91

Seeing-As in the Light of Vision Science NED BLOCK New York University Introduction I will discuss the issue of what we see, using the framework of Tyler Burge’s wonderful and monumental book Origins of Objectivity (Burge 2010). On Burge’s view, every percept is constituted by a “perceptual attributive” (that represents an attribute) and a singular element (that represents an individual). The format of a percept is map-like or iconic, and could be symbolized as “That F”.1 The most basic perceptual judgments and beliefs are conceptualized and propositionalized versions of percepts in which the attributive need not be bound to a time and place. What are the actual perceptual attributives? Vision science picks out a small set of basic low level attributives that are products of sensory transduction and are causally involved in the production of other visual attributives: shape, spatial relations (including position and size), geometrical motion, texture, brightness and color. Burge discusses a higher level attributive for objects (“integrated body”) and considers that there may be some higher level attributives for some biologically important properties like food, danger and shelter. However, he notes (and I agree) that probably there are no culture-specific higher level attributives for teacups and recessions (p. 101). Terminological note: I will abbreviate the list of low level environmental attributives with “color, shape and texture” and I will refer to all attributives that are not low level as “high level”. 1 More fully, the singular element in the percept is an occurrent context-bound application of “that” referring to a non-repeatable property-instance such as an object or event or a “trope”. SEEING-AS IN THE LIGHT OF VISION SCIENCE 1Seeing and Concepts One tradition in philosophy holds that all seeing is conceptual. Brian O’Shaughnessy (O’Shaughnessy 2012) argues (p. 42) that according to Wittgenstein, “…the work of the Understanding lies at the center of visual perceptual experience.” This point of view appears in a different form in Jerry Fodor’s forthcoming review of Burge’s book (paraphrased with permission). Fodor’s argument is: (1) No seeing without seeing as; (2) No seeing as without conceptualization; (3) No conceptualization without concepts. So seeing is inherently conceptual. Fodor has justified (2) on the ground that “…perception involves constancies and constancies require inferences, so those visual attributions involve concepts” (Fodor 2007). Since Helmholtz, computations whose successful operation depends on facts about the world have been called “inferences” but this is a metaphorical use of the term that does not justify the attribution of concepts to the visual system. Premise 2 is false. Burge refutes the idea by appealing to the fact that we share some perceptual attributives with lower animals, even insects. (Note that insects have color vision). Further, given that we know that some insects can see but we do not know (and indeed doubt) that they are capable of conceptual thought, we should not suppose that there is any a priori argument that visual attribution requires concepts. “Concept” is used in different ways so this disagreement may seem verbal. However, Burge, Fodor and I agree that there is a joint in nature between percepts and concepts, so there is an important theoretical question of how to characterize it. Burge and I agree that percepts do not constitutively function in propositional structures, whereas concepts do. More specif ically, as Burge notes (2010, p. 545), perceptual attributives function to modify perceptual applications to particulars, whereas conceptualized attributives function in what Burge calls pure attribution, to attribute properties to something already picked out by other means. And the abilities involved in pure attribution involve potential to function in propositional inference. I do not have the space to discuss controversies over whether there is in fact a joint in nature between perception and cognition (but see (Firestone 2013; Firestone and Scholl 2013)) except to note that such a joint is compatible with causation from one to the other and with the existence of borderline cases. There is a joint in nature between living and non-living things but also borderline cases such as viruses and prions. Acquiring New Visual Attributives? Endre Begby describes Burge’s view as a “peculiarly reductive account of what we are capable of perceiving, properly speaking” (Begby 2011). One thing that is supposed to be reductive is that there are no visual attributives specific to teacups and CD-cases. 2 NEDBLOCKThe claim that there are culture-specific perceptual attributives is often based on the phenomenological consequences of perceptual expertise—e.g. learning to recognize pine trees changes our experience of them (Siegel 2010). However, there is an alternative explanation of this effect in terms of low level attributives. Early vision is highly influenced by contours in the environment. For example, Mary Peterson has shown that viewers are more likely to see the white part of the left item of Figure 1 as figure (rather than ground) compared to the white part on the right item (Peterson, Harvey et al. 1991). (The border appears to shape the figure but the ground appears shapeless, which suggests something for which there is ample additional evidence (Peterson and Skow 2008), that the figure-shape suppresses the ground-shape.) The difference between the left and right no doubt reflects greater familiarity with contours of upright standing women in western dress than inverted standing women. Recognitional Coextension If moderately complex pictures—e.g. scenes or faces—are presented one after another with no advance information, subjects can perceive them moderately well at a rate of 6 per second and better than chance at 12 per second (Potter, Staub et al. 2004; Potter, Wyble et al. 2014). We must do this on the basis of bottom-up processing of low level features since there is no time for a substantial top-down influence (VanRullen and Koch 2003). Are there clusters of low level features that are what I will call “recognitionally Figure 1. Viewers are more likely to see the white parts of the left item as f igure than for the inverted version on the right though Gestalt principles favor the small, closed, convex and symmetric black portions in both cases. From (Peterson, Harvey et al. 1991) courtesy of the American Psychological Association. SEEING-AS IN THE LIGHT OF VISION SCIENCE 3coextensive” with, e.g. faceness that is, coextensive with faceness-to-theextent-that-we-can-recognize-it? If so—as it must be for quick context-less recognition to be possible—the question arises as to how we can possibly distinguish the use of a face-attributive from a recognitionally coextensive congery of low level attributives. Susanna Siegel (Siegel 2010) has argued for culture-specific perceptual attributives and even perceptual attributives for named individuals, e.g. being John Malkovich. Her “method of phenomenal contrast” appeals to the best explanation of pairs that differ phenomenally, and so in principle allows for experimental evidence as part of the best explanation, but when she actually applies the method, only armchair considerations are used. However, only empirical evidence can distinguish among the following: (1) complexes of low level attributives, (2) the high level attributives that are recognitionally coextensive with them, (3) conceptual attributives that are part of perceptual judgments rather than perceptions themselves. I will discuss three empirical phenomena. The first will provide prima facie evidence for perceptual attributives for facial expressions but will not settle the recognitional coextension issue. The second will show that there are high level attributives for some aspects of face perception but will not settle the perceptual rather than cognitive nature of those attributives. The f inal case will consider the issue of whether the technique used in the first two discussions really isolates perceptual attributives as opposed to cognitive attributives instantiated in an occurrent perceptual judgment. Unfortunately these three cases concern three different classes of attributives, the last is not high level and none of the high level attributives are cultural along the lines of teacups and CD-cases. My concern is more with the methodology of answering the questions rather than the actual answers. That methodology is based on perceptual adaptation, what John Frisby (from an idea of John Mollon) called the “psychophysicists’ microelectrode” (Mollon 1974; Frisby 1979). Just as the neuroscientist can first raise, then reduce a neuron’s firing rate by direct stimulation with a microelectrode, the psychophysicist can first raise, then reduce a neural system’s activity by stimulating it with its preferred stimulus. Perceptual adaptation was known to Aristotle who described (in “On Dreams”) what we now call the “waterfall illusion” in which “…when persons turn away from looking at objects in motion, e.g. rivers, and especially those which flow very rapidly, things really at rest are seen as moving” (Aristotle 1955, p. 731). Staring at something moving down raises the threshold for detecting downward motion, biasing the percept towards upwards motion, so stationary things look like they are moving upwards. The middle picture in Figure 2 is ambiguous between anger and fear. When one stares at the fearful face on the right, the threshold for firing of neural systems (Butler, Oruc et al. 2008) that code for fearfulness is raised, 4 NEDBLOCKFigure 2. Cover the two pictures on the left with a blank piece of paper. Stare intently at the picture on the right for 1 minute. Then look at the center picture. Now cover the two pictures on the right and stare intently at the picture on the left for one minute. Now look at the center picture again. It will appear to have a different expression. Similar experiments show adaptation that is at least coextensive with facial aspects of race, gender and individual identity. From (Butler, Oruc et al. 2008) with permission from Elsevier. so the perception of the middle picture is biased towards anger. And the opposite happens when one stares at the angry face first. Similar phenomena occur for other facial expressions, face identity and face gender. This phenomenon grounds a prima facie case that we have visual attributives for facial expressions. And there is additional evidence that the effect is not mainly due to recognitionally equivalent low level attributives. One can vary the low level properties and so long as the emotional expressions are kept constant, adaptation obtains, though diminished somewhat, suggesting that face perception utilizes both low and high level attributives. It would take baroque congeries of low level properties to explain this fact (Butler, Oruc et al. 2008). (See also (Rossion and Boremanse 2011)). A further line of evidence (Susilo, McKone et al. 2010) is based on two ideas. The first idea is to compare adaptation effects between inverted and upright faces on the assumption that adaptation effects that work for inverted faces are likely to derive from low level attributives, whereas extra adaptation for upright faces is likely to involve high level perceptual attributives specific to faces. Susilo, et. al. note that adaptation for height transfers from one shape to another. For example, staring at an elongated ellipse (the “adaptor”) makes a rectangle (“adaptee”) look shorter and staring at a short adaptor ellipse makes an adaptee rectangle look longer. Susilo et. al. designed an experiment that examined transfer of adaptation from the letter ‘T’ of various heights to faces whose eye to mouth distance also varied as shown in Figure 3 and also the reverse transfer of adaption. They used SEEING-AS IN THE LIGHT OF VISION SCIENCE 5adaptor items (both ‘T’s and faces) of three different elongations, testing the effects of these differences in adaptors on a variety of elongations of “adaptee” faces and ‘T’s. The technique was to ask the subject to stare at a face or a ‘T’ of one of the three elongations for 4 seconds, then to view a face or a ‘T’ of one or another elongation, judging whether the test item was longer or shorter. A subject who has adapted to an elongated stimulus will see another stimulus as shorter than it would otherwise have looked, so to the extent that perception of upright and inverted faces is low level, transfer of adaptation within stimulus types should be the same as between stimulus types. Using ‘F-F’ to mean transfer from face to face, there are four types of transfers that should all be the same if perception is entirely low level, i.e. F-F = F-T =