Saturday, October 9, 2010

Gibson and Mach Bands

In one version of the Mach bands, one has a dark field and a light field separated by a contrast gradient.  Yet, at the right edge of the bright field, there is an apparent brighter band, while at the left edge of the dark field there is an apparent darker band.

From Purves Lab

The standard account of Mach bands has to do with lateral inhibition in the retina.  But, Gibson writes things that seem to be highly dismissive of a) not just of this account, but of any such accounts and of b) attempts to understand them.
Not even the current theory that the inputs of the sensory channels are subject to "cognitive processing" will do. The inputs are described in terms of information theory, but the processes are described in terms of old-fashioned mental acts: recognition, interpretation, inference, concepts, ideas, and storage and retrieval of ideas. These are still the operations of the mind upon the deliverances of the senses, and there are too many perplexities entailed in this theory. It will not do, and the approach should be abandoned.
         ...
Let us remember once again that it is the perception of the environment that we wish to explain. If we were content to explain only the perception of forms or pictures on a surface, of nonsense figures to which meanings must be attached, of discrete, stimuli imposed on an observer willy-nilly, in short, the items most often presented to an observer in the laboratory, the traditional theories might prove to be adequate and would not have to be abandoned. But we should not be content with that limited aim. (Gibson, 1979, pp. 238-9).
http://www.purveslab.net/research/explanation/brightness/mach.html#f1

42 comments:

  1. If you want to really get ecological on this, psychologists can simply ignore the fact that, to a photometer, the contrast ramp is in the shape described in the middle graph, and focus on the fact that to an eye, the contrast ramp is in the shape seen at the bottom. Physics is not a privileged level of description, remember.

    This is clearly just a sensible feature to add to the retina part of a visual system; contrast is interesting and should be 'enhanced' (compared to the relatively poor contrast detection of the CCD in a light meter, for instance, which has to deal with the function in the middle).

    Gibson denied that perception began at the retina, but he never claimed the organism wasn't involved in useful ways. Clearly the visual system doesn't care what the 'physical' gradient is (in the same way it doesn't care about distance in terms of mm or in).

    ReplyDelete
  2. Let's go back to this: "Not even the current theory that the inputs of the sensory channels are subject to "cognitive processing" will do. The inputs are described in terms of information theory, but the processes are described in terms of old-fashioned mental acts: recognition, interpretation, inference, concepts, ideas, and storage and retrieval of ideas."

    Now, Gibson is not so specific here about what a "sensory channel" is, e.g. whether a single neuron is a sensory channel or whether the retina plus optic nerve is a sensory channel. But, he apparently wants to reject the idea that this does "cognitive processing", since he thinks this involves invoking mental operations, such as recognition, interpretation, inference, concepts, ideas, and storage and retrieval of ideas. But, the lateral inhibition account doesn't work this way. So, it looks like the "cognitive processing" account of the retina is ok after all.

    ReplyDelete
  3. Hardly. The wiring of the retina is hardly an example of any of the 'mental operations' he's rejecting. Gibson's talking here about rejecting enrichment theories of perception based in sensation, and it's the classical mechanisms of enrichment (eg a mental representation containing disambiguating content) that's the target here.

    There's nothing anyone would call cognitive processing here (well, unless you want to say that the retina's wiring embodies a transformation of the signal from the physical to the psychological function, which would a slightly odd and imprecise way to go about things).

    ReplyDelete
  4. Dude! I think that using lateral inhibition for edge enhancement is a kind of paradigm of what cognitivists will count as cognitive processing. To enhance a differences in luminance is to invoke a kind of assumption about the world, namely, that such differences indicate edges.

    ReplyDelete
  5. So is the polar planimeter a cognitive processing device too?

    ReplyDelete
  6. Probably not, because it does not do the right kinds of information processing. Remember, the Adams & Aizawa view (which is a weak form of cognitivism) is that cognitive processing is a specific sort of manipulation of non-derived content bearing representations. The polar planimeter doesn't to the right kind of manipulations. Lateral inhibition is ok; rolling along an edge is probably not.

    ReplyDelete
  7. But the lateral inhibition can be described in two ways: the 'cognitive' way, in which the retina takes one signal and transforms it into another signal, and the 'smart' way, in which the retina directly detects the signal in the second form. Does the retina ever take the initial signal on board, so to speak, and then transform it? I don't think it does, thus the smart description seems more appropriate.

    I think your hunch that it is cognitive is based in the assumption that the physical function is the baseline, and that it must be transformed into the psychological function. Remember, the physical function is simply one way to characterise that signal; Mach bands reveal that this is the incorrect way to characterise the signal when detecting it with a retina.

    ReplyDelete
  8. Well, I take it that lateral inhibition involves interactions among cells. To simplify, one cell provides one signal about light intensity/frequency at one physical point, another cell provides another signal about light intensity/frequency at another physical point, and the yet another cell takes those signals and produces another.

    But, yes, there are multiple levels of description. One level is the physiological, then the higher level is psychological. That's pretty much standard stuff, right? So, there's this physiological process that realizes the psychological process.

    ReplyDelete
  9. In fact, this is the kind of commentary that I think that Fodor and Pylyshyn would apply to the case:

    "The current Establishment theory (sometimes referred to as the "information processing" view) is that perception depends, in several respects presently to be discussed, upon inferences. Since inference is a process in which premises are displayed and consequences derived, and since that takes time, it is part and parcel of the information processing view that there is an intrinsic connection between perception and memory. And since, finally, the Establishment holds that the psychological mechanism of inference is the transformation of mental representations, it follows that perception is in relevant respects a computational process. "

    The signals from the neurons are mental representations and the way in which one neuron responds to others is a kind of inference. This is the kind of thing they are talking about when they talk about information processing.

    ReplyDelete
  10. But what if the relevant physiological unit isn't the cell, but the broader system of multiple cells? It's fairly uncontroversial that single cells aren't that interesting; maybe the psychological level should inform the search for the physiological mechanisms for a change.

    ReplyDelete
  11. Well, it isn't the single cell that matters in this case, but the neural network.

    As for psychology informing physiology, Fodor, for example, largely dumps on physiology. He's Dr. Autonomy of Psychology. So, he would probably let psychology take the lead.

    Personally, I think that psychological studies of visual illusions often provide understanding of vision that is beyond the current reach of neuroscience.

    All that, however, is not a matter of denying that vision begins with the retina and that there is computational processing or information processing in the retina. That's what Gibson at least appears to be denying.

    Maybe ask Gary what he thinks Gibson would say about the Mach bands. There is a lot of contemporary philosophy that draws inspiration from Gibson and appears to want to avoid this kind of retinal information processing.

    ReplyDelete
  12. Andrew (from another post):
    "2. ... The detection and transduction of light happens, but that's not the starting point for perception, simply part of the chain of events in detecting visual information. "

    I just don't get this at all. Can you explain this a bit more?

    ReplyDelete
  13. "For our present purposes, I will review several
    categories of illusion that have been more traditionally explored in history and in modern
    neuroscience. The purpose of this article is to
    illustrate how such illusions have helped to
    guide neuroscience research.To that end, I will
    attempt to trace the growth of intellectual
    threads that directly led to insights into — and
    placed constraints on — the underlying neural
    mechanics of vision."

    David Eagleman, "Visual Illusions and Neurobiology". Nature Reviews Neuroscience. (Available via Google Scholar).

    ReplyDelete
  14. I'm certainly suggesting that Mach bands aren't information processing (physical input -> transformation -> psychological output) but direct detection of the psychological output using a smart contrast detector.

    Gibson denies any need for computation or mental operations because he has a theory of information that makes them unnecessary; the problem they solve doesn't exist in Gibson's theory of perception. The organism doing the perceiving is clearly part of the overall act of perception, but it's not doing what you keep claiming it's doing (eg computation).

    "2. ... The detection and transduction of light happens, but that's not the starting point for perception, simply part of the chain of events in detecting visual information. "

    I just don't get this at all. Can you explain this a bit more?

    For Gibson, visual perception begins in the structuring of ambient light by the environment. This move enables him to have information that specifies something in the world.

    For everyone else, there is no structure in light, there is only structure in the activity of the retina (the retinal image, however you precisely want to characterise this). This proximal stimulus is deeply impoverished, and entirely ambiguous.

    So, Gibson denies that the retinal image can possibly be the basis for perception. All I was pointing out is that this doesn't mean the retina isn't part of the functioning of a visual perceptual system. It's job is clearly the transduction of light into neural signals, which is clearly part of what happens during the act of vision. Gibson (and I) just deny that the retina is a bottleneck in this act of the type assumed by traditional theorists.

    That was a little long winded, but I'm aiming for context :)

    ReplyDelete
  15. Sorry, I missed your last comment for some reason.

    "Gibson denies any need for computation or mental operations because he has a theory of information that makes them unnecessary"

    But, wait a second. I thought you said that you accept the standard account of Mach bands. When I originally asked the Gibsonian account of Mach bands you wrote,"I've no idea. Aren't they just a feature of the organisation of the retina that enhances contrast? The kind of thing you might build into a visual system interested in differences? That seems uncomplicated for everyone. "

    So, do you accept the standard explanation of Mach bands as computations involving lateral inhibition or not?

    ReplyDelete
  16. "For everyone else, there is no structure in light"

    What? Don't you think that mainstream vision scientists think that there are differences in, say, intensity of light at different points in an camera image projected on the retina? Why is that not structure in light?

    ReplyDelete
  17. "For Gibson, visual perception begins in the structuring of ambient light by the environment."

    Ok. So, I get this, but it is pretty misleading. One can say that visual perception does not *begin* with photon capture, but the crucial part is that perception will still be mediated by photon capture. Photon capture is an intermediate step in perception. It is these intermediate stages of photon capture, lateral inhibition, etc. that seems to explain the principal features of Mach bands.

    ReplyDelete
  18. So, do you accept the standard explanation of Mach bands as computations involving lateral inhibition or not?
    No. As I said, "Mach bands aren't information processing (physical input -> transformation -> psychological output) but direct detection of the psychological output using a smart contrast detector.". Taking a 'smart' approach means conceiving of the job differently.

    Don't you think that mainstream vision scientists think that there are differences in, say, intensity of light at different points in an camera image projected on the retina? Why is that not structure in light?
    That's structure in the retinal image.

    Also, ok, intensity differences might be structure of a kind; but it's not the kind that can support information (as everyone knows, such a proximal stimulus is ambiguous).

    And it's not misleading - it's being precise. Visual perception has to somehow end up as capable of supporting all the behaviour we know it supports. Gibson's definition of terms is entirely about scientifically laying out what that system must be like. Photon capture, lateral inhibition, aren't the basis for perception because they lead to a proximal stimulus that cannot support visual perception.

    ReplyDelete
  19. Ok. Mach bands aren't information processing. Right. Mach bands are a phenomenon standardly *explained* by lateral inhibition, which is the kind of thing mainstreamers also describe as computation or information (of a non-Gibsonian type) processing. Mach bands are due to lateral inhibition in LGN cells in the retina. You reject this?

    So, what is your explanation of Mach bands then? That they are the direct detection of the psychological output using a smart contrast detector? That doesn't seem to me to be much of an explanation, but instead a redescription of the phenomenon.

    What's the explanation of the Herman grid illusion? That it the direct detection of the psychological output using a smart smudge at intersection detector?

    What' the explanation of the Ponzo illusion? That it is the direct detection of the psychological output using a smart line detector?

    Light structure: But, it is differences in the intensities of light that give rise to intensity differences in a retinal image. Or think of sine wave gratings. These are described in terms of orientation, contrast, and frequency. These properties of objects structure light in the environment. This structure propagates with the light to the retina giving a retinal image that is at least in part a function of the structure of the light. (There is also the contribution of certain aberrations in the physiological optics of the eye.) And, you originally said "For everyone else, there is no structure in light, there is only structure in the activity of the retina"

    It's misleading to say that perception does not begin with photon capture, when one is really after the idea of photon capture playing an explanatory role in, for example, Mach bands. It's misdirecting attention from the issue at hand.

    "Photon capture, lateral inhibition, aren't the basis for perception because they lead to a proximal stimulus that cannot support visual perception." Ok. Let's be precise here. What you mean is that Photon capture, lateral inhibition, aren't the basis for perception *as Gibson defines it* because they lead to a proximal stimulus that cannot support visual perception *as Gibson defines it*, but that's different than saying that photon capture, lateral inhibition, aren't the basis for perception as defined by mainstreamers because they lead to a proximal stimulus that cannot support visual perception as defined by mainstreamers.

    But, what is the Mach bands phenomenon an instance of perception as Gibson defines it? You've said this before, but I dont' see it.

    ReplyDelete
  20. Light structure: Heck, why go as far as sine wave gratings. Look at the top panel of the figure above. That physical object gives rise to light having the structure shown in the second panel. The light in the second panel gives rise to an experience of the sort shown in the third panel.

    ReplyDelete
  21. So, you earlier commented, "It's not like we don't think the retina is involved at some point in the proceedings; it's just not the first thing to structure light. "

    So, granting that the retina is not the first thing to structure light (which seems is odd, since I don't think the retina structures light), what is it retina doing in there? What is that lateral inhibition doing if not processing non-Gibsonian information to generate, in conjunction with other visual system apparatus, Mach bands>

    ReplyDelete
  22. So, what is your explanation of Mach bands then? That they are the direct detection of the psychological output using a smart contrast detector? That doesn't seem to me to be much of an explanation, but instead a redescription of the phenomenon.
    I think Mach Bands are a side effect of the wiring of the smart device which perceives 'enhanced contrast'. In fairness, I have no specific evidence for this; it's just a first swing at a 'smart' analysis.

    What's the explanation of the Herman grid illusion? That it the direct detection of the psychological output using a smart smudge at intersection detector?
    Another side effect of the smart detection system (I would suggest).

    I don't really know the Ponzo illusions that well so I don't have a suggestion for that one of the top of my head.

    But the question to keep in mind is whether you think the polar planimeter is processing information to arrive at area, or whether it's directly detecting area as the 'simple' variable. That example is the example I have in mind wrt the Mach Bands, etc.

    So for instance:
    Look at the top panel of the figure above. That physical object gives rise to light having the structure shown in the second panel. The light in the second panel gives rise to an experience of the sort shown in the third panel.
    The second panel only arises if there is something like a photometer involved somewhere. Only to that device does the middle panel show up. To a retina wired the way Mach Bands reveal they are, perhaps the physical object in the top panel only gives rise to the experience in the third panel. This is what Runeson and Gibson mean when they talk about how physics is not necessarily the correct level of description. If there's nothing like a photometer involved then there's no need to convert from it.

    ReplyDelete
  23. "But the question to keep in mind is whether you think the polar planimeter is processing information to arrive at area, or whether it's directly detecting area as the 'simple' variable. That example is the example I have in mind wrt the Mach Bands, etc."

    Ok. My turn to give you an answer. I think the polar planimeter is processing information. It is converting the change of positions in the tracing point into rotations of a dial that *represents* the area as computed by the distance the point has traveled along with changes in the positions of the arms.

    So, this seems to me at least to involve representations (what is recorded on the dial represents area). Moreover, there seems to there have to be representations over time to somehow preserve a record of what has been scanned so far. (This goes back to my example a while back of perceiving a flat head screw driver by touching the end, then the middle, then the handle. There need to be records of the past.)

    But, the PP does information processing, only not (Adams and Aizawa conjecture) the kind of information processing that humans do in their brains. So, it is non-cognitive information processing. (Recall that the A&A mark of the cognitive includes a condition on the types of processing performed.)

    ReplyDelete
  24. "The second panel only arises if there is something like a photometer involved somewhere. Only to that device does the middle panel show up. To a retina wired the way Mach Bands reveal they are, perhaps the physical object in the top panel only gives rise to the experience in the third panel. This is what Runeson and Gibson mean when they talk about how physics is not necessarily the correct level of description. If there's nothing like a photometer involved then there's no need to convert from it."

    Ok. I'm not sure how much of this you mean as you say it. I take it that it is straight physics that the ink pattern (or whatever) in the top panel objectively produces a structure in the light of the sort shown in the second panel. That looks to be just physics and it is that way whether there's a photometer around to measure it or not. That's just what philosophers call "realism" about the external physical world.

    Now, what you can say is that the human eye does not respond to *that* structure in the light. But even that is kind of a strange thing to me. What is this *other* structure of light if not that shown in panel #2? I mean you can say it's "ecological optical structure" rather than physical optical structure, but then what's the difference between eos and pos?

    ReplyDelete
  25. I'm perfectly familiar with realism, which you should know given Gibson's ontology is a direct realism.

    But the problem is with the object 'objectively produces a structure in the light of the sort shown in the second panel.'. You're missing 'when measured with a photometer' and thus your description is incomplete. When measured with a retina, it is objectively the case that it produces a structure in the light of the sort shown in the third panel'. Given we have retinas and not photometers, the pattern of (well, some limited) interest to psychology is the latter.

    And if you think the polar planimeter is an information processor then it's not clear to me you will ever understand what the dynamical and ecological types are up to. There is a difference between a processing information and a calibrated smart measuring device with a dynamic of it's own. One is rote; the other is smart. One can be general; the other is task-specific. I presume you think van Gelder's steam governor example is information processing too.

    ReplyDelete
  26. To follow up a little: there is computational description of the polar planimeter (I'll see if I can dig up a reference). But as I recall,

    a) it wasn't worked out until about 200 years after the device had been made
    b) it's ridiculously complicated, and
    c) there's nothing in the polar planimeter that implements most of the specific steps in the calculation.

    The whole device 'implements' the whole computation, in that they do the same thing, but it's not at all clear that 'implement' is the right relationship here. As per van Gelder's analysis of the steam governor, the much better level is in terms of dynamics and calibration (and this is the level any engineer would happily use for either device).

    ReplyDelete
  27. Well, Gibson uses "information" idiosyncratically, "perception" idiosyncratically, so why not "realism". But, ok.

    Well, I don't think my understanding of the polar planimeter is that it is an information processor and an analog computer. I've posted a query about this at another blog I write for from time to time: http://philosophyofbrains.com/2010/10/19/is-the-polar-planimeter-an-analog-computer.aspx. Let's see how that turns out. The blog is mostly philosophers like me.

    Yes, I think the steam governor is an analog computer.

    Well, I do think that there are differences between analog and digital computers, so it's not as though I don't see any differences. I just don't make out the differences in terms of an informal notion of information processing and computation. But, it's true that I don't get what you are Runeson are saying about the polar planimeter being a "smart" mechanism. That's not a very helpful description to me.

    ReplyDelete
  28. Oops, I should have written: "Well, I don't think my understanding of the polar planimeter is that it is an information processor and an analog computer is peculiar to me." Better might have been, "I don't think I'm the only one to hold that the polar planimeter is an information processor and an analog computer."

    But, you know, if you can't explain it to me, you're going to have a much tougher time winning over other cognitivists. Most will not invest the time to master all the jargon. I wouldn't be dedicating this much time to Gibson just to satisfy my intellectual curiosity. I feel somewhat compelled to get on top of it, since it seems to be driving certain segments of the extended cognition movement.

    ReplyDelete
  29. I take it that the item in panel one produces the structure described in panel two, regardless of whether there is a photometer to measure it or not? Physics explains this.

    I also take it that the item in panel one produces some structure *other than that described in panel two*, regardless of whether there is a human visual system there or not. So, what is the EP account of how the thing in panel one generates this other structure?

    ReplyDelete
  30. To be honest I don't spend a lot of my time trying to convert anyone unless they're interested enough to keep going. My hunch is that you win a scientific revolution by getting on with your empirical work and trying to out-do the opposition in grad students, rather than conversion.

    But if you can't even conceive of how the polar planimeter and the steam governor might be best described as dynamical systems rather than computational systems, then you won't ever 'get' the ecological basis to any EC arguments and there's not a lot I can do to help (although I'm happy to keep trying). Disagreeing with that assessment is one thing, and obviously fine - but you keep saying you just can't wrap your head around the idea.

    I take it that the item in panel one produces the structure described in panel two, regardless of whether there is a photometer to measure it or not? Physics explains this.
    I'm not sure this is true. Prior to the invention of the photometer, the only device that measured the object in panel 1 was a retina and it detects the pattern in panel 3. Photometers are clearly not the only way to measure that object.

    ReplyDelete
  31. "But if you can't even conceive of how the polar planimeter and the steam governor might be best described as dynamical systems rather than computational systems,"

    Well, maybe the polar planimeter is best described as a dynamical system, rather than a computational system. But, that doesn't mean it isn't a computational system. Nor does it mean that it's not an information processing system. I haven't responded to your commentary that the pp is best described as a DS, rather than a computational system, because I don't see any inconsistency.

    I take it that the sun produces structured light regardless of whether there are photometers or eyes to detect it, right?

    ReplyDelete
  32. But then you can compare the descriptions. In the case of the polar planimeter, there aren't any bits doing the individual computations required for the full computational description; so while the computational description might end up with the same result, it's not describing the steps by which the device itself achieves that result. So while it's an adequate description of the process, it's not an adequate description of the device, and if you want the latter than you have good reason to reject the computational description as inadequate.

    The sun is a light source so it's producing radiant, not ambient light. Ambient light contains ecologically/psychologically relevant structure by virtue of it's interactions with surfaces; radiant light does not.

    ReplyDelete
  33. Were rocks structuring light before there were organisms?

    ReplyDelete
  34. Yes. But who knows what it meant until something evolved a detector.

    ReplyDelete
  35. Oh oh oh. I do! I do! It meant nothing! =)

    But, the light and its structure were still there.

    ReplyDelete
  36. So, are you now willing to say that rocks the sun shines on structure light these days, even when no photometer is detecting this structure.

    ReplyDelete
  37. Sure - You need a niche to evolve into. There was all kinds of structure available. One kind can be detected by a photometer; one kind gets picked up by a retina. And so on.

    ReplyDelete
  38. Ok. So, the structure that can be detected by a photometer is shown in panel two of the figure, but the structure that can be detected by the retina is ___________?

    ReplyDelete
  39. It's a fair cop; I was imprecise.

    There is structure in light as the result of it's interactions with surfaces.

    One example of that structure looks like Panel 2 when detected by a photometer, and like Panel 3 when detected by a retina. (This is analogous to measuring a length that has a real size using either an inch or cm ruler.)

    Does that clarify? You wanted a role for the organism in perception other than detection; building a specific smart device could be it.

    ReplyDelete
  40. Ok. So, according to Gibson, the thing in panel one generates a structure in light like that shown in panel 3. But, how does the thing in panel 1 do this? For "lamestream" vision scientists, physics spells out how panel one generates panel 2 in the light, right? What is the story for Gibsonians?

    Or, maybe a better way to ask the question is this. It looks like panel two specifies panel one, but panel three specifies something other than panel one, something with "extra" dark and light bands in it.

    ReplyDelete
  41. This is where I was being imprecise. The object structures light; what that structure looks like depends on the measuring device.

    The 'extra' idea only makes sense if Panel 2 is somehow the 'true' visual experience of the object.

    ReplyDelete
  42. Wait a second.

    The mainstream analysis is:
    Panel one describes the ink and paper object.
    Panel two describes the structure of the light that produces the visual experience.
    Panel three describes the visual experience.

    The Gibsonian analysis is:
    Panel one describes the ink and paper object.
    Panel two misdescribes the structure of the light that produces the visual experience; instead, something like panel three describes the structure of the light that produces the experience described in panel three. Right?
    Panel three describes the visual experience.

    Nobody, I don't believe, thinks panel two describes the visual experience.

    ReplyDelete