Tuesday 2 November 2010

Gibson vs Physics: Gibson Wins, at the Ecological Scale

One of the interesting questions that popped out of our discussions with Ken Aizawa about Runeson and the Ames Room is this: did Gibson and his followers banish physics and geometry from his psychology? And if so, is Runeson breaking this prohibition by talking geometrically about the Ames Room? And how can you banish physics anyway - what the hell?

I think I've convinced Ken that there is no such prohibition in the comments at the posts linked to above (I think). But it's a topic of fairly central importance to the ecological approach, so I wanted to summarise some of these ideas and examples here.

(Psycho)physics
Physics describes objects in the abstract, as approximations in units that aren't relative to an observer. Remember, in the joke, it's the physicist who assumes the cow is a sphere. But we tend to treat the physical description as 'the way things actually are', the objective yardstick against which all science should be evaluated. Indeed, the science of psychophysics is explicitly premised on the idea that the job of the sensory systems is to convert the objects of the physical world into the objects of psychological experience, because the former are 'what there is to be known'. Vision science is therefore fascinated by illusions because they are supposed to reveal how our psychological experience often fails to correspond with the physical reality, due to the assumptions we have to impose in order to convert from the one to the other.

But recall the ecological response to illusions; first, they are generally readily revealed by exploration that breaks the limitation creating the problem. Second, and most relevantly here, they are only 'failures' if you think the job of the sensory systems is to bring your psychological experience into correspondence with an external world as described by the science of physics. It's precisely this that Gibson and his followers reject.

Gibson (1979)
Gibson does indeed spend quite a bit of time railing against physics; specifically, against the argument that the objects of perception are best described using the language of physics. Gibson spends the first three chapters of the 1979 book defining what things like 'surface', and 'ambient light' mean to an organism, rather than to physics. Sabrina has covered this in her Reading Group posts on Chapter 1Chapter 2 & Chapter 3. These chapters are unique to Gibson (1979); no other perceptual psychologist feels the need to (re)define what the world is made of, given that this is what physics is supposedly for. I find them fascinating, precisely because no-one else thought it needed doing, while Gibson clearly felt he had to, in order to get his theory up and running. This is the first hint that Gibson is truly proposing something different. 

Turvey, Shaw, Reed & Mace (1981)
But Gibson is not denying that information is rooted in physics; he's simply pointing out that the correct level of analysis for the information available to a perceiving organism is ecological. What an organism needs to know is not how far away something is, but whether it can reach that thing - in other words, affordances. You can't get to affordances via the objects of physics, because affordances are personal and contain meaning while physics is neither of these. You therefore have to get to affordances by detecting information about affordances, not units of physics.

The Establishment concern with this was summed up in the Fodor & Pylyshyn (1981) paper which I've discussed before as 'the dick move'. The dick move is to simply say this information could be trivial; if you detect the property of 'being a shoe' you will of course perceive a shoe. The reply by Turvey et al (1981) is the 'ecological laws' paper that lays out how information can arise via lawful relations, and how this information will therefore specify affordance properties within the scope of the law. (Ken is blogging about this paper here, and given how that's going I think I'll post about that paper in some detail again.) 'Being a shoe' isn't a candidate property because it doesn't lawfully give rise to a specific pattern in a perceptual array which an organism might detect.

So where does this leave physics?
Physics presumes a privileged, third-person perspective and describes things from this position. The science of perception is interested in what it is possible for an individual to know about, and the mechanism supporting that knowledge. Individuals do not have access to such a third person perspective, and thus the description of the world from that point of view is not appropriate for describing perception. Individuals exist in niches, with a specific set of capacities and opportunities for behaviour - these are what perception must be about, because these are what we must remain in contact with in order to behave. Gibson proposed that a) perception is of affordances, and occurs b) by virtue of information. Ecological optics is an attempt to scientifically engage with the topic of information, and is therefore itself an abstraction, like physics. However, it is an abstraction of the ecological level, and thus remains a more appropriate abstraction than physics.

Physics (and geometry) still have a key role to play, however. Ecological laws must obey the laws of physics. Information (spatial-temporal patterns in energy arrays) must arise from interactions between energy and surfaces and these interactions will be describable according to the laws of physics. The systems for detecting these patterns will evolve in line with the physical constraints of detecting that energy. Geometry is still an excellent tool for describing the resulting information, because geometry is concerned with form, shape and location and provides tools for discussing these qualities. There is more than one geometry, of course; the Euclidean geometry we're all familiar with from high school is just one of many, each making different assumptions about how form is preserved (or not) over various transformations. Euclidean geometry, for instance, is not a particularly good description of reach space; a better description is affine geometry (e.g. Todd et al, 2001).

Physics and ecological optics are abstractions whose purpose is to aid scientific analysis. Mathematics (especially geometry) is a system for formally interacting with these abstractions during that analysis. Like all tools, you either have the right one for the job, or you don't. The science of perception has to account for how individual organisms, who are situated in specific niches and who have specific evolutionary trajectories related to that niche, can come to have knowledge about the possibilities for behaviour within that niche. Physics is not equipped to answer these questions; ecological optics is, and some flavour of geometry will come in handy because geometry is just a way of talking about precisely the kinds of things described by ecological optics. It will remain, as ever, an empirical question as to which geometry is best suited, and when.

References
Fodor, J. A., & Pylyshyn, Z. W. (1981). How direct is visual perception?: Some reflections on Gibson’s “Ecological Approach”. Cognition, 9,139-196. DOI

Todd, JT, Oomes, AHJ, Koenderink, JJ, & Kappers, AML (2001). On the affine structure of perceptual space. Psychological Science, 12(3), 191-196. DOI

Turvey, M. T., Shaw, R. E., Reed, E. S., & Mace W. M. (1981). Ecological laws of perceiving and acting: In reply to Fodor and Pylyshyn (1981). Cognition, 9, 237-304. DOI

32 comments:

  1. "'Being a shoe' isn't a candidate property because it doesn't lawfully give rise to a specific pattern in a perceptual array which an organism might detect."

    Ok. Can you give me an example of something that does give rise to a specific pattern in a perceptual array which an organism might detect?

    ReplyDelete
  2. A really simple example* would be the focus of expansion (FoE) in the optic array. As you locomote, the optic array flows around you from a point (the focus of expansion) which is precisely the point in the optic array which corresponds to your heading. So 'heading' is a candidate property because it gives rise to this pattern which is detectable by anyone with a visual system.

    *(It's slightly more complicated than this, of course. The eyes move too and add a component of motion to the optic array flow field. There's currently two ecological models of steering in the literature: Fajen and Warren suggest you recover the FoE and use that, while Wilkie & Wann suggest you use the 'retinal flow' (optic flow + eye movements). But the FoE is still a nice clear example of the kind of thing I'm talking about.)

    ReplyDelete
  3. Isn't the (FoE) in the optic array a specific pattern in a perceptual array, rather than a property that lawfully gives rise to a specific pattern in a perceptual array?

    ReplyDelete
  4. "You therefore have to get to affordances by detecting information about affordances, not units of physics."

    "Ecological laws must obey the laws of physics. Information (spatial-temporal patterns in energy arrays) must arise from interactions between energy and surfaces and these interactions will be describable according to the laws of physics. The systems for detecting these patterns will evolve in line with the physical constraints of detecting that energy."

    Detecting energy sounds like detecting a physical quantity or a "unit of physics".

    ReplyDelete
  5. Yes, sorry; heading is the property; the FoE is the pattern.

    Detecting energy sounds like detecting a physical quantity or a "unit of physics".
    No one's denying that light energy is involved in vision. But perception doesn't begin with photons.

    ReplyDelete
  6. >Detecting energy sounds like detecting a physical quantity or a "unit of physics".


    Gibson famously said that "what a thing is is not separate from what it means".

    My favorite example is the perception of a rock. Say I perceive a rock in terms of how it affords me a place to sit. Now, detecting the property of "sit-uponable" is not a matter of detecting changes at the molecular level. I do not need to know anything about physics in order to learn that the rock affords a place to sit down. The property of "sit-uponable" is not something intrinsic to the rock, yet it's ability to support me is determined by its actual physical properties. The physicist would say that what is perceived is the object spatio-temporal structure of the rock. This is true, but only in a trivial sense. What is *perceived* is the fact that the rock supports me. Sure, I am detecting "physical units" insofar as the rock is a physical thing and the light bouncing off it is a physical thing. But as Andrew nicely showed, this would be that wrong level of description for psychology. The level of description relevant to psychology would be in terms of the learning and detection of the affordance of the rock, not in a description of how the rock is held together at the atomic level. Like Bergson, Gibson thought that perception operates by means of a subtraction or selection rather than construction. There is too much information out there. In order to perceive the affordance, I need to subtract from the given stimulus the abstractions relevant to my needs. I subtract away from the molecular flux and instead perceive the ecological level of reality, which isn't necessarily tied to the present-moment, but rather, can persist across time. Ecological information has both an adjacent and successive order.

    ReplyDelete
  7. Andrew, it seems to me that the passages I've quoted are contradictory. The first one says that one detects information, not units of physics. The second one says that one detects energy, which looks like a unit of physics. What's up?

    ReplyDelete
  8. Isn't heading the same thing as direction, so that FoE would not specify heading? Instead maybe something like forward motion?

    ReplyDelete
  9. Forward motion is specified by the global optic flow. The FoE specifies heading.

    Maybe the latter 'detect' could be replace with 'interacting with'? Would that help? I'd hate for this to get derailed by a sentence I added at the last minute.

    ReplyDelete
  10. Let me put it this way. We detect information (spatial-temporal patterns in energy arrays) and so we perceive the world (by virtue of the specification relation between the information and the property of the world). Does that help?

    The perceptual systems clearly interact with energy and optimising the capacity for that interaction is obviously of value. I added that link because I found that NYT article right after I posted this and it's interesting :)

    ReplyDelete
  11. I find your account thoroughly baffling.

    So "Gibson is not denying that information is rooted in physics'

    but, then,

    "You can't get to affordances via the objects of physics, because affordances are personal and contain meaning while physics is neither of these. You therefore have to get to affordances by detecting information about affordances, not units of physics.You therefore have to get to affordances by detecting information about affordances, not units of physics."

    You can't get affordances from physics, right? Information is about affordances, not physics, right? So, how *does* physics get into the picture then?

    ReplyDelete
  12. I sort of expected that Gibson would have physics specific either an affordance or information, so that he would reject at least one of the points I've just listed.

    So, for example, looking at the first page of Chapter 2 of Gibson, 1979, we find "Air is 'insubstantial' and so is water, more or less. It thus affords locomotion to an animate body." It is on the force of such comments that it seems that Gibson is ok with physics. But, then you give him views that seem to rule out a role for physics, only they are not supposed to rule out a role for physics.

    ReplyDelete
  13. Perhaps, while at a very base (and rather reductionist) level, affordances have to deal with physical properties, and hence Physics. But they can also go further to deal with Psychological or biomechanical properties (like sitting down)

    ReplyDelete
  14. "affordances have to deal with physical properties". I'm not sure what "deal with" means here. There are plenty of non-reductive relations one could be talking about here. For example, it could be that affordances causally interact with physical properties. It could be that affordances are realized or implemented by physical properties.

    ReplyDelete
  15. Er, "I sort of expected that Gibson would have physics specific either an affordance or information" that should probably be "I sort of expected that Gibson would have physics *determine* either an affordance or information"

    ReplyDelete
  16. Take the electric field the shark uses to detect 'edible fish' in the Turvey et al 'laws' paper (Ken and I have been discussing it on his blog beginning here).

    The electrical field is physical energy, out in the world. A physicist would analyse it in terms of volts, current, or whatever else is required. An ecological psychologist takes the same field and analyses it in terms of the spatial-temporal pattern which lawfully specifies the affordance property, 'is edible by the shark'. It's the same physical 'world stuff', but different scientific abstractions. Only the latter is of use to a theory of perception, because the former has objects (volts, etc) that contain no meaning or ecological relevance - this would all have to be added back in if the physics version was the basis of perception. Gibson's argument is that organisms don't perceive in terms of volts, but in terms of information specifying affordances, because it's available and solves the otherwise unsolvable enrichment problem.

    Now, that spatial-temporal pattern is composed of electrical energy, so physics is in there and issues of the detection of electrical energy (as talked about in that NYT article I linked to) will clearly impinge on evolution. But that's not the same as claiming that the world as described by physics is the correct way description for a theory of perception!

    With respect to, say, locomotion, certainly properties describable by physics (length, load-bearing, etc) are relevant. But critically, when analysing locomotion, you should really be describing those properties in affordance terms (i.e. with respect to organism effectivities) because that is what the organism needs to perceive, not the properties as described by physics.

    Long winded, I know, but is it helping? The point that I think is not clear is that physics and ecological optics are two ways to abstractly describe the same bits of stuff in the world, and neither is intrinsically more privileged or useful than the other (although one will pop out as better for a given task). This is even true within physics; there are multiple geometries and physicists just move around to use whichever one works best. Gibson's move was to free perception from having to start with the wrong description, and all the problems that arise from that initial error then go away (and we have new, but potentially much more tractable problems to do science on).

    This is helping me, Ken! I had mixed feelings about how well I was making my point in this post and clearly it wasn't working.

    ReplyDelete
  17. "The electrical field is physical energy, out in the world. A physicist would analyse it in terms of volts, current, or whatever else is required. An ecological psychologist takes the same field and analyses it in terms of the spatial-temporal pattern which lawfully specifies the affordance property"

    Here it sounds like you are saying that the physics *does* specify the affordance property.

    But, earlier you also said, "You can't get to affordances via the objects of physics, " which seems to me to contradict this story.

    All it seems to me you are doing to avoid this contradiction is to merely stop repeating one side. But, tell me which story is wrong and why.

    So, that's one thing...

    ReplyDelete
  18. "The electrical field is physical energy, out in the world. A physicist would analyse it in terms of volts, current, or whatever else is required. An ecological psychologist takes the same field and analyses it in terms of the spatial-temporal pattern which lawfully specifies the affordance property"

    Personally, I am unsure that EP has a coherent account of the relationship between physics, affordances, and information, but let me try to set that aside and offer what I hope will be some constructive criticism.

    Go back to Chapter 2 of Gibson, 1979. Gibson talks about earth, water, and air and suggests that scientists have focused on the properties of these. He, however, apparently wants to shift attention away from the typical properties of earth, air, and water and focus on other properties. He wants to focus on the surface created by the air-water interface, for example. This is not a matter of redescribing, I would think, but instead a matter of shifting attention from the usual properties to other properties.

    Similarly, I would think, looking at optical flow is not a matter of redescribing things. It is a matter of shifting scientific attention from static viewing properties to dynamic viewing properties.

    Similarly, in looking at ledge's "stand-on-ability" one is not redescribing the ledge as much as drawing attention to another property (or better relation) of the ledge.

    Similary, in looking at water in terms of affording hydration one is shifting attention away from physical properties, such as its index of refraction or heat capacity, to a "meaningful" property of the water.

    This "shift of attention" analysis seems to me to get some familiar Gibsonian cases right, however, it also seems to break down when it comes to, say, the Mach bands and light. Where we can see how air and water generate a surface, it's mysterious, to me at least, how the physical ink gradient on paper could generate the kind of information in the light that you have proposed. Moreover, I think this "shift of attention" analysis breaks down for the electrical field case.

    ReplyDelete
  19. Here it sounds like you are saying that the physics *does* specify the affordance property.

    But, earlier you also said, "You can't get to affordances via the objects of physics, " which seems to me to contradict this story.

    The field isn't the object of physics; the analysis of the field, into volts etc, produces the objects of physics. Maybe 'objects' isn't how you'd refer to them; units? Properties?

    Ecological optics is a way to analyse an energy array to identify the spatial-temporal patterns which specify affordance (organism-relevant) properties. That pattern is information and is an object produced by the abstraction of the world into ecological optics. There is an energy array involved, which physics is capable of talking about. However, when it talks about it, it analyses it into organism-neutral properties; these are the objects of the abstraction of the world into physics.

    This is not a matter of redescribing, I would think, but instead a matter of shifting attention from the usual properties to other properties.
    It's a reanalysis. Shifting attention implies all these properties are equally available; they are to science, but not to a perceiving organism.

    ReplyDelete
  20. "This is not a matter of redescribing, I would think, but instead a matter of shifting attention from the usual properties to other properties.
    It's a reanalysis. Shifting attention implies all these properties are equally available; they are to science, but not to a perceiving organism. "

    Shifting attention (and by this I mean shifting the attention of psychologists, not of perceivers) implies that all the properties are there--that they exist. It doesn't say that they are equally available for perception. I take it that there are all these properties in the world and some are perceptible (by, e.g. humans) and some are not.

    So, for example, high UV radiation exists just as much as does red light, but on the mainstream vision science approach, High UV is not perceptible by humans, where red light is.

    ReplyDelete
  21. Ok. I'm assuming that electrical fields are physical objects. So, if you are getting affordances from electrical you are violating your principle "You can't get to affordances via the objects of physics".

    Now, you can say that electrical fields are not physical objects or maybe you can say that electrical fields have non-physical properties, such as affording fish (or whatever), in addition to their physical properties.

    I'm getting the picture that physical objects and properties aren't ecological objects and properties, but I think you've got a problem if you just say that they are unrelated or that they are not unrelated. The first seems false and the second is unilluminating.

    ReplyDelete
  22. Here's a simple picture might have that I can at least understand.

    Physical objects and properties
    realize/implement
    Affordances
    which are specified by
    Information
    which is picked up by
    Perceivers.

    Or, here's another picture.


    Physical objects and properties
    are redescribed as
    Affordances
    which are specified by
    Information
    which is picked up by
    Perceivers.

    Those pictures at least make sense to me (though they seem to me to be false). But, then they both conflict with "You can't get to affordances via the objects of physics"."

    ReplyDelete
  23. Please find for me, within the modern science of physics, the property "edible-by-a-shark". Then I might be interested in arguments that affordances are in physics.

    ReplyDelete
  24. Neither account above presupposes that "edible-by-a-shark" is part of modern physics.

    Both are ways in which a)"edible-by-a-shark" is *not* a part of physics, but b) still related to physics.

    This seems to be the kind of picture you sometimes want. Sometimes, because you also want to say "You can't get to affordances via the objects of physics".

    ReplyDelete
  25. In fact, it seems to me that both of the accounts I've sketched above are ones that you have articulated yourself. (You don't use "realize", but instead "implements".)

    ReplyDelete
  26. I don't mean to be insulting here, but you seem to be overlooking a fairly standard philosophical view, namely, that there are psychological properties that are distinct from physical properties, are not reducible to physical properties, but are nonetheless realized by physical properties. So, maybe pain is not a physical property, but a property in psychology. It is, one might propose, not reducible to a physical property, but it is nonetheless realized by physical properties.

    Set aside the merits of the view. I'm just saying that this is at least a live option among philosophers and there are times when EPists can embrace it.

    ReplyDelete
  27. As I just said on your blog: if all you have is physics, all you get are patterns in energy. For that pattern to specify an affordance (ie be information) you need the ecological laws within which the specification relation can be supported for that niche.

    Perception only has access to affordances, via information; it does not have access to the physical process by which the information came to be, nor the 'physics' properties of the object or event in the world that affords a behaviour. So the physics route is the incorrect ontology for a theory about perception; ecological laws are simply a viable alternative.

    ReplyDelete
  28. Apologies for bumping an old post. I'm a physics grad student with an interest in cognitive science. I'm really enjoying your blog, it's making me think!

    I understand what you're getting at with your comment that "You therefore have to get to affordances by detecting information about affordances, not units of physics."

    A typical Physics 101 problem goes like this: There's a 25kg object hanging from a massless rope that runs over a frictionless pulley, connected to a 5kg block on a 30 degree incline, coefficient of friction 0.5 What's the acceleration in meters per second?"

    A typical problem in life goes like this: I'm trying to decide whether I want to get a large case of beer (more beer!) or a smaller one (less difficult to carry home). Kilograms & meters per second don't come into this. Rather, "That looks heavy, it's a long way to my house, the last time I took two jugs of milk & my shoulders hurt," etc.

    The main differences I see are these: (1) Information gathering: Measurement gets replaced by estimation, because that's all the organism has (2) Analysis: exact calculation based on mathematical laws established by scientific experiments gets replaced by estimation based on personal prior experience some sort. (3) The goal: being able to quantitatively predict the behavior of a simplified system under controlled conditions, versus having a robust 'algorithm' that almost always gets you close to the optimum quickly, and always has *some* answer, even in unfamiliar situations, or in situations with insufficient information.

    The similarities: both processes are exercises in model-building. Both get some information from the world, then act on that information according to models, which in turn have been built on prior information that came from the world. This information is filtered in very specific ways, however; most of the information about either sort of problem is disregarded, and only a few bits enter either analysis.

    I think this goes along well with your ideas about calibration. (You'd be surprised how large a role calibration plays in physics: scientists are reluctant to build devices that are significantly bigger than existing ones because we often don't have the ability to predict accurately how experiments will scale.)

    I'm coming at this from the perspective of being a physics teaching assistant/tutor for several years. I've spent many hours trying to convince people that it's worth taking a look at the world from that 'privileged 3rd person perspective.' It's not always easy for them. And I have heard my fair share of comments such as "But how are we going to use this in the 'real world'?" My initial response, as a physicist, is to say "Well, you might not use it, but it still works that way! So why not learn it?" But I don't think people see it that way. I suspect that many students roll their eyes, because how many times in life do you have a frictionless, massless pulley?

    & speaking of perspectives, I think the difference between Cartesian & spherical polar coordinate systems makes an interesting analogy to all this: spherical coordinates are natural because vision works that way, and the origin of the coordinate system has a special place in the system, which the origin in Cartesian coordinates lacks. (That is, 'everything revolves around me.')

    PS: thought you might like this cartoon, related to the 'outfielder problem:' http://www.gocomics.com/foxtrot/2006/01/08
    Or this one, about frictionless, massless etc: http://xkcd.com/669/

    ReplyDelete
  29. There's a few things going on here

    1. Perception is an act of measurement, not estimation, the same as any other device. However, it's measurements are calibrated by 'capability to act', not SI units. Exactly what this action scale is, we aren't quite sure; two first swings have been body scale (eg eye height for walking, leg length for stair climbing) and 'effort' (see this post). The calibration will be task specific and about our ability to do the task, it's the details we have yet to work out.

    2. In the ecological approach, the end result of this measurement is not model building - it's action. We directly perceive the world in action scaled units and use this information to coordinate and control action, without mediation by anything resembling a model.

    These (important) details aside, you are on the right track here. Calibration is a new concept for psychology, because people haven't been treating us as devices of any kind; embodiment gives you information to calibrate a measurement, though, rather than having to compute anything.

    Love that Foxtrot cartoon - I'm stealing that :)

    ReplyDelete
  30. 1) Point taken. I should have said, inexact measurements in organism-centric 'units', rather than 'estimation.' Estimation would be saying 'That looks like 6 ft.' Not what we typically do. (Although I know a guy who can estimate distances up to 40 ft to less than 1/2 inch, pretty accurately by my non-scientific test of his ability! The rarity of that kind of talent ought to tell us something.)

    So I pose a question: is this measurement quantitative or qualitative? Do these distinctions even mean anything in this context? My first pass at an answer: they are qualitative measurements about quantitative things.

    2)Right, so the model is unnecessary & gets whacked by Occam's razor, according to your thesis.

    Calibration is obviously how we learn (motor) skills: we take a stab at a new task, we fail, we form some correction to our behavior based on new information, & we try again. (I say obviously, but of course I risk sounding like an Aristotelian.) It surprises me that this isn't well-accepted.

    ReplyDelete
  31. Everyone thinks that's how learning works; the tricky bit is how the correction is implemented.

    is this measurement quantitative or qualitative?
    It has to be quantitative in some sense. A simple example is prehension - reach-to-grasp. People scale their grip aperture to the size of the object ahead of contact (they go slightly larger) and this scaling is controlled by the perception of the maximum size they might encounter (if you come in at an angle you're on a hypotenuse, not the distance straight across). At the end of the day, the hand actually has to make contact with an object so you have to actually match the size quantitatively.

    But, that said, the way we do it is to control things on the fly and make adjustments. So the initial qualitative fit of the maximum grip aperture gets converted to a quantitative fit as we come into contact and close off the grip.

    Hm. I'll have to think about that some more. Interesting.

    ReplyDelete
  32. So I pose a question: is this measurement quantitative or qualitative? Do these distinctions even mean anything in this context?

    The quantitative vs. qualitative distinction should ever be on the "suspect" list. Things that seem like obvious "qualities" can be quantified if need be, and pure quantitative talk typically hides a bunch of qualitative stuff that everyone simply takes for granted for the purpose of measurement.

    Now, sometimes it is useful to make that distinction... but we should always keep in mind that it is useful for that purpose, and not some deep deep truth about the world.

    ReplyDelete