onto your hat. First, he shows that perception can’t be based on sensation. Then he shows that we can’t see light. Finally, he shows that we don’t see images. Here we go...

Chapter 4: The relationship between stimulation and stimulus information

The first three chapters describe the environment (see here, here, and here). Now, Gibson describes the information available in the environment. For visual perception, this information comes from ecological optics. Ecological optics is not the same thing as either geometric or physiological optics.

The distinction between luminous and illuminated bodies

Luminous bodies emit light. The sun, a desk lamp, and a fire are luminous. Non-luminous bodies do not emit light, and most things in our environment are non-luminous. We can often see non-luminous bodies because they are illuminated by ambient light. We can tell the difference between luminous and illuminated bodies.

The distinction between radiation and illumination

Radiant light travels from the sun to the earth in parallel rays. Some of this light is scattered in the atmosphere. When the light hits the ground it is further broken up by scatter reflection. This scatter-reflected light bounces between the sky and the ground until the ray is completely dispersed. It is because of this property that things are illuminated even if they are not directly under the sun. For instance, a room with windows is illuminated during the day. So, radiant light travels to earth in nice parallel rays, and illumination occurs when these rays are scattered through their interaction with the ground, etc. Ambient light is a consequence of illuminated light.

The distinction between radiant light and ambient light

Light rays scatter so completely and quickly that light rays completely fill the air. “Each point in the air [is] a point of intersection of rays coming from all directions. It would follow that light is ambient at every point” (p. 51). The amount of energy from light in a space changes all the time as a consequence of changes in the source, like when the sun goes behind a cloud. However, the structure in the reverberation of light is preserved through these changes. This structure is a more reliable source of information about the environment than the absolute flow of energy.

“Radiant light from a point source is not different in different directions; ambient light at a point is different in different directions. Radiant light has no structure; ambient light has structure. Radiant light is propagated; ambient light is not, it is simply there. Radiant light comes from atoms and returns to atoms; ambient light depends on an environment of surfaces. Radiant light is energy; ambient light can be information” (p. 51).

The structuring of ambient light

The ambient optic array refers to ambient light with structure. The structure mostly comes from differences in ambient light intensity coming from different directions. Without structure there is no information. For instance, during a white-out, light reflects between tightly packed particles of snow to create more or less homogeneous illumination. Being in a situation like this is stupidly disorienting. Structure rules.

Stimulation and stimulus information

If an eye (hopefully attached to an animal or else that’s just creepy) intercepts some ambient light, then photons will stimulate photoreceptors in the retina. In a white-out, the stimulation will be uniform and the observer won’t be able to perceive anything. For perception to occur there must be structure in the light. This means that stimulation doesn’t lead to perception. In people, our two mobile eyes, set in a turn-able head, on top of a movable body comprise a perceptual system (the visual system). This system is activated in the presence of stimulus information. This construal of the visual system essentially abandons the old idea of sensation. Sensation (e.g., stimulation of photoreceptors) doesn’t tell us anything about the world. Sensation cannot be the basis of perception.

Do we ever see light as such?

The sensation-based theories of perception assert that all we can ever see is light. Light is all that gets into the eye (in the form of rays) via the retina, so this must be what we see. This would mean that we don’t see the environment directly, that we don’t see objects or surfaces. The other possibility is that we never see light, which sounds crazy. But, consider, if we see a single point of light in a dark room we don’t think “My, look at that point of light,” we think “Oh, there must be a light source somewhere in the distance.” We don’t see a beam of light in the air unless it illuminates particles in the medium. We don’t see light that fills empty space – we only see things that are illuminated. So, Gibson concludes that “all we ever see is the environment or facts about the environment, never photons or waves or radiant energy” (p. 55). The stimulation of receptors in the retina is not the data of perception – it is not sufficient for perception.

The concept of the stimulus as an application of energy

People use the word “stimuli” to mean all sorts of things. The original physiological sense of the word meant something that fires a nerve cell, or triggers a reflex arc. So, something that caused a photoreceptor to fire, that caused a sensation, would be a stimulus. Psychology generalised this meaning to include something that causes a response, as well as a sensation. However, we respond to lots of things, which means that we can call pretty much anything a stimulus. This makes it a pretty useless designation. For clarity, Gibson uses only the physiological meaning of “stimuli.” By this definition, any type of energy that causes a response, in a receptor, say, is a stimulus. Photoreceptors are very sensitive to light energy, but other types of energy will cause a response as well. This is also true for chemorecptors (in the nose) and mechanoreceptors (in the skin). So, stimulation cannot tell us anything about the source of the energy – it carries no information about the world. Stimulation is also temporary, and cells quickly adapt to sensation and decrease their firing rate. But, our experience tells us that objects in the world persist. This means that objects cannot be specified by sensations because sensations do not persist. Also, according to the physiological definition used by Gibson, objects can’t be stimuli (although this is how psychologists tend to use the term). Photons are stimuli, not the objects from which they reflect. For visual perception, the task is to find invariants in the changing structure of the optic flow.

The orthodox theory of the retinal image

The story goes like this: We form images of objects in the back of the eye, like a camera. All objects reflect light from all directions – in effect, they are covered in tiny sources of light. These rays enter the eye and are focused onto the back of the eye. This set of focus points is the retinal image, which connects object and image, point by point. So, psychologists tend to think of the stimulation of photoreceptors in the retina as an image projected on a screen. Gibson points out an interesting entailment of this conception: If the stimulation of receptors constitutes an image, then who looks at that image? Why would an image inside our eye be useful? Presumably images are meant to be seen with eyes, and putting the thing to be seen inside the eye itself seems like a dumb idea. Another, slightly more refined conception, is to think of an image as being transmitted, bit by bit, as signals along the optic nerve fibres. However, this gets back to the problem of sensation-based perception. If images are transferred as signals, bit by bit, then they would simply reflect the correspondence between brightness of things in the world and firing intensity or between colour and wavelength. This isn’t very useful and it places an enormous burden on the brain to infer what these signals could be about. And, the whole idea of the brain inferring things about signals entails something like a little man waiting to receive information from the outside world. In other words, this conception implies a separation between mind and body. Gibson’s main point is that we don’t need to think of anything being transmitted along the optic nerve during perception. The proper view is to construe vision as a perceptual system engaged in a circular process – not a one way transmission. What does this system do? It “registers invariants in the structure of ambient light”, that’s what (p. 61).

A demonstration that the retinal image is not necessary for vision

Insects have a completely different type of eye than mammals. I’m sure everyone’s heard of the compound eye, but it’s worth taking a second to appreciate how different it is from our type of eye. Unlike our eyes, which have a lens, and a sensory surface (which many people interpret as acting like a screen), insect eyes have tightly packed receptive tubes, all pointing in different directions, like in this close-up picture of a fly eye. It is simply not possible to conceive of these eyes producing anything like an image. First, there is no “screen” to project anything on to. Second, all the points would end up jumbled. Despite this, insects see quite well. What their eyes are suited for is taking lots of samples of ambient light – precisely the kind of thing you’d want to do if visual perception entailed registering invariant structure in ambient light.

The concept of optical information

It’s important to distinguish perceptual information from the common use of the word “information”. We usually think that information is something conveyed via some medium. For instance, we can convey words via speech, books, email, etc. This type of information has senders (who transmit the signal) and receivers (who, um, receive the signal). Perceptual information is not conveyed and it doesn’t necessitate a sender and a receiver.  The world isn’t trying to communicate with us – “the world is specified in the structure of the light that reaches us, but it is entirely up to us to perceive it” (p. 63).

Got it? We don’t see images or light, and sensation isn’t the basis of perception. By focusing on perceptual information, Gibson removes an enormous burden from the brain, which no longer is expected to make complex inferences on the basis of under-specified sensations. He also sidesteps the pesky mind-body dualism that emerges from sensation-based accounts of perception. The next chapter goes into detail about the ambient optic array.

24 comments:

  1. Gibson has such a good turn of phrase: talking about light 'reverberating' around a room is such an evocative description that entirely captures what it is he wants to convey.

    “the world is specified in the structure of the light that reaches us, but it is entirely up to us to perceive it”
    Ha! I like this too :)

    ReplyDelete
  2. Thanks so much for this great summary.

    ReplyDelete
  3. "The sensation-based theories of perception assert that all we can ever see is light. Light is all that gets into the eye (in the form of rays) via the retina, so this must be what we see. "

    I'm sure that there must be some sensation-based theories that go like this, but then there are also others that merely say that we see things like table and chairs, but we perform computations on incoming patterns of light that enable us to see tables and chairs. These are indirect theories of perception, right?

    ReplyDelete
  4. Yes. But indirect perception is only warranted if your input is impoverished, ie if it's sensations. Any theory that requires computations, etc, is at heart a sensation-based theory of vision.

    ReplyDelete
  5. "Any theory that requires computations, etc, is at heart a sensation-based theory of vision."

    As usual, I need some vocabulary clarification. I assume "sensation" is equivalent to "sensory stimulation", ie, what occurs in sensory neurons. And with that understanding, I don't see how any theory of vision - or any other sense - can be other than "sensation-based". Based on one of your recent comments at BofC, I wonder if you are interpreting "sensation" as having something to do with the concept of "sense data". If so, given that Sellars killed - or at least seriously crippled - that idea half a centry ago, I don't see why.

    And although for convenience almost everyone uses digital computer terminology - processing, memory, database, computation, etc - and I do as well when talking about architectural issues, I always try to keep in mind that our physiology is analog. Hence, I might describe Gibson's "feedback architecture" (discussed at BofC in the context of system engineering) as "computational" or being a "processor" notwithstanding being fully aware that it's implementation is analog. Is there something more than that digital-analog distinction underlying your objection to "any theory that requires computations"?

    I'm happy to adopt any vocabulary, so I'm not arguing for one or another - just trying to establish a common one.

    ReplyDelete
  6. I assume "sensation" is equivalent to "sensory stimulation", ie, what occurs in sensory neurons. And with that understanding, I don't see how any theory of vision - or any other sense - can be other than "sensation-based".
    As Sabrina notes above, Gibson talks about this distinction in this chapter; stimulation of the receptors is necessary but not sufficient for visual perception. Obviously you have to have the energy interact with a receptor; but light is stimulation, not stimulus information. The example is a fog, or white out conditions: plenty of light, plenty of stimulation, but you literally perceive nothing (hence white out is such an incredibly dangerous situation).

    I wonder if you are interpreting "sensation" as having something to do with the concept of "sense data".
    I had a quick look at the SEP entry; Gibson actually explicitly rejects this idea in this chapter (pg 54), calling it a 'naive theory' :)

    Gibson is offering a third alternative to the current options of sensations or sense-data being the objects of perception: perceptual information. See the next posts on Chapter 5 for more on that (or Gibson 1966 and 1979, obviously)

    ...for convenience almost everyone uses digital computer terminology - processing, memory, database, computation, etc...
    But if it's a rubbish metaphor and we have alternatives (eg dynamical systems) why keep using it? Sure you can describe things representationally/computationally, but the question must always be, are you describing correctly?

    Is there something more than that digital-analog distinction underlying your objection to "any theory that requires computations"?
    My main objection is that computation is only ever necessary to invoke if the stimulus is impoverished and must be enriched to restore our perceptual experience to match the contents of the world. If you don't have this problem you don't need to do computation, or at the very least, not the kind of computation generally invoked.

    ReplyDelete
  7. Sabrina correctly notes that the popular sense of "information" as something conveyed from source to sink isn't relevant, but doesn't say what sense is relevant. So, let's see if we can agree on that.

    The technical sense of "information" is a measure of uncertainty about the outcome of a random event (random variable). Eg, before flipping a coin there are two possible outcomes. After the coin is flipped, an observer knows which outcome occurred, and that observer has become, shall we say, "epistemologically enriched" to the tune of one bit of information. This general idea can be expanded to more "information rich" events by considering random variables with more possible outcomes, and vectors or "arrays" of multiple random variables. But the basic idea of the acquisition of information as the resolution of uncertainty is the same. I assume that the term "perceptual information" is being used in this sense.

    Clearly, in order to apply this concept one has to define the random events in question and their statistics. Some discussion of such attempts are here:

    http://www.consciousentities.com/?p=292.

    I infer from Sabrina's description that you are envisioning current models of visual processing that involve events that are neural activity (sensations?) occurring somewhere in the processing chain, and that each event is assumed to be closely correlated with the stimulation of an individual receptor in the retina. (This inference is probably wrong since my understanding is that such a model would be based on a serious misunderstanding of visual processing). I infer that Gibson's model also assumes that the events are neural activity in the processing chain, but that each event is, so to speak, the result of "integrating" the results of stimulation of multiple receptors in a way that somehow captures the perceiver-centered geometry of the environment and changes consequent to movements within that environment. These inferences are mostly to motivate clarification of the differences in processing assumptions for the two models. I don't think comments to follow about processing will depend on the specifics of the processing model.

    ReplyDelete
  8. Actually Gibson specifically says that he's not talking about information a la Shannon. Gibson doesn't want perception to be a process of resolving ambiguity, because he hypothesises information variables which aren't ambiguous. Information consists of spatial-temporal structure in energy arrays (eg light, sound) which arise lawfully from objects and events in the world; it's information if that structure specifies that object or event (via natural law, cf the TSRM paper). So we aren't in Shannon territory.

    current models of visual processing that involve events that are neural activity (sensations?) occurring somewhere in the processing chain, and that each event is assumed to be closely correlated with the stimulation of an individual receptor in the retina.
    No - sensations are the proximal causes of the stimulation of individual receptors; the neural activity is the implementation of the computations required to turn those into a perceptual experience.

    I infer that Gibson's model also assumes that the events are neural activity in the processing chain, but that each event is, so to speak, the result of "integrating" the results of stimulation of multiple receptors in a way that somehow captures the perceiver-centered geometry of the environment and changes consequent to movements within that environment.
    Gibson had very little to say about the neural implementation of detecting information, because a) he was busy establishing that information made sense and b) he knew he had no basis for claims about implementation. That said, he didn't want it to be a 'construction' process: perception is not the end result of the brain re-assembling pieces, this smacks too much of the sensation-based theories he was rejecting.

    ReplyDelete
  9. Charles, some attempts to post your comment above ended up in the spam folder. They seem to be versions of what actually made it through; I won't let them through unless there was something missing. Blogger traps things sometimes, if your comment doesn't appear right away I'll catch it.

    ReplyDelete
  10. Andrew - I started with a comment that was rejected as being too long (who would ever have imagined!), and the rest were attempts to break it into postable pieces. Anyway, the part that got through is plenty for the moment. I'll do the rest after we work through this much.

    "sensations are the proximal causes of the stimulation of individual receptors"

    This may seem like quibbling, but I infer that Gibson was proposing a major upheaval in the way we look at the whole perception process, so I want to understand the process in his terms and in as much detail as possible. So, while I'm very happy to put "sensations" near the interface between stimulus and receptor, in order for "sensations" to be biological they presumably need to be on the output side of that interface, ie, a result of the stimulus rather than a cause. Then in the case of vision, the stimulus is impinging energy in the form of light, the receptors are rods and cones, and sensations are the electro-chemical outputs from those receptors. OK?

    FWIW, this is consistent with wiki's definition under "Sensation (psychology)":

    Sensation is the function of the low-level biochemical and neurological events that begin with the impinging of a stimulus upon the receptor cells of a sensory organ.

    (Although "the function of" seems superfluous and confusing.)

    I can sympathize with resisting ideas like "neural implementation of detecting information", "construction", and "the brain reassembling pieces" - my distinction between digital and analog was intended to address that issue. But even from the highest implementation-detail-free system perspective, there has to be an input and there has to be processing. So, if we agree that "sensations" are the most primitive biological entities available, how can all "sensation-based" theories be rejected? That's why I distinguished "sensations" and "sense-data"; the latter are widely rejected, but some concept of the former seems unavoidable for any reasonable definition of the word.

    ReplyDelete
  11. More on information. Besides informal uses of "information", the only sense of the word with which I'm familiar is the technical one (which I hesitate to call "Shannon information" because that phrase can get overloaded with the communication-specific baggage attendant to his famous work). So, I submit that if it is unquestionable that "information" in the technical sense is inapplicable to whatever in Gibson's scheme is the word's referent, then it would be preferable to choose a different, more descriptive word or phrase. Otherwise, its use seems yet another "rubbish metaphor".

    But just in case "information" in the formal sense might be applicable, please indulge what unfortunately will be pedantry if it unquestionably isn't. Information in the formal sense isn't about resolving ambiguity but rather uncertainty. Eg, interpreting a bistable illusion one-way or the other doesn't result in any information. But anytime there is a possible "random" (in the sense of unpredictable) event, there is a priori uncertainty, in which case a posteriori resolution does provide information. If the statistics of the possible outcomes aren't known, quantifying that information may be hard or even impossible, but that's a separate issue. And contrary to contemporary, digital computer-infused thinking, the events can be analog, ie, continuous. Hence, even if we view perception as a system the input of which is a continuous, time-varying, and unpredictable flow of sensory output (sensations), the processing of which can be modeled as an analog feedback loop, and the output of which is an analog stream of "perceptual experience", the analysis could conceivably include concepts of information, at least in principle.

    ReplyDelete
  12. So, while I'm very happy to put "sensations" near the interface between stimulus and receptor, in order for "sensations" to be biological they presumably need to be on the output side of that interface, ie, a result of the stimulus rather than a cause. Then in the case of vision, the stimulus is impinging energy in the form of light, the receptors are rods and cones, and sensations are the electro-chemical outputs from those receptors. OK?
    Actually sensations tend to get talked about in terms of the light; I see what you're saying and it makes sense, but I think the more standard nomenclature is that sensations cause the firing of receptors which initiates a sequences of neural activity.

    So, if we agree that "sensations" are the most primitive biological entities available, how can all "sensation-based" theories be rejected?
    The question is not 'does light cause cells in the retina to fire?' but 'what is the basis of perception?'. One question is about anatomy, the other is about psychology. No one denies the retina is involved in vision: but visual perception, for Gibson, isn't about anatomy. It's the activity of a perceptual system (eyes+head+body, all mobile) who's purpose is to detect information to support action. So you can deny that perception is grounded in sensations without rejecting the fact that there is light impinging on retinal cells.

    Wikipedia doesn't understand psychology very well, unfortunately; it's all written by people who took Intro and nothing else.

    I started with a comment that was rejected as being too long
    That 'URL is too long' error doesn't stop the post from going through :)

    ReplyDelete
  13. So, I submit that if it is unquestionable that "information" in the technical sense is inapplicable to whatever in Gibson's scheme is the word's referent, then it would be preferable to choose a different, more descriptive word or phrase. Otherwise, its use seems yet another "rubbish metaphor".
    It's not a metaphor; Gibson is describing the information content of energy such as ambient light that is available to a mobile eye mounted on a mobile head, body, etc. That information is structure in the energy (defined over space and time) which specifies organism-relevant properties of the world, eg affordances. So there's nothing metaphorical about it, but it's not generic the way Shannon/information processing people use the term and the information content isn't measured in entropy.

    This issue of the 'technical sense' reminds me of Gibson vs. physics. The use of information that you have in mind is not privileged above others; it's perfectly sensible but within bounds, and it's of no use for a theory of perception. (Read all the comments, Ken's nagging was quite useful for me getting what I wanted to say out and my final comment sums up my point quite well, I think :)

    But just in case "information" in the formal sense might be applicable,...
    You're right, I was wrong when I said resolving ambiguity. It's been a while since I've done this sort of information. And this approach does get used: people studying perception from a sensation based view do exactly these kinds of analyses (or at least they used to, in the 60s and 70s when people still acknowledged the assumptions of their approach; but that's a rant for another day :) The good ol' days of psychophysics was full of this type of analysis, ROC curves, etc etc. So yes, it's a legitimate way to go at the problem; it's just not appropriate to Gibson because he doesn't 'view perception as a system the input of which is a continuous, time-varying, and unpredictable flow of sensory output (sensations)'. I need to think about this some more, but my first pass as to why is that the flow isn't unpredictable; it contains all this structure which makes the formal information analysis the wrong approach.

    ReplyDelete
  14. OK, I'm beginning to get tuned in. For Gibson, "the system" is the total environment, and the body is just a subsystem. And as in all system engineering, all one needs to address in the beginning are interfaces and functions: what goes in and comes out of each subsystem and how inputs and outputs are related. From that perspective, light (in one of its avatars) goes into the body via the eye, action comes out.

    So, I need to go back and reread Sabrina's summaries and the Gibson v. physics post with that in mind. "I shall return!"

    ReplyDelete
  15. We'll get you 'resonating' to Gibson yet (a little ecological humour there :)

    ReplyDelete
  16. As I get more in tune with Gibson's project, it seems to me that part of the problem is that there are some inconsistencies in the presentation (as opposed to the actual content). (In the Gibson vs physics comments, Ken made the same observation.) So, I'm going to tear into the presentation in this chapter pretty aggressively. To be clear, my target is only the content. The problems I'll identify may be due to any combination of Gibson's presentation in the book (to which I have no access), Sabrina's interpretation of that, some of Andrew's commentary above, presumed misconceptions held by some (all?) of Gibson's target audience, or my own misconceptions. But my objective is solely to learn, not to criticize anyone. So, hopefully no feelings will be hurt.

    The summaries of Chapters 1-3 are fine. I think I see what Gibson is trying to do and have even toyed - at a superficial level, of course - with some of the general ideas myself. And although I'm less clear of where he's going with all the "light" business, I have no complaints with the review of Chapter 4 up to the paragraph titled "Stimulation and stimulus information".

    The paragraph gets off to promising start with "If an eye ... intercepts some ambient light, photons will stimulate photoreceptors in the retina", with which I assume we all concur. (Although I don't get the restriction to ambient light. Is it significant? Or even correct?) And although I think it depends on one's precise definition of "perception", I'll accept the idea that perception requires "structure in the light". (I'd be more comfortable with something like "texture", or better still a short descriptive phrase - eg, "variation in one or more parameters across the FOV") - that removes all ambiguity.)

    In order to keep the rest of my critique in one comment, I'll break here.

    ReplyDelete
  17. Unfortunately, things now seem to fall apart. The next sentence is misleading:

    "This means that stimulation doesn’t lead to perception."

    I understand the distinction between stimulation by light that has no structure - hence, no information content - and light which does have structure and thus information content. But the logically sound way to express this is something like:

    Stimulation by light per se doesn’t lead to perception; the light must have structure in order to have information content.

    And this is not a quibble. The retina is the visual interface to the "body" subsystem. As such, that interface (as well as all others, of course) has to be described thoroughly and accurately since in the top level view of a closed system, the subsystem interfaces are all you have to work with.

    If one accepts all that, the rest of the paragraph becomes a jumble. The description is ambiguous as to how the eye, head, and body relate to one another and to the "perceptual system". I'm assuming that the body subsystem includes the head as a component which in turn includes the eye as a subcomponent. But in any event, with my current understanding of what Gibson is trying to do, it's not clear that detailed decomposition of the body is either necessary or desirable. (For example. in accounting for relative motion, is it necessary to distinguish between body and eye movements? If so, can that be captured in the interface specification without separating them into distinct subsystems? I would hope so.)

    Finally, if one continues to agree with my description, the rest of the paragraph should read something like:

    The input at the visual interface to the body subsystem - ie, at the photoreceptors - is "stimulus information", by which we mean light that has structure owing to the interplay between light in the environment and the other contents of the FOV.

    Note that I have eliminated any reference to "sensation". I infer that Gibson emphasized the contrast between his view and other contemporary views based on some concept(s) meant to be captured by that word. But as I have noted in earlier comments, if the word is meant to capture the concept of "sense data", that idea has been mostly abandoned, so that Gibson denigrating it is a matter of only historical interest. And if its meant to capture some other concept, in the absence of a description of that concept the word merely confuses.

    I'll have more to say on the general theme of current irrelevance of attacking outdated views when I go after the "image" stuff. (After I get some feedback on this much.)

    ReplyDelete
  18. Neither of us have thin skins, Charles; swing hard, we'll cope :)

    (Although I don't get the restriction to ambient light. Is it significant? Or even correct?)
    It is significant; only ambient light can contain information for perception, because ambient light is that which has reflected off something. Gibson's interested in the basis for visual perception, so radiant light isn't relevant to the case for 'stimulus information'.

    I'll accept the idea that perception requires "structure in the light". (I'd be more comfortable with something like "texture", or better still a short descriptive phrase - eg, "variation in one or more parameters across the FOV") - that removes all ambiguity.)
    Actually 'structure in light' is the less ambiguous term, because that's exactly what it is. If you're familiar with topology (things like div and curl) then that's what's going on here; literally structures in the flow. Koenderink and van Doorn are the two names that spring to mind for that more formal analysis.

    Some comments on your comments:
    1. For Gibson, perception is not the end result (somewhere in the system) of a process initiated by the detection of light energy at the retina, so it's quite correct for Sabrina to say that stimulation by light doesn't lead to perception. Perception is an act of the broader perceptual system, in which the system resonates to the presence of stimulus information. Yes, it is trivially the case that light and retinal cells are involved, but anatomy is not the right way to split the system out. So when you say 'The input at the visual interface to the body subsystem' you are not correctly describing Gibson's suggestions; visual perception is instead information interacting with a visual perceptual system (comprising the eye, the head, the body).

    2.Gibson isn't decomposing the system; he's laying out the idea that these components (eye, head, etc) are all identifiable pieces but that, as far as visual perception is concerned, they are functionally a single system. Hence Sabrina correctly says 'In people, our two mobile eyes, set in a turn-able head, on top of a movable body comprise a perceptual system (the visual system). This system is activated in the presence of stimulus information. '. This is a clearer way of saying what I was just saying.

    Your description is indeed one way to go about describing things. But it's not what Gibson was trying to convey, so while I'm happy for you to think about things this way, don't make the mistake of translating Gibson into that description in order to interact with him. For example, as I have said, Gibson explicitly does not mean 'sense-data' when he says 'sensations', and while the field has moved on a little in the terminology it uses since 1979, this actually only disguises the fact that Gibson's target is still alive and well, just not talked about clearly. Make sure you don't lose anything key in the translation; it's been a long term problem, with people interacting what they thought Gibson said once they had expressed his claims from their frame of reference.

    ReplyDelete
  19. Andrew -

    As a first step in trying to get a grip on this material, I need to construct a descriptive model of the scenario that Gibson is addressing. That descriptive model is solely for the purpose of getting a general idea about the relationships between the pieces involved in perception. It does not need to map directly onto whatever analytical model is used. Eg, although I understand that in the analytical model the light is treated as structured and continuous, and that the structure provides information, I don't need to address that level of detail in my descriptive model.

    (Abstracting is actually one of my strengths, so moving from the descriptive model to the analytical should be easy for me - although if the analytical model is too mathematical, I might rebel. Despite being pretty math-literate, I burned out on detailed analysis long ago.)

    So before tackling the analytical model, I'd like to get my descriptive one right. Currently, it's something like this:

    There's an environment containing entities that reflect light. At any point in the environment, the light received from a specified FOV is a composite of ambient light reflected by entities in that FOV. The visual receptors of a perceiver with the specified FOV will be stimulated by that composite light.

    The environment is assumed to be dynamic. Entities in the environment can move, and a perceiver is assumed to be fully mobile (eyes, head, body) and can change it's FOV (and consequently, the stimulation of the visual receptors). So, the content of the visual stimulus can be time-varying

    I take the input to the perceiver viewed as a perceptual system to be the content of the light received from the FOV. But I'm still missing a description of the results of the perceptual process. (I don't know how to get from "the system resonates to the presence of stimulus information" to the description I need.) In the analytical model, what is measured to determine how well the system is perceiving, and to what - if anything - in my descriptive model might that correspond?

    ReplyDelete
  20. In the analytical model, what is measured to determine how well the system is perceiving
    Action. If you want to measure perception, you generally want to measure the stability of the action that information is supporting. So, for example, in rhythmic movement coordination, the information is hypothesised to be the relative direction of motion of the oscillators, conditioned by the relative speeds as a noise term. To assess this, I need to manipulate the availability of information and measure the consequences on behaviour (often we do judgement studies first, eg Wilson & Bingham 2008, but they are always only a preliminary to the real deal, eg Wilson et al 2005).

    and to what - if anything - in my descriptive model might that correspond?
    So it doesn't correspond to anything in your model, because your model is not yet complete :) You need action!

    ReplyDelete
  21. Given our "failure to communicate" so far, this may come as a surprise, but I couldn't have hoped for a better answer! I won't (probably can't yet) elaborate, but I place tremendous importance on language (a Rortian influence) and am moving toward interpreting language in terms of action (a Davidsonian influence). So, several things seem to be converging for me.

    Is it fair to say that the "output" of the perceptual system is action toward some objective? Ie, "behavior"? And if so, won't some object that this makes Gibson's project "behaviorism", which I gather is a term that in modern psychological circles is considered pejorative?

    I don't know if you read all comments at BofC, but in case you don't here's part of one that was directed at you in absentia:

    "this seems an appropriate opportunity to say how grateful I am for the opportunity to interact with people like you (and Andrew over at his blog, Peter at CE, et al) - ie, to triangulate. It probably isn't the primary intent when specialists set up blogs, but for unaffiliated people like me, they function as classroom, dorm bull session, one-on-one tutor, et al. Many thanks."

    ReplyDelete
  22. Glad you're finding the blog useful! I keep tabs on most of what goes on at BoC although I tend to zone out on some of the topics which are less relevant to me :)

    Is it fair to say that the "output" of the perceptual system is action toward some objective? Ie, "behavior"?
    You could say this, although I always worry this sort of talk draws attention away from the dynamical system aspects, namely that said output directly creates the information that drives behaviour, etc etc. It always sounds hand-wavy to just draw 'perception-action loops' when I give lectures, but it's just true! :) But yes, so long as you keep one eye on the fact that the perception-action-perception cycle is the deal, then your description is perfectly ok. There's always devil in the detail, but that's ok too.

    And if so, won't some object that this makes Gibson's project "behaviorism", which I gather is a term that in modern psychological circles is considered pejorative?
    Gibson was a behaviourist, although more Holt than Skinner (see this post on Harry Heft's book). Gibson was also quite happy to be one, in line with Holt and Holt's other student Tolman. I also have no problem with being a behaviourist; I think many of the objections have become obscured over time and don't hold as much water as people like to think. So yes, it is about behaviour, but purposive behaviour, behaviour that has goals, etc.

    ReplyDelete
  23. Not to worry. In my younger days, I frequently encountered control systems employing feedback, which sounds like what Gibson's view amounts to. So. when I say "objective", I don't mean that in the sense of reaching a destination but in the sense of staying on the road instead of ending up in a ditch.

    Are there more than the five chapters Sabrina has already covered, and if so is she planning more installments?

    ReplyDelete
  24. There are, and she is. Stay tuned!

    ReplyDelete