Wednesday, August 27, 2014

Visual experiences aren’t always transparent
In 1976, when I delivered the John Locke Lectures at Oxford, I often spent time with Peter Strawson, and one day at lunch he made a remark I have never been able to forget. He said, "Surely half the pleasure of life is sardonic comment on the passing show".  This blog is devoted to comments, not all of them sardonic, on the passing philosophical show.
Hilary Putnam

Back on 6/11 (Escher Colors) I wrote;
“My own experience thus prepared me to accept one of the points that Block makes: the view (common to ‘representationalists’ and at least some ‘disjunctivists’ in the philosophy of perception) that the phenomenal quality of a subject’s visual experience upon looking at (hearing, feeling, smelling, etc.) a certain portion of her environment is exhausted by the objective appearance-properties (e.g., looking such-and-such a shade from such-and-such a point in space under such-and-such lighting conditions) of that portion of the environment is untenable. This view is, of course, a strong form of ‘naïve realism’, and while I think naïve realists are right to say that what we see (hear, feel, smell, etc.) when we perceive objects and events in our environment are properties of those objects and not properties of our qualia (something Block also thinks is right, as his paper makes clear), it is a mistake to say that describing what we perceive in objective terms also completely describes the phenomenology of the perceptual experience.”

In today’s post I want to discuss an argument representationalists (aka “strong representationalists”,  and “intentionalists”) often use to defend the view that I criticized, the argument from the “transparency” of phenomenal character. Here is how a leading representationalist, Michael Tye, opens his presentation of the argument:

 “Representationalism is a thesis about the phenomenal character of experiences, about their immediate subjective ‘feel’. At a minimum, the thesis is one of supervenience: necessarily, experiences that are alike in their representational contents are alike in their phenomenal character. So understood, the thesis is silent on the nature of phenomenal character. Strong or pure representationalism goes further. It aims to tell us what phenomenal character is. According to the theory developed in Tye 1995, phenomenal character is one and the same as representational content that meets certain further conditions. One very important motivation for this theory is the so-called transparency of experience. [1]
—and here is how he goes on to explain “transparency”:
“Focus your attention on the scene before your eyes and on how things look to you. You see various objects; and you see these objects by seeing their facing surfaces. Sense- datum theorists claimed that the facing surfaces of the objects are themselves seen by seeing further immaterial surfaces or sense-data. The sense-datum theory is unacceptable, however, for a whole host of familiar reasons. Intuitively, the surfaces you directly see are publicly observable physical surfaces. They are at varying angles to the line of sight and varying distances away. They can be photographed. In seeing these surfaces, you are immediately and directly aware of a whole host of qualities. You may not be able to name or describe these qualities but they look to you to qualify the surfaces. You experience them as being qualities of the surfaces. None of the qualities of which you are directly aware in seeing the various surfaces look to you to be qualities of your experience. You do not experience any of these qualities as qualities of your experience. For example, if blueness is one of the qualities and roundness another, you do not experience your experience as blue or round.”
For normal adult viewers, this paragraph does describe “how it seems”; in philosophers’ jargon, it describes part of the phenomenology of visual experience. But Escher Colors did criticize an assumption that is clearly implicit in what Tye writes, the assumption that the “qualities” I seem to perceive are qualities that a physical surface could actually have, (NB: Tye, like Byrne is a physicalist about colors). In Escher Colors I argued that there is no such thing as “really” being the exact shade that something seems to be when I look with my left eye (the right eye being closed) as opposed to “really” being the exact shade that it seems to be when I look with my right eye (the left eye being closed); those “Escher colors” may, phenomenologically, seem to be qualities of physical surfaces but they can’t be. That was the point of that post. But the purpose of the present post is not simply to remind you of that. Tye’s intention is to use the phenomenological “transparency” of visual experience to argue for the thesis that “phenomenal character is one and the same as representational content that meets certain further conditions”; and for that purpose, transparency (understood as Tye does) needs to be a necessary property of all visual experiences - an intrinsic  property of such experiences. The purpose of this post is to argue that there is impressive empirical evidence that transparency is not intrinsic, but is the result of learning.
Held’s experiments (1)
There is strong experimental evidence that in certain cases, described by Richard Held in two different series of experiments many years apart, some non-human mammals and even some humans do not visually perceive colors and shapes as “out there”. I conclude that “transparency” is (normally) the result of early learning (but, contrary to  Block’s view, mentioned in my previous post, not something we can override if we try and we know how to do it. At least I see no evidence for this claim.) The first visual experiences of many mammals, including both cats and humans, are not experienced by those organisms as properties of surfaces “out there”. “Out thereness”, recognition of things as having locations in places accessible to both sight and touch, requires that the correlations between visual space and sensorimotor (tactile, or “haptic”) space be learned. The disposition to learn such correlations quickly may well have been selected for in the evolutionary history of our visual and haptic systems, but some hours or days of learning are still needed. Here is some background.
In 1963, when Richard Held and I were colleagues at MIT, he showed me the following experiment that he and Alan Hein performed[2]: two kittens were placed in baskets that were at the opposite ends of a pole that was free to rotate around a vertical axle[3]. One basket had holes that permitted the kitten in that basket, kitten A, to push against the floor and thus to determine how the basket would move, at least to a limited extent. Kitten B’s basket had no holes; willy nilly, kitten B experienced spatial motion when kitten A moved – the mirror image of kitten A’s spatial motion, in fact - but no sensorimotor feedback. When kitten B was taken out of its basket, it had no recognition of any sensorimotor affordances at all. If one poked a finger towards its eyes, it stuck out its paws (an innate reflex), but not in the direction of the approaching finger. If it were put near the edge of the table, it would fall off (or would have fallen if there were not something to catch it) as often as not. But when Kitten A was taken out, it was able to position its paws in front of the threatening finger, it never walked off the edge of the table, etc. The results strongly suggest that the kitten that had not learned those correlations did not see the visual data it experienced as “out there”, and so the phenomenal aspects of its visual experiences were not transparent.
Held’s experiments (2)
In 1688 William Molyneux sent a message to the philosopher John Locke asking: “Suppose a man born blind, and now adult, and taught by his touch to distinguish between a cube and a sphere of the same metal ... Suppose then the cube and sphere placed on a table, and the blind man be made to see: query, whether by his sight, before he touched them he could now distinguish and tell which is the globe, which the cube ...?”[4]
In 2011 a team of researchers including Held and led by Pawan Sinha of MIT, published a negative answer to Molyneux’s question.[5] Similar results were reported earlier by a group referred to by Held in “Visual-Haptic Mapping and the Origin of Crossmodal Identity”[6]. The research involved studying youngsters who had been blind from birth, after they had lenses surgically implanted under the auspices of Project Prakash in India. Here is  the Abstract of “Visual-Haptic Mapping”:
“We found that the congenitally blind person who gains sight initially fails to identify seen objects with their felt versions: a negative answer to the Molyneux question. However, s(he) succeeds in doing so after a few days of sight. We argue that this rapid learning resembles that of adaptation to rearrangement in which the experimentally produced separations of seen and felt perceptions of objects are rapidly reunited by the process called capture. Moreover, the original ability to identify objects across modalities by the neonate may be assured by the same process.”
—And here is a description of the subjects studied by Ostrovsky et al:
“subjects’ responses were driven by low-level image attributes; [when asked to point to objects] they pointed to regions of different hues and luminances as distinct objects. This approach greatly oversegmented the images and partitioned them into meaningless regions, which would be unstable across different views and uninformative regarding object identity. A robust object representation is difficult to construct on the basis of such fragments”  (Ostrovsky, Y., et al. 2009. “Visual parsing after recovery from blindness”. Psychological Science 20, 1484-1491. 
In sum, these subjects’ visual experiences were anything but “transparent” to them. The representational content of visual experiences is something we learn to recognize. It is not intrinsic, and a fortiori not identical with their phenomenal character.









[1] Michael Tye, “Representationalism and the Transparency of Experience”, Noûs 36 (1), 137-51.

[2] Held, R. and Hein, A (1963) Movement-Produced Stimulation in the Development of Visually-Guided Behavior. Journal of Comparative and Physiological Psychology, vol. 56, 5: 872-876.
[3] For a picture, see: http://pirate.shu.edu/~hovancjo/exp/kit_caro.htm
[4] Morgan MJ. Molyneux's Question: Vision, Touch, and the Philosophy of Perception. Cambridge University Press, 1977. Quoted by Held in “Visual-Haptic Mapping and the Origin of Crossmodal Identity”, http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2749576/.
[5] Richard Held, Yuri Ostrovsky, Beatrice de Gelder, Tapan Gandhi, Suma Ganesh, Umang MathurPawan Sinha. “The Newly Sighted Fail to Match Seen with Felt”, Nature Neuroscience 14 (2011), 551-553.
[6] Ostrovsky, Y., et al. 2009. Visual parsing after recovery from blindness. Psychological Science 20, 1484-1491.

5 comments:

  1. "What (Black and White) Mary Didn't Know", the music video:

    ReplyDelete
  2. According to what I might call here the "Kant- Putnam critical view" (please correct me if I am wrong or off the mark), the world is the way it is independently of the purposeful efforts of any organism trying to understand that world and say anything about it, and to say anything about it, even to distinguish between an object and its properties "out there", is dependent on a language and categorial system, etc. Thus an observer might say that the reflectancy profile of the object, the effect of the bounced light rays on the visual system of an organism, all that is the way it is independently of the organism and its attempts to understand the world. But the organism acts in order to understand the world and makes distinctions in the optic information coming into the visual system, distinctions that are, we might say, adaptive or useful, by moving, manipulating an object, changing point of view, etc. Some of these distinctions, whose existence depends on action, are marked by endogenous signifiers, and linguistic categories are constructed and are used to talk about visual phenomena. But the recognition of distinctions in the exogenous signifiers, the reflected light, is not dependent on the constructed linguistic categories or restricted by them. It would, however, be dependent on action. I want to suggest the importance of action (and inter-action with the world) not only for learning, but for knowing anything about the world. Action and the critique of action, the development of criteria for the evaluation of action, from perceptual activity to attempts to make sense of the world and expressing the results, seem to be pretty central. I know I haven't quite got to what you are talking about here. I'm struggling to get there.

    ReplyDelete
  3. The visual information is not identical to its significance, and so is not "seen" in the same sense. An organism can distinguish between two phenomena involving a difference in the reflectancy profile of the surface, and even distinguish between the two associated experiences, and this can be described as "seeing the difference", etc. But, for a human subject (but not only a human subject), if the subject distinguishes the experience AS a distinction between, e.g., blue and green, and the observer says that the subject "sees the difference between the blue and the green", then the observer is not using the term 'see' in the same sense as in the first case, since the latter involves the significance of the perceived difference. One could say as well, that the predator "sees2" the difference between the prey and the leaf by relying on being able to "see1" the slight difference in the reflected light from the leaf and the prey. This is something that always bothered me about JJ Gibson's theory (affordances, etc.): he didn't seem to make this distinction in the use of the term 'see' and then equivocated.

    ReplyDelete
  4. Is 'representational content' what is actually being learned here? Or is it the reduction of perceptual uncertainty? Certainly the problem faced by the *brain* isn't the problem of 'what to make of these colour splotches,' but rather how to successfully navigate environments given concurrent and past sensory feedback.

    The question then, is one of where the problem of colour-splotches fits in the explanatory food-chain. The presumption is that this how we 'successfully navigate environments given concurrent and past sensory feedback,' via representations, and that 'learning transparency' simply is this learning to successfully navigate. The presumption, in other words, is that phenomenology discharges the kinds of functions it appears to.

    It's worthwhile, at this point, to recall Dehaene's 'law,' the neuroscientific maxim (important in theory building) that we are generally unaware of the glaring gaps in our metacognitive awareness. You could build a strong empirical case that intuitive phenomenological judgments, exercises like Tye's, are likely little more than exercises in self-deception. It is almost a certainty that Tye, all transparency theorists actually, are drawing conclusions on the basis horrifically incomplete information.

    Think of how much evolutionary history and biomechanical complexity is required to reliably identify a loaf of bread in the relatively simply environment that is your kitchen. Given this, how much evolutionary history and biomechanical complexity is required to reliably identify 'phenomenal content' in the astronomically complicated environment that is your brain?

    There's plenty of research in metacognition out there, and the picture it's revealing is not one that philosophers possessing any phenomenological sympathies will like. The more we learn, the more clear it's becoming that our metacognitive capacities are fractionate and specialized, which is to say, heuristic. It should come as no surprise that evolution provided us with just the metacognitive werewithal we required to discharge a variety of practical tasks. The information available for 'phenomenological reflection' is this problem-specific information - what else could it be? The million dollar question becomes, why should we assume that this information is suited to solve theoretical problems? Because it 'feels' like it can is the nearest answer I can come up with, because heaven knows precious few of these problems have ever been solved!

    I know these tilting-at-windmills type criticisms trigger the sense that the critic is simply trying to play cricket on the pitch, but the fact is the science has swallowed up this particular branch of speculation. It is simply no longer the case that one can ground claims on the presumption of phenomenological adequacy short of considering the growing amount empirical data suggesting, if not establishing, that no such adequacy can be presumed. (I have a roundup of many different sources at, http://rsbakker.wordpress.com/2014/06/25/the-philosopher-the-drunk-and-the-lamppost/ for those who are interested.)

    ReplyDelete