Preattentive Precursors to Phenomenal Properties

What are the relations between preattentive feature-placing and states of perceptual awareness? For the purposes of this paper, states of "perceptual awareness" are confined to the simplest possible exemplars: states in which one is aware of some aspect of the appearance of something one perceives. Subjective contours are used as an example. Early visual processing seems to employ independent, high-bandwidth, preattentive feature "channels", followed by a selective process that directs selective attention. The mechanisms that yield subjective contours are found very early in this processing. An experiment by Greg Davis and Jon Driver is described; it seems to show that multiple subjective figures can be coded in these preattentive, parallel stages of visual processing. I propose that some of these preattentive states might register the very same differences that, were one aware of them, would be phenomenal differences. Some arguments pro and con on this possibility are assessed. Today I want to describe some of the relations between feature-placing and phenomenal consciousness. At the feature end I will focus on some of the representations found in "early", or "preattentive" vision--the representations that are operative before you have selected something to which to attend, and indeed the representations on whose basis you make those selections. On the phenomenal consciousness end I will focus on the simplest kind of episode worthy of that label: an episode in which one perceives something and is aware of some aspect of what one perceives. Its perceptual origin gives the episode its distinctive phenomenal features, which often include that of which one is aware. Examples include seeing and thereby becoming aware of the red glow of the sky at sunset, or feeling a cool breeze commence as the sun goes down. One might call them episodes of "perceptual experience" or "perceptual awareness". In these states one is (typically) aware of some aspect of what one senses. One may, or may not, also be conscious of being in that state. It is possible that one is so absorbed by the sunset--the red glow of the sky, the cool breeze--that one is not also simultaneously aware of seeing the red glow, or of feeling the cool breeze. So a state of perceptual awareness (as used here) is not necessarily a state of which one is conscious. Instead it is a state in virtue of which one is conscious of (some aspect of) something one perceives. I want to try to convince you that the following state of affairs is possible: a subject S is appeared-to in some determinate fashion even though at that moment S is not conscious of any aspect of that appearance, or for that matter, even of its occurrence. To put it in other words: something can appear somehow to someone, even though the someone in question is not aware of the appearance, at all. Being appeared-to can be entirely divorced from being aware-of. If the argument works, then after today you will never again need to worry about what constitutes the kind of mental states that we call phenomenal consciousness. If the argument works, it isn't one kind, but (at least) two; and they can separate. Not only is this possible, but I think there is good evidence that it actually Clark: Preattentive Precursors 2 happens. But before we get there I need to explain what I mean by "phenomenal". This is best done with an example. Fig. 1. A Kanizsa square I. Phenomenal properties In fig 1, there seems to be a square. Its edges appear to be continuous, straight lines. It looks like there is an edge, even in the spaces between the pac-men. Finally, (depending on the contrast of the copy), the square might appear to be a little brighter than the background. It looks like a white square, lying on top of the four pac-men, somewhat brighter than its surround. Each sentence just uttered employs what Chisholm would call a "verb of appearance": looks, seems, appears. One traditional interpretation of the term "phenomenal property" is that they characterize the facts of sensible appearance. So the appearance as of a square, the apparent continuity and straightness of each edge, and the seemingly greater brightness of the square compared to its background are all instances of phenomenal properties (or relations). They characterize how something appears, where that something is something that one perceives. I'll (try to) restrict my use of the term "phenomenal properties" to apply only to attributes of sensible appearance: how something looks, appears, feels, or (generically) seems. This follows C. D. Broad's (1927) usage. Phenomenal properties characterize the "facts of sensible appearance": how things (in the broadest sense of the word) appear. At least some of these appearances are, as we say, mere appearances. Some of the characteristics attributed to what one sees when one looks at figure 1 do not in fact characterize figure 1; they are not true of it. For example, there is no continuous edge, and no straight edge, found in the gaps between the pac-men. If having four continuous straight edges is a necessary condition for being a square, then there is no square in figure 1. And the part of the figure that seems brighter than the background has exactly the same whiteness. The standard move, which I endorse, is to say: well, the clever arrangement of the pac-men in this figure puts us into a perceptual state in which there Clark: Preattentive Precursors 3 appears to be a square in front of us, even though in fact there is not. "It looks P" is explained as "I am visually representing it as being P." The figure causes me to enter a state that represents there to be a square, with continuous, straight edges, brighter than its background. But this representation is not veridical. On this line "phenomenal properties", in my restricted sense, are properties that the figure is represented to have. They are properties of the figure as represented. Sometimes, as here, those representations go awry. One difficulty with the terminology is that the term "phenomenal" is sometimes applied, not only to characteristics of the world as represented (as here), but also to characteristics of the representings of the world. Some philosophers will say that the stimulus causes one to enter a mental state that has "phenomenal character" (or "qualitative character"), where these "characters" are now thought of as properties of mental states, not properties of perceptible things as represented. I'll try to use the term "qualitative character" when talking of the properties of the mental state.1 A second difficulty with the word "phenomenal" is that sometimes it is used in the sense "merely phenomenal"--an appearance which is non-veridical, or illusory--and sometimes it is used in a sense that embraces any episode in which some manner of sensible appearance is apprehended, whether veridical or not. The latter sense is the one typically in play in discussions of phenomenal similarity. So for example one might say that the Kanizsa square in figure 1 is phenomenally similar to, or presents the same phenomenal properties as, some figure in which there is in fact a slightly brighter white square lying on a slightly dimmer background, with its corners aligned with pac-men. (We might use photographic paper and make the central square slightly brighter, so that there are real luminance differences at edges that are both straight and continuous.) The appearances presented by the illusion might be quite similar to (at the limit, perhaps identical to) those presented by the real thing. In such a case the appearances of the latter are not "merely" phenomenal; they are real. We can call them "sensible features", or "discriminable features"--here, specifically, visually discriminable ones. Something that (merely) looks like an edge can look eerily similar to something that is an edge. Ascription of the phenomenal property does not demand or say much of anything about the thing to which it is attributed. Instead it says something about the character of the occasion on which that thing was perceived. That occasion resembled occasions on which some real edges were seen. This gives us a clue for unpacking the traditional notion of phenomenal properties. Phenomenal properties characterize what is common across episodes in which (for example) something merely looks like an edge 1 I believe that early 20th century usage of the word "qualia" matched what I am here calling "phenomenal properties". The apparently greater brightness of the square was a quale presented by the figure. But as we all know, the word "qualia" has been hijacked, kidnapped, held for ransom, and released many times in its short lifetime, and now if anything I think the predominant usage is not as of a quality one might see something to have, but rather a property of mental states, in virtue of which it is like what it is like to see those things, or (alternatively) to be conscious of seeing those things. The "quale" is now, not the brightness, but the "feel" of the sensation of brightness-what it is like to entertain the state in which the world appears that way, for example. Clark: Preattentive Precursors 4 (but is not), and in which something is seen to be an edge. Both episodes might present the same appearance, but in only one of them is the appearance sustained. I tend to use "phenomenal" in this broader sense, in which it marks a commonality between occasions in which something looks like and is an edge, and occasions in which something merely looks like an edge. To mark the distinction when it is important, I will (try to) call the latter sorts "merely phenomenal" properties. They are the ones presented in occasions of nonveridical or illusory appearance.1 II. Explaining Looks One strategy experimental psychologists often use to try to explain such subjective contours is relatively straightforward: they try to find what it is about the stimulus that affects the system in a way that is similar to the way a real contour would. Edges are detected and "registered" quite early in visual processing, and what one would hope to find is that the array presenting a subjective contour is registered the same way as some kinds of stimuli containing real contours. I am using "register" in Fred Dretske's sense: these are internal, information-bearing states of the system, exploited by the system for the information they contain. What we aim to show is that Fig 1 causes the same such visual states as a stimulus containing real contours. Note that this strategy partially endorses the idea that "x looks P" is explained as "x is represented as being P". Instead of saying the state represents the stimulus as having an edge, oriented thusly, thereabouts, vision scientists are more likely to want to say that the stimulus causes an internal state which carries the information about a discontinuity of intensity, oriented thusly, thereabouts. The idea is that if we can show how fig 1 causes the same internal registration of such a state as does a real edge, then we can explain why fig 1 looks as if it has continuous, straight, edges. What might come as a big surprise to philosophers is that the edge detection mechanisms that yield subjective contours do their work very, very early. Much of it is already done by the time we are a mere four synapses away from the retina itself. It has been known since 1984 that there are cells in visual area V2, just four synapses up from the retina, that respond selectively to oriented contours, and respond almost as enthusiastically to "subjective" contours as to real ones (von der Heydt, Peterhans, & Baumgartner, 1984). Factors known to block formation of subjective contours also temper the enthusiasm of these neurons. (See figures 2 & 3). The stimuli for neuron 1 are shown on the left; spikes are shown by the dots in the horizontal traces. In (A) the neuron is responding to a (real) white bar that moves across its receptive field. The ellipse shows the receptive field 1 We speak readily of states of the system that serve to "register information" about visually discriminable features of the stimuli it confronts. We might say (as I sometimes do) that some of them register "phenomenal differences", but only if "phenomenal" is taken in the broader sense, and not as "merely phenomenal". In an illusion there is no real difference to be registered. Clark: Preattentive Precursors 5 Fig. 2. From von der Heydt, Peterhans, & Baumgartner, 1984 of the neuron, and the "+" sign is the point of fixation. In condition (B) we present just the two ends of that bar; no luminance difference ever crosses the receptive field of neuron 1. Yet it responds almost as enthusiastically to the "subjective" figure as to the real one. Formation of a subjective contour is disrupted if we present only one vertex of the two needed to define an edge, as in (C) and (D); here the neuron fails to respond. Likewise, neuron 2 responds to a real bar (shown on the right, in F) and almost as enthusiastically to a "virtual" one (in G). If however the open ends of the two white bars are blocked (end-stopped) with thin black lines, the subjective contour is not induced; in (H) we see the effect of this stimulus on neuron 2. III. Preattentive processing V2 is one of prime areas containing various examples of the "feature maps" of which I am so fond. It is part of what people think as "early" vision or "preattentive" vision or "visual preprocessing". The work therein must get done before selective attention can start its selecting. In fact the oriented edges, the shape, and the brightness of the shape against its background are just a few of many of the families of features whose registration, processing, and discrimination occurs preattentively. Models of early visual processing produced in the last forty years have grown increasingly sophisticated and complex, and in detail they differ markedly from one another. But certain global features of the landscape are found reflected in all of them (see Neisser 1967). They are: 1. That distinct visual attributes of stimuli are initially processed largely Clark: Preattentive Precursors 6 independently of one another, in parallel, for the entire visual field; 2. That these processing streams, "channels", or "feature maps" extract information of varying orders of complexity, have complicated linkages with one another, may reside in distinct portions of the visual nervous system, and in cases of brain damage may be selectively dissociated from one another; 3. That at some point there is a competitive selection process, whereby some results of some of these parallel processes are selected for additional, capacity-limited, processing; 4. That a result of winning that selection is that discriminations in the selected domain can be made more quickly and more accurately; that the additional processing allows for the resolution of details and structure that might otherwise be unresolved, and that there may be features that are resoluble only as a result of such selective attention; and 5. That sensory representations which are not selected do not gain all these benefits, though they may gain some. The initial stage is high-bandwidth analysis (e.g., lots of data processed per second) over the entire visual field, extracting "local" or low level visual features, with different "channels" carrying on the analysis of different visual attributes. For example, motion, hue, luminance, and shape are likely candidates for being distinct attributes, analyzed in distinct channels. Each such channel carries on a high bandwidth analysis of its favorite feature for the entire visual field. Fig 3. Pathways to V2. From Livingstone & Hubel, 1987. There are many different strands of evidence that support each of the five assertions, and many different points of disagreement between different models on each of the five. So for example the evidence for the first point-Clark: Preattentive Precursors 7 that distinct attributes are processed independently, and in parallel--comes from various sources. The first is neurophysiological: the careful tracing of visual pathways and different cortical areas connected by those pathways (see figure 3). A second strand of evidence is neuropsychological: particular types of brain damage can "knock out" a particular dimension of variations in visual appearance, while leaving others intact. Particularly compelling is evidence of double dissociation: some types of brain damage knock out A, while leaving B; while others knock out B, and leave A. Achromatopsia and akinesia show this dissociation: one type of central nervous system damage can leave a patient color blind, but still able to perceive motion. Another impairs discrimination of motion, while leaving color perception intact. A third type of evidence is purely psychological, and is based on various types of experiments, including, most prominently, reaction times in visual search tasks (Treisman 1998, 1996, 1993, 1988; Wolfe 1994, 1996a, 1996b) . In a visual search task a "target" is to be located as quickly as possible within an array of varying numbers of "distractors". Particular combinations of attributes of targets and distractors can make an enormous difference to the speed of such searches, and these differential times give evidence that some classes of features are processed independently of others. For example it is found that in some classes of features, a target with a unique value within that class can be located in a fixed (and brief) interval, no matter how many distractors are present. If the target is the red letter, and we have a single red H among green H's, blue T's, and green X's, the red H will "pop out", and reaction times for its identification will be low and basically constant, no matter how many distractors are present. Similarly if we have a single letter "X" among H's and T's, of whatever colors, the X will pop out. But if the target can be identified only by a conjunction of features--a single red H among green H's, red and blue T's and red and green X's--then search becomes much more arduous, and reaction times increase linearly with the number of distractors. It is as if each item in the array must be examined in turn, and such examination takes some finite time. The latter is sometimes called "serial search", or simply "less efficient" search. The relatively straightforward inference from this result is that feature families which allow "pop out" are processed independently of one another, and such processing is high-bandwidth, covering the entire visual field. From the results above, for example, color and orientation (which allow the diagonals of an X to pop out from the horizontals and verticals of H's and T's) would seem to be features processed independently of one another. Each allows pop-out. Only if a target is defined by a conjunction does search become less efficient. Pop out indicates a special kind of "preattentive" processing of the feature in question. Attention is effortlessly drawn to the singleton, no matter how many distractors are present. The processing to do this must happen before attention arrives on the scene, and it must encompass the entire visual field, since pop out occurs no matter how many distractors are present. So features Clark: Preattentive Precursors 8 in that group seem to be favored with independent, high-bandwidth, full field processing, completed preattentively. There are other tests for the independent processing of distinct families of visible attributes (Treisman 1988; Wolfe 1996b). One is "effortless texture segmentation", or the ability to discriminate regions defined by contours across which the feature in question changes. There is some boundary across which feature F changes; and the segments of that boundary in turn form a continuous closed curve. In this case the entire region might "pop out", or be immediately noticeable. The feature difference defining the texture can be minuscule: minute differences in orientation, closure (whether we have full squares or just three sides), and so on. Whereas other differences one might think would suffice do not: N's cannot be preattentively segmented from regions of backward N's, E's and F's are texture-wise indistinguishable, and so on (see Julesz 1984). The inference is that certain feature values are the subject of preattentive parallel processing, while others are not. Orientation and closure yes; the difference between E and F, no. Fig 4. The architecture of preattentive processing. Parallel feature channels, followed by a selective process. From Wolfe 1994. With these and other sorts of experimental probes available, a considerable amount has been learned about the different kinds and categories of features processed preattentively. Jeremy Wolfe (1996b) has provided a useful list of visual features that pass both of the two tests just mentioned. He calls them "basic features": color orientation curvature vernier offset size/spatial frequency/scale motion pictorial depth cues Clark: Preattentive Precursors 9 stereoscopic depth gloss various form primitives So for example a curved line (with basic feature number three) can be picked out quite efficiently if it is surrounded by straight lines, however many they may be (that's "pop out"); and it allows preattentive texture segmentation: a region filled with curved lines is effortlessly seen to have a different texture than one composed of straight ones. All of the features on the list pass both tests. Some are perhaps unfamiliar. We get "vernier offset" by breaking a straight line, and giving the second segment a slight offset, or sideways jog, relative to the first. (Human vision is extraordinarily sensitive to vernier offsets, detecting them at visual angles smaller than the angle between adjacent receptors in the retina.) "Pictorial depth cues" include cues for occlusion, perspective, shape from shading, and others that artists might use to make a two dimensional canvas represent a three dimensional scene. The last ("form primitives") is the most controversial and complicated. For reasons that will become important later (see Wolfe & Bennett 1997) it is not exactly "shape", but includes local features of contours and surfaces that might together yield "cues" or evidence for shape. Wolfe mentions, for example: line termination, intersection, and closure (e.g. is the contour a closed curve?). Others are less well attested: topological features such as holes, convergence, and perhaps containment. IV. Preattentive Counterparts? --No Now we are close to the crux of the question for today. What are the relations between these pre-attentive registrations of discriminable features, and the post-attentive phenomena found (for example) in phenomenal consciousness? The simplest hypothesis, which I will favor, is that the differences registered by some of these preattentive visual states are the very same differences that, were one aware of them, would be phenomenal differences. Or: the features registered by some of these preattentive states are the very features that, were one aware of them, would be phenomenal features.1 This hypothesis is simplest because it urges that the only way the properties registered by some of these preattentive states differ from phenomenal properties is that here the subject is not aware of them. Otherwise they have all the same properties as your run-of-the-mill phenomenal properties. The hypothesis is apt to met with incredulity, sputters of outrage, and then finally an argument: "But that's not what the word means! Phenomenal properties must be conscious. The proposal that there exist unconscious phenomenal differences, or that (for example) DB is being appeared-to in a matter of which he is unaware, is speculative, useless metaphysics. It distorts language and can only sow confusion." Now everyone has Humpty-Dumpty's right to use a word however they 1 Some of the preattentive states occur so early that they are not possible candidates for selection by selective attention, so this claim is not true of features registered by all preattentive states, Clark: Preattentive Precursors 10 please, and one can, if one likes, reserve the words "phenomenal property" for just those ones of which someone is conscious. A frontal assault on that position will not succeed. But here's a new argument. The critical question is whether or not there can exist a family of other properties, which differ from these "phenomenal properties" and "appearances" in only one respect: that no one is aware of them. Could there be differences, that are exactly like those differences that you call "phenomenal", excepting only the fact that no one is conscious of them? Call them "counterparts" or "cousins" to the normal phenomenal properties, if you like; the idea is that the only way they differ from the kind normally called "phenomenal" is that no one is aware of them. Could there exist such creatures of darkness? If you think the answer to that question might be yes--that a "yes" is conceivable-then you are open to the possibility of an informative explanation of (these simplest varieties of) phenomenal consciousness. If you think the answer must be no, then I fear you have erected an a priori roadblock against the possibility of our ever understanding phenomenal consciousness. Consider the implications of a "no". To say "no" is to say that it is not the case that there exists a family of properties that differ from the normal phenomenal properties in only one respect: that no one is conscious of them. So for any phenomenal property P, any such candidate cousin Q must differ from P in some other respect besides those consequent upon the fact that someone is conscious of P, and no one is conscious of Q. In other words, quite apart from the fact that P characterizes something of which someone is conscious, and Q does not, there is also some other difference, or differences, between them. This implies in turn that that of which one is aware when one is aware of a phenomenal property cannot be identical to any property that can exist even if no one is aware of it. All the candidates must differ, not only in not being conscious, but also in some other respect as well. One might think that one could give a characterization of that of which one is conscious which is independent of the fact that one is conscious of it. In the typical two-term relation, one can specify the two terms quite independently of their relationship to one another. But the claim that phenomenal properties must be conscious implies that such a project must fail: phenomenal properties cannot be identified with any properties for which such an independent characterization is possible. Not only is the consciousness of them a mystery, but that of which one is conscious when one is conscious of them is a mystery as well. A solution to either requires the other be already solved. This is not a mystery inside an enigma, but something worse: two mysteries, each of which is wholly inside the other.1 1 This argument mimics one invented by David Rosenthal (1997, 735-36), in a different context. He argues that if we think of the difference between a mental state being conscious and being unconscious as a difference in the properties of the state itself, then we will never get an informative account of what it is to be conscious of a mental state. For that of which one is conscious could not be characterized independently of the consciousness thereof. Clark: Preattentive Precursors 11 V. Preattentive coding of multiple subjective figures Why might one think that some of the properties registered by some preattentive feature maps are nonconscious counterparts of phenomenal properties? The answer is that some very surprising experiments seem to indicate as much. The experiments feature our new friends, the Kanizsa figures. Early ones show that their merely subjective contours can be detected in parallel, preattentive stages of vision. The simpler sort of experiment shows "pop out" for merely subjective polygons. For example, Davis and Driver (1994) used a stimulus array in which some portions had the "pac-men" circles aligned so as to create the impression of a Kanizsa square, while in other portions the circles were rotated ninety degrees. In the latter condition the cut-outs are not correctly aligned, no potential edge is given two vertices, and no impression of a subjective square arises. They found efficient search and effortless segmentation of the subjective square, regardless of the number of nonsubjective (e.g., merely actual!) distractors. Enclosing each pac-man with a boundary also disrupts the formation of subjective contours, and Davis and Driver showed that the same stimulus, but with pac-men encircled, also gave inefficient search. There was still one place where the alignment was correct, but without the subjective contours it had to be found by laborious, item-byitem scanning. Davis and Driver (1998) provided further controls to show that multiple subjective figures are indeed "coded simultaneously", in parallel, preattentive stages of vision. Furthermore, this parallel encoding does not depend upon the intentions of the agent, but is automatic, or "obligatory", even in cases (such as theirs) in which it interferes with the intentions of the agent. These are beautiful experiments, worth describing in some detail. Figs 5 & 6 The target is a notched circle. In 6 it is occluded by a Kanizsa square. In all of them the target was a brown circle with notch cut out of it, aligned at one of the corners of a Kanizsa square. A simple way to see the problem is Clark: Preattentive Precursors 12 to compare fig 5 with fig 6. In fig 5 it is easy to detect the target--the notched circle. It is the only part of the stimulus that contains straight lines and a corner, so it "pops out" from among all the circles. Why is the same notched circle so hard to detect in fig 6? Notice that the array of circles, including the notched one, is identical in fig 5 and fig 6. But if you glance at figure 6 quickly, you will see what seem to be nine complete circles. It takes direct examination to determine that the one in the lower left is in fact notched, not complete. The problem is that in fig 6, the notched circle appears to be occluded by the square. This square is not a real square, but a Kanizsa square; that is, a "subjective visual surface". We can revert to ordinary verbs of appearance. In the lower left corner, there appears to be a circle occluded by a square. We have what psychologists call "amodal completion"; it does not look like a notched circle, lying on the same plane as the pacmen, but instead it looks like a complete circle, lying beneath the pac-men, and occluded by them. This is "occlusion by a subjective visual surface". The experiment has one other wrinkle, to ensure that we get appearances of occlusion. The appearances of brown circles are produced by stereoscopic fusion, of red through one eye and (a slightly offset) green through the other, viewed through red-green "3-D" spectacles. As fans of 3-D horror movies already know, this can generate a convincing impression of depth. In experiment 1, all of the brown circles, both whole and notched, were presented so as to appear on a plane "below" or "deeper than" the plane on which the other pacmen and the subjective surfaces appear (see fig 7, A). By reversing the color filters in the spectacles, one can reverse the disparity, and then the same array of brown circles, both whole and notched, will appear to float "above" the plane on which the pacmen lie. This was experiment 2 (see fig 7, B). Fig 7. In experiment 1, the 3D glasses were set so that all the brown regions appeared as in (A). In experiment 2, the disparities were reversed, so they all appeared as in (B). Clark: Preattentive Precursors 13 Results were what one might expect given the effect seen in figure 6. Search in experiment 1, with the notched targets apparently occluded by the Kanizsa squares, was difficult or "inefficient". (See fig 8). It takes a fair amount of time to find each target, and increasing the number of distractors increases that time in a linear fashion. The interpretation of such results dates back to Treisman (1988). Increasing the number of distractors slows performance because each potential target must be examined in turn. In contrast, when the same stimulus array is presented with reversed disparities, so that the brown circles appear to float "above" the plane of the page, they no longer appear to be occluded by subjective figures on that plane, and search is very efficient. The notched one now "pops out", in a relatively constant and short interval, no matter how many distractors (fig 8). Fig 8. Experiment 1 yields slow, inefficient search. 2 demonstrates "pop out". Why do the notched targets pop out in experiment 2, but fail to do so in experiment 1? That is, why is search so inefficient in experiment 1 (condition A)? Davis and Driver argue that there is just one viable explanation: in experiment 1, the notched circle target appears to be occluded by a Kanizsa square. Because of that apparent occlusion, it is hard to tell that the brown circle is in fact notched: it looks like a whole circle, lying below, and with a portion hidden by the square floating above it. As they put it, "the presence of the notched large circle target is not immediately apparent, because of amodal completion behind by the abutting subjective square" (Davis and Driver 1998, 176). Experiment 2 provides the critical data supporting this interpretation. It uses exactly the same stimuli, but changes the filters in the 3D glasses, so that depth relations are reversed, and the brown regions now occlude the figures "below" them. In that case, the notched targets pop out. All that changes is the occlusion--or rather, the appearance of occlusion--but that is enough. They run some other experiments to rule out a few other rival interpretations, but in the end the intuitive explanation seems well-confirmed: search is difficult in experiment 1 because in it the targets appear to be occluded. Clark: Preattentive Precursors 14 VI. Appeared-to, but not aware-of Now this result has some rather stunning implications. Drawing them requires reference to some of the assumptions made by models of selective attention, described in section III. To wit: If a stimulus array yields a unique instance of a feature within some feature-family, we would have pop-out; and if we had pop-out, then search in experiment 1 would be efficient instead of difficult. Hence the stimulus array in experiment 1 does not yield any singletons among its feature families: no feature that appears in it appears just once in it. Now, as Davis & Driver argue, the reason it is difficult to detect a target in experiment 1 is because each such target appears to be partially occluded by a Kanizsa square. That Kanizsa square is characterized by the various phenomenal properties described in section I: it appears to be an opaque square, with straight, continuous edges, slightly brighter than its background. If it were the only instance of a square, or of something brighter, or of something with continuous, straight edges in the stimulus array presented in experiment 1, then that one instance would "pop-out". Because it does not, we can conclude that to the subject looking at that stimulus array there must appear to be multiple squares--multiple opaque white surfaces, brighter than their background, with straight, continuous edges. This is why Davis & Driver say that their results demonstrate "parallel coding for multiple Kanizsa subjective figures" and that the experiments "are the first to show that several concurrent subjective figures could be coded together in parallel." (182). If there weren't several, simultaneously, then we would have pop-out. It is important also to notice that the target is not the subjective figure, but something occluded by subjective figures. Here in fact the subjective figures interfere with the search, make it more difficult; so if deliberate intention or focused attention were required for the "construction of subjective figures", one would think that they would not be present at all. This is why Davis & Driver say the coding of multiple subjective figures is "obligatory" or "automatic". It is not something these subjects could deliberately suppress. If we refer back to the architecture of selective attention depicted in figure 4, this combination of attributes--obligatory coding of multiple figures, in parallel, in an automatic, bottom-up fashion--yields just one locus where such coding could take place. "Parallel coding" happens in the preattentive feature channels of early vision. Such high-bandwidth, preattentive processing yields the information necessary to direct the selections of selective attention; it happens independently of, and before, those selections are made. Indeed, those selections are often made on the basis of the information contained in these preattentive registrations. So for example in experiment 2, the very same notched circles do not appear to be occluded by the Kanizsa squares, but rather to float "above" them; and in that circumstance, they "pop out". Information that the notch is a singleton must be made available to the "activation map" that selects that to which attention is about to shift. There are multiple Kanizsa squares apparent in the stimulus arrays presented in experiment 2, and so the Clark: Preattentive Precursors 15 coding for those multiple Kanizsa squares occurs as well in these preattentive, high-bandwidth, parallel channels. So the argument can be put in terms of information registered prior to, and independently of, the activation of selective attention. In fact it could be put in terms of the information that is made available to the selective mechanisms themselves--the ones that select that to which attention is about to shift. You certainly cannot attend to that information before the selective mechanisms receive it. I don't see any plausible argument for the claim that one is somehow "aware of" the content of these preattentive states. They have to occur earlier, and be finished before, any states arise that might be candidates for awareness. They provide the information guiding the selections of selective attention, determining that to which attention is about to shift. There is one more step needed to complete this part of the argument. If Davis & Driver are right, then multiple subjective figures are coded simultaneously, at these preattentive stages of visual processing. But a subjective figure is by definition a collection of nonveridical phenomenal properties; a Kanizsa square is something that looks like a square, seems to be brighter than its background, appears to have continuous straight edges, and so on. It must present such appearances if it is to produce the occlusion effect. As they say: Kanizsa figures could only act as occluders in this way if there was indeed some construction of an illusory surface between the inducers. (Davis & Driver 1998, 174) This way of putting it is amusing (What materials are used to construct this illusory surface? Who are the workmen? Where do they put it? etc) But Davis & Driver are both smart enough that I think they are probably just tweaking their noses at the ordinary language philosophers in the crowd--and doing this deliberately, knowing that they are doing it. What they mean is clear: we have multiple registrations of the sensory features as of squares-edges, contours, continuity, brightness, etc--and here many of those registrations are non-veridical. They represent features that are not present-that are "merely phenomenal". One might say that the "intentional object" of them is "constructed"; in any case it is clear that a subject who enters such states is being appeared-to. If the subject is not aware of that which is coded by preattentive stages of visual processing, then these experiments provide good evidence for a subject being appeared-to in manners of which that subject is not aware. QED. Let us back up and review. Edges and contours are basic features that are processed preattentively, in parallel over the entire visual field. The mechanisms that register edges can sometimes be put into that state even though the stimulus confronting it contains no real edge. It is then representing there to be an edge thereabouts when there is not. It responds perceptually just as it would if there were an edge. Such is a case of nonveridical appearance. The Davis & Driver experiments seem to show that the system can enter Clark: Preattentive Precursors 16 states registering a square (with continuous edges etc.), in region delta, prior to, and without, the ministrations of selective attention. So, in those delicate moments, the subject can be appeared-to in a determinate fashion--so as be presented the appearance of a square in region delta--without necessarily being aware of any aspect of that appearance. My hypothesis is that some of these preattentive registrations of sensible features are registrations of the very features that, once one is aware of them, are the "normal" phenomenal features. I think it is plausible to identify some of those psychological states as just the ones that subserve the discriminability of, and relative similarities among, various sensible features. For example, they carry the information necessary and sufficient to render lightness differences discriminable; those in a different feature-family carry information necessary and sufficient to discriminate a straight edge from one that is slightly curved. Now, as argued above (in section IV) if we are ever to get a non-mysterious account of phenomenal consciousness, at some point we will find a family of states and properties that differ in just one way from the normal phenomenal properties, that difference being simply: that no one is aware of them. Suppose that some of the states of preattentive registration of perceptual information subserving discriminabilities of sensible features are the states representing just that family. Then differences they represent are just the differences that, if one were aware of them, would be the differences of which one is aware when one is aware of a phenomenal difference. These states represent phenomenal differences; all they lack is some standing in the competition for selective attention. VII. Preattentive counterparts? -Yes Let us return to the main question -could there be a family of "counterpart" phenomenal properties -and consider some of the virtues of saying "yes" . A "yes" answer to that question allows that there might be candidate properties that differ from normal phenomenal properties in only one respect: that no one is conscious of them. The very same difference of which a subject is aware when that subject is aware of the difference between the appearance of straight and of curved, or between more bright and less bright, can exist even if no one is aware of it. If so, it is possible--conceivable--to analyze a phenomenal property of which one is conscious independently of being conscious of it. This mimics a strategy applied to understanding many perceptual or cognitive capacities. In order to explain how a subject might first doubt that P, and then come to believe that P, we must understand what it is to represent the content that P. We have to be able characterize this representation independently of the doubting or believing of its content. Otherwise we would fail to explain the existence of different propositional attitudes toward the "same" proposition. Likewise, to explain how a subject can both feel the shape of the coffee mug and see the shape of the coffee mug, we have to give some account of the Clark: Preattentive Precursors 17 "shape of the coffee mug" that is independent of the feeling or the seeing of it. If some shape properties are accessible to both modalities, we should be able to characterize the constitution of the properties that make this possible. A similar divide-and-conquer approach applied to awareness of phenomenal similarities and differences would suggest that we first try to understand that of which one is aware, independently of the awareness of it; and then try to understand what instantiates the relation that makes it true that some subject S is aware of some such thing. The guiding assumption is that "phenomenal consciousness" does not simply wink into existence, in one inscrutable singularity. Instead, one first registers the differences that, were one conscious of them, would be phenomenal differences. We get some prior and independent understanding of how the content is registered: of that of which one is about to become conscious. Then to this we add the capacities that suffice sometimes to make one conscious of the differences already registered. So divide-and-conquer works like this: 1. Give an account of that of which one is about to become aware. In the perceptual cases in question, it is a particular manner of sensible appearance. I am being appeared-to as if a square is present. More prosaically, a preattentive state has already registered sensible features as of a square in that vicinity; and one is about to become aware of that which this state represents, nonveridically. 2. Give an account of the process of becoming aware of it. Per hypothesis, this is the process of entering and engaging the competition for selection by selective attention.