Perceptual Information and Dual Reflexive Functionalism

Perceptual Information and Dual Reflexive Functionalism

John Dilworth (Draft only, 1/05)

Cognitive naturalists seek to explain perceptual and other kinds of mental representational states in terms of exclusively causal transactions of an organism with the world. The main kind of previous attempt to do so uses a concept of information that is explained in terms of nomic causal covariation.[1] Cognitive states, on such an account, involve representations of the world that are to be explained in terms of causal intake of such nomic information. To date, any inadequacies in that account have been handled by supplementing nomic information with extra elements such as teleological factors.[2]

The problems in such nomic informational approaches are well known. Just to mention two major failings, the mismatch between always correct nomic information and possibly misrepresenting cognitive states is a fundamental structural flaw in such accounts, while the disjunction problem resulting from the fact that worldly causal states do not always cause the same cognitive response is also endemic in such accounts.[3]

I shall argue that a radical rethinking of naturalist accounts is needed, in order to cope with these fundamental failings, which among other things involves outright abandonment of nomic concepts of information. However, fortunately such an attempt does not need to start from scratch, in that a serviceable fundamental structure for such a radical rethinking is already available in terms of broadly functionalist approaches to cognition and the philosophy of mind--that is, in terms of approaches that seek to explain cognitive states in terms of their overall causal inputs and outputs. Further, that functionalist structure can also be used to diagnose another prime inadequacy of nomic informational views (see below.)

On a genuinely functionalist view of mental representation, any representational state of an organism, such as its perceptual state when it is perceiving object X to have property F, must be explained in terms both of its causal inputs and its causal outputs. From such a functionalist point of view, a fundamental failing of nomic informational views is that they attempt to entirely explain perceptual etc. states in terms only of antecedent causality, or causal inputs, which explanations give no integral role to the subsequent causality of causal outputs from an organism, such as its interactions with the worldly items that caused its perception of them. Or in logical terms, on a functionalist view both the concept of mental representation, and the associated concept of information, must be triadic, or higher, concepts involving transactions between all three items of causal inputs, cognitive states, and causal outputs--possibly involving as well further relations among the functional components of those three broad categories--whereas the concept of nomic information is merely a dyadic concept.

Now if nomic informational views counterfactually actually worked as successful explanations of mental representation, this functionalist criticism would be ineffective, in that any such hypothetical successes would in effect show that functionalism itself was an inadequate and dispensable basis for explaining mental representation. But since nomic informational views are in actuality unsuccessful, the shoe is on the other foot, and arguably now only fully functionalist views of mental representation are promising for further investigation at this stage. Or in terms of broad scientific paradigms, the failed naturalistic nomic information paradigm of mental representation must be replaced by the broadly successful--at least in other areas of the philosophy of mind--functionalist paradigm, even though this paradigm has not yet been adequately applied to mental representation issues, and so requires appropriate extensions.

Fortunately, such a functionalist program of extension and replacement need not abandon intuitively attractive informational concepts entirely, in that on any naturalistic view of information and mental representation the concepts are closely connected. It is just that a functionalist account must explain concepts such as that of antecedent or incoming perceptual or sensory information in terms of its functional role as one relevant factor in mental representation, in place of the failed nomic informational approach, which instead attempted to explain mental representation itself in terms of the nomic information involved in it.

One further functionalist preliminary is as follows. A useful way to express functionalist views, in broadly causal terms, is in terms of a cognitive state S having both characteristic kinds of causes which explain its antecedent functional characteristics, plus of those causes being such that they cause the creation or activation of appropriate kinds of causal dispositions D associated with that cognitive state S, which dispositions are closely associated with the characteristic kinds of subsequent causality associated with the state.

A more general preliminary issue should also be briefly mentioned. The view to be proposed will not assume that perceptual or cognitive states may be identified with, or just are, representational states, and so it is capable of accommodating recent arguments to the effect that much cognition involves skilled interactions with the world rather than representations of it.[4] Nevertheless, in this initial presentation it will be convenient to restrict the discussion to cases in which cognitive and perceptual states do primarily function in a purely representational way. Also, for initial simplicity, only specifically perceptual cognitive states will be considered, though arguably this involves no significant losses in explanatory strength or generality, since a broadly empiricist theory would want to build its account of general cognitive representation on purely perceptual foundations in any case.

After these preliminaries, an initial overview of the approach to be proposed can now be presented. The basic account to be given, in an initial simplified form, will be that an organism Z is in a perceptual state S involving its perceiving of object X as having property F, just in case property F of object X causes Z to acquire or activate some F-related disposition D toward object X, which disposition D is the content of the relevant perceptual state S. Or, in terms of mental representation, the relevant perceptual state S represents that X has property F just in case property F of object X causes Z to acquire or activate some F-related disposition D toward object X.

Thus so far the theory is a reflexive functionalist theory, in that a worldly property causes a perceiver to acquire a disposition directed toward that very property itself. But section 3 will argue that not one but two kinds of content are needed in an adequate theory, so that overall the theory is a dual reflexive functionalist perceptual theory.

1. Misrepresentation and Intentionality Issues

The functionalist account briefly summarized above will now be expanded and explained in more detail. One immediately required expansion of such a reflexive view of perception is as follows. Any adequate view of perception must allow for the possibility both of veridical perceptual representation, and non-veridical misrepresentation. A theoretically convenient way to do so is by initially conceptualizing the determinate property F in the definition under its generic determinable rather than determinate description--for instance, as a color or shape, rather than in terms of its more determinate description as a property of redness or circularity etc. Then perception of, for instance, the color of object X by an organism Z involves the acquisition or activation of some color-related disposition D toward object X, but the determinate description of disposition D might provide evidence either of veridical or non-veridical perception of object X's color--such as if a person Z sees object X, and then puts it in one of a series of colored boxes, whose color either matches or does not match the actual determinate color of object X. Thus if person Z puts object X in the green box, even though X is red, this would provide evidence that Z has misperceived the determinate red color of X as instead being a green color. But if Z instead had put X in the red box, this would give evidence of her having veridically perceived X to be red.

To be sure, all actual properties are determinate properties, so that it is some determinate property such as redness[5] that causes a color-related disposition in the perceiver of it. Also, all color-related dispositions are determinate too, and directed toward the very same object X having the determinate color that caused them. But the utility of introducing determinable descriptions of the relevant determinate properties is to bring out how even a non-veridical misperception or misrepresentation of a determinate property F could still be caused by, and be relevant to, that property F--rather than to some other determinate property G of X, which is instead a determinate kind of shape, mass etc. rather than a determinate kind of color. Thus the current functionalist reflexive theory of perception already has a fundamental advantage over nomic covariation accounts, in that both veridical perceptual representation, and misrepresentation or misperception, can be explained as being equally basic and as having the same structural theoretical analysis. (See section 4 for a complementary account of how some perceptual error is epistemically likely, as well as being structurally possible as here argued.)

A second immediately required expansion or explanation of the reflexive analysis involves the issue of what it is for a disposition D to be directed toward, or related or relevant to, an object X. A potential objection is of course that such an analysis might be circular, or at least not a purely naturalistic one, if it invokes a concept of intentionality, directedness, relatedness or aboutness that cannot be explained in purely causal terms. This issue is a large topic, but for the present two considerations having the potential to defuse such concerns may be cited. First, as in the color example given above, in particular paradigm cases descriptions of a disposition as being 'directed toward' an object X and its properties may be explained in terms of actual behavioral evidence of physical manipulation by organism Z of the object X in a way that is relevant to one of its properties, such as when a person Z puts object X in one color box rather than another. Thus at least in paradigm cases, appropriately caused direct physical interaction with an object X in an F-related way can be used to reductively explain the object-directedness of a disposition D with respect to some related property F.

Second, arguably any disposition, including paradigm purely physical cases, is in a naturalistically harmless sense 'directed toward' those conditions that would activate it.[6] The solubility of salt is a dispositional property that is manifested in the presence of water. Similarly, on the present reflexive view, perception involves dispositions that are, in paradigm cases, manifested in the presence of the relevant perceived objects. In both cases arguably the concept of their being 'directed toward' their manifestation-conditions is a heuristically useful description only, having no additional non-naturalistic ontological implications.[7]

2. Why is a Functionalist Analysis Needed at All?

An important broader, background theoretical issue will now be discussed. Though it is generally accepted that nomic covariation accounts of cognitive information are problematic in some of the major ways already discussed, merely replacing such an account with an alternative reflexive functionalist account would not, in and of itself, give any explanation as to precisely why--from a broadly epistemic point of view--nomic informational approaches are inadequate. For example, consider a hypothetical ideal world in which there always was a nomic relation between properties of worldly objects and properties of raw perceptual data, such as the red color of an object X in relation to some red-related stimulation patterns on the retinas of perceivers of the redness of X, in which world perception consequently was always veridical. In such a world, at least from an epistemic perspective, it seems that a nomic informational account of perceptual representation would be perfectly adequate, in that neither misrepresentation nor disjunction problems could ever arise--horses would always be perceived as horses, and perception of a horse, even on a dark night, would always produce a tokening of 'horse' rather than 'cow'.

In such a world a functionalist account, that also considered the subsequent causality of a perceiver's dispositions with respect to the relevant property F of object X, would seem to be redundant, in that it would have no additional explanatory value in explaining the representational information available to a perceiver, over and above that already provided by the nomic information account. (See section 5 for further discussion of this issue.)

Thus an important issue for present purposes is, given that the actual world is not an epistemically perfect world--hence not an ideal world--exactly what epistemically relevant theoretical resources are added by a functionalist account, which could not also be provided by a nomic information account? For after all, it might be argued, the practice of science often uses ideal models that aren't precisely applicable to the world. Why shouldn't a modified form of a nomic information account, that makes appropriate allowances for worldly imperfections, similarly make any functionalist account redundant? Presumably some such thought might have sustained the efforts of Dretske, Fodor and so on, who, obviously being very familiar with miscellaneous non-teleological varieties of functionalism in the philosophy of mind, could easily have adopted some functionalist alternative to a nomic informational view if it had seemed necessary or preferable.[8]

I shall now sketch two lines of reply for the functionalist to the above arguments. The first has already been mentioned previously, namely that perceptual states may sometimes either not involve information or representational states at all, or involve other additional elements, in either case involving factors of interaction with objects in the world in some way. Thus even if functionalist theories had no additional, strictly representational epistemic value over nomic information accounts, their ability to smoothly theoretically integrate with 'enactive' or interactive views of many perceptual activities would make them overall theoretically preferable in a comprehensive perceptual theory.[9]

The second line of reply returns to the initial theme of this essay, that a radical alternative to the nomic informational account is needed. It will be outlined here, and then detailed in succeeding sections. To begin, I shall claim that nomic informational accounts, far from being roughly or approximately true in the real world, are hopelessly inadequate, in that, for instance, there is an indefinitely large class of different raw retinal stimulations that could be involved in perception of a given property F of an object X under uncontrolled, real-world rather than experimental conditions, so that the actual probability of covariance between the two of them is closer to zero than it is to 1 under normal perceptual conditions. But an actual probability of 1 is required for genuine nomic covariance, as convincingly argued by Dretske.[10]

Hence--and here is where some specifically functionalist factors enter into the analysis--perception must involve skilled inferential processes of a broadly functional kind, the acquisition of which inferential skills would typically involve the setting up of appropriate dispositions toward objects of the relevant perceived kind, which dispositions would be activated during any particular perceptual episode involving an object of that kind. Thus on this view, insofar as perception is representational, i.e., insofar as it involves an informational content with respect to a property F of object X, it is so in virtue of the overall functional interactions of this network of dispositions with raw perceptual data, so that the raw data by themselves cannot explain that informational content, as would be required by alternative nomic information accounts. Thus on this functionalist view, a legitimate concept of perceptual information would be explained in terms of the role of perceptual states in the whole functional process just outlined.

3. The Aspect-Relativity of Perceptual Information

However, there is one further epistemic problem that still needs to be resolved. If, as I claim, raw perceptual data radically underdetermines perceptual content--with a covariance probability closer to zero than one, as claimed above--arguably this would leave functional inferential factors with too much work to do. For if sensory data as such were epistemically almost worthless, then functionalist inferential processes would have to assume almost all of the burden of information acquisition. But epistemically speaking, such a view would be dangerously close to a coherentist rather than a foundationalist account of perceptual information.[11]

Also, it would be unclear, on such a near-zero view of incoming perceptual information, how skilled, as opposed to incompetent or hopelessly bad, inferential abilities could ever be acquired through a trial and error process, because the utility for learning of functionalist interactions with worldly objects would itself require that reliable feedback could be obtained as to the results of such interactions, which results themselves would have to be perceptually obtained via substantive and reliable incoming sensory data, in order for any prior mistakes to be rectified.

This epistemic problem situation calls for a solution that is significantly more radical than just the move to a functionalist rather than nomic covariance view of information, in that, among other things, it seems to have previously been untried, and it also requires the postulation of two kinds rather than just one kind of mental representation or infomational content. The solution may be introduced as follows.

To begin, perhaps one reason as to why nomic covariance views of information have retained their theoretical hold upon many writers, in spite of their problems, is that it clearly is possible to set up experimentally controllable and repeatable conditions under which at least de facto covariance--if not true nomic or lawlike covariance--can be observed. For example, under fixed conditions of lighting and observation, changes in a stimulus property, such as the color of the surface of an object X, will indeed produce reliable covariation in the raw data impinging on the retinas of a perceiver's eyes, in that repetition of the experiment under the same conditions would always produce the same results. Thus, under fixed conditions, a given retinal stimulation will serve to uniquely identify a particular color that it covaries with.

By contrast, my claim that instead only near-zero covariance typically holds under real-world, non-experimental or uncontrolled conditions is quite consistent with this experimental result, in that if contextual or aspectual factors such as lighting conditions, perceiver movement and so on are not controlled for or kept fixed, then their influence would typically swamp or ruin the correlations between object properties and resulting raw retinal data. Thus for example, a real-world situation such as perception of a single color F of an object X under varying conditions of daylight from dawn to dusk would produce a wildly varying range of retinal stimulations produced by that same color F. Thus a nomic covariation account is utterly hopeless as an account of how in fact we are able to perceive that it is the same color under those radically different lighting conditions.

However, a positive lesson may be learned from this failure. The situation just described is one in which the relevant aspectual conditions of lighting etc are changed, with the color being kept fixed. Also, the resulting retinal stimulation would covary with each change in those aspectual conditions, and what is more, it would do so in a reproducible or repeatable way. Hence, the resulting situation is one having a structure that is dual to that of the initial covariation case, in which instead the conditions were fixed but the color was changed. What this structural duality shows is that, just as a given retinal stimulation pattern P would serve to uniquely identify a particular color under known fixed conditions, so also would P serve to uniquely identify a set of aspectual conditions for a given known fixed color.

Applied mathematicians will readily identify the general theoretical situation here as being one in which a particular determinate retinal stimulation pattern P is a function not of just one but instead of two independent variables, namely the actual color of object X and the aspectual conditions under which it is perceived. Two logical features of such functions P=f(A,F) of two variables, with the retinal stimulation pattern P being a function both of the aspectual conditions A and the property F, are of current interest. The first logical feature has already been remarked on, namely that for the retinal stimulation pattern P, if the value of either A or F is already known (or fixed), then the value of the other variable, F or A respectively, may be deduced.

A second logical feature is also important. It is that in general a given retinal stimulation pattern P could have been produced by many different pairs of values of aspectual conditions A and property F. Hence I claim that the primary inferential work to be carried out by the current functionalist account of perception has as its goal the estimate of a particular such pair (Ai, Fi)--selected out of the range of all currently possible pairs that might have caused this same resultant stimulation pattern P--which pair (Ai, Fi) is functionally inferred to be the most likely pair to have caused the actual stimulation pattern P.

In support of this claim, from an epistemic point of view, the introduction or uncovering of two, rather than just one, information-related factors in perception clearly can lead to a marked improvement in the ability of antecedent perceptual causality--sensory stimulation--to provide meaningful and substantively useful information for perceptual purposes. For, though the standard concept of a single kind of nomic information about object-related properties has had to be entirely abandoned, a two-kind, non-nomic functional replacement is now available, because a given retinal stimulation pattern P can provide information about a range of correlative aspectual versus object-related property pairs associated with the relevant perceived object X, any pair of which could have caused P.

But, this being so, there finally is available some genuine and characteristic work for a functionalist process of inference to carry out, which is not merely a potentially unnecessary addendum to a standard nomic informational account--namely, to estimate which pair (Ai, Fi), out of all of the possible aspectual plus object-related pairs, is most likely to have been the cause of P. Thus, to state the main theoretical innovation of the current account more explicitly, on the present functionalist view the single standard concept of nomic information has been completely replaced by two mutually correlative concepts of aspectual versus object-related, non-nomic, functionally provided information. And this theoretical move is required, because the raw retinal--and in general, sensory--data available to a perceiver provides no genuinely nomic information whatsoever that is specifically about the relevant property F of object X.[12]

To be sure, a vestigial concept of nomic information could be preserved in our overall theoretical structure, by arguing that the given stimulation P necessarily covaries with the whole range {(A1,F1), (A2,F2), ...(An,Fn)} of possible pairs of aspectual versus object-related values, in that changes in P and changes in the range of relevant pairs would presumably be mutually co-determinative. But, though a cognitive system must be able to process data concerning all of these items taken individually, its functionally based estimate of which pair (Ai, Fi) is most likely to have caused the sensorily received stimulation P involves no nomic information of any kind concerning any of the relevant available pairs, or with respect to which pair should be chosen, or concerning the particular correlative aspectual versus object-related components in whichever pair is functionally chosen.

Thus, in a broader perspective, the new picture is of a perceptual cognitive system, not as a processor of sensorily received nomic information, but instead as an inferentially-based estimator--in virtue of its functional structure--of likely causes of items of raw, non-nomic sensory data.

4. The Informational Indeterminacy of All Raw Perceptual Data

A concrete example will be useful at this point. Consider an object feature F such

as the shape of an object, e.g., the circularity of a flat circular disk X. My claim is that there is no nomic relation between that shape F and raw retinal image patterns P of the shape, because the shape-related retinal data P will vary, depending on various aspectual conditions, including the angle at which a perceiver views object X--so that, in particular, most retinal images of such a disk will be elliptical in various ways, not circular, for someone viewing object X from various angles.

Now an initial, but potentially misleading, statement of the current functional view is that any epistemically adequate perceptual processing of that raw information P must involve breaking it down into two correlative parts--of aspectual versus object-related data--where the primary aspectual data is data about the appearance of the disc X as viewed from different angles, while the object-related data is data about the actual circular shape of X. However, that description could potentially be co-opted by a traditional nomic informational view, which could happily accept that sometimes two independent kinds of nomic information become intermixed, hence requiring a decomposition into its nomic components.

The crucial issue, in the present case, between a traditional nomic view and the current functional view is as to whether a unique decomposition is always possible in such cases. To be successful, a nomic view would have to hold that all such cases permit a unique decomposition into what are, on the nomic view, completely independent items of information. But the functional view instead claims that there are an indefinite number of different possible aspectual versus object-related decompositions of that retinal data, each of which is consistent with that data as such, but each of which would involve putative information about, or putative representations of, different correlative aspectual versus object-related properties of object X. Thus the nomic theorist must assume that the retinal data P already, in and of itself, supplies completely determinate information of both kinds which would permit a unique decomposition, whereas the functionalist denies that the retinal data P supplies completely determinate information of any kind.

For example, a given raw retinal elliptical shape P might be decomposed into an aspectual representation of a perpendicular aspectual view of X, with X being represented as itself having exactly that same elliptical shape, or as an oblique aspectual view of X, with X being represented as having a circular shape, and so on through an indefinite range of possible representational pairs that are consistent with the raw data P. Or in other words, on the functionalist view the raw retinal data P as such is significantly representationally or informationally indeterminate, and in need of further inferential processing. The retinal data P only acquires a determinate, dual aspectual versus object-related representational interpretation, and hence a fully determinate status as a source of two correlative kinds of informational content, in virtue of its functional role in activating shape-related dispositions toward X within the person Z who is perceiving the shape of object X.

In support of this functionalist view, which admittedly might initially sound somewhat radical in its rejection of nomic informational views, the everyday concept of 'limited information' may be appealed to. It is a familiar idea that someone may have only limited perceptual access to a situation, and so have to infer, on the basis of the limited perceptual information available to her, what the actual properties of some object or objects are, and possibly make some mistakes in doing so. The current functionalist claim is simply that all perceptual situations similarly involve only limited, incomplete or partially indeterminate raw sensory data, which requires functional interpretation to become fully determinate, and which inferential process may well be fallible in that, in some cases, the actual functional role of the relevant raw data items may involve the activation of incorrect or inappropriate property-related dispositions toward the relevant objects.

The current connection between the indeterminacy and fallibility of raw perceptual data is also important in filling out the account of perceptual misrepresentation given in section 1. That previous section showed how the basic functionalist account being presented provides the logical resources for perceptual misrepresentation to be possible. But the current point provides some substantive epistemic backing as well, in that it shows why some misrepresentation is likely to occur, in addition to its being structurally possible in a functionalist account, because of the indeterminacy of raw perceptual data, and the fallibility of inferential choice of a determinate double-content resolution of that indeterminacy.

The current functionalist view is also strongly supportive of well-known 'poverty of the stimulus' claims concerning perception,[13] according to which perception must involve inferential elements because, for instance, perception of an item as a book goes beyond the low-level, purely perceptual contents of shapes and colors. But the functionalist view is more comprehensive in its claims concerning the impoverishment of the stimuli, in that it claims that even the lowest-level kinds of perceptual data, including simple shapes and colors, are similarly indeterminate prior to inferential processing.

A further example involving colors may be useful at this point. In the case of color, we must distinguish the intrinsic color C of a given surface of an object X from the aspectual color A of the light impinging on its surface. The color information in a raw retinal stimulation P, resulting from the reflection or refraction of aspectual light A from

colored surface C of X, is physically compatible with a range of different combinations of aspectual light A and intrinsic color C of the object surface. For example, a yellowish retinal stimulation P might be the result of white light A1 impinging on a yellow surface C1, or yellow light A2 impinging on a white surface C2, or any other physically possible combinations--possibly involving other colors as well--that would result in the same retinal stimulation P.

Thus both low-level retinal shapes and low-level retinal colors provide only indeterminate or impoverished retinal data, requiring functional interpretation. But since arguably all retinal stimulation consists of nothing more that various combinations of shapes and colors, the functionalist case against any raw perceptual information involving truly nomic informational elements is complete--all of them are intrinsically indeterminate, and such that they acquire a determinate content only as part of their fallible functional role in activating a network of related perceptual dispositions in the perceiver.

5. The Perceiver-Relativity of Functionalist Information

It was convenient, in introducing the current functionalist and non-nomic concepts of aspectual plus object-related information, to write as if the concept of nomic information itself was a respectable or acceptable one, with its only problem being that it is never actually applicable to cases of raw perceptual data patterns P, because such raw data patterns are never in fact nomically related to the distal properties of worldly objects which a perceiver is attempting to perceive. However, the kind of reflexive functionalism being argued for here makes the concept of perceptual information or content essentially a perceiver-relative one, in that a perceptual state S of a perceiver Z counts as representing or providing information about object X having property F just in case property F caused perceiver Z to acquire or activate F-related dispositions with respect to object X. But a perceiver Z might or might not be in such an informational state, whether or not her raw perceptual data pattern P is nomically related to property F of object X. Or in other words, on the present conception of information, the holding or not holding of nomic connections between items--whether purely in the external world, or linking a perceiver's sensory state to that world--is completely irrelevant to issues of perceptual information or representation.

Consider, for instance, the issue of transduction, or nomic conversion.[14] A transducer is, roughly speaking, a device which converts one form of energy into another form in a lawlike way. Now it seems likely that there is indeed a nomic relation between a beam of light photons impinging on the eye, and the resulting pattern of sensory stimulations in the retina, which relation, on the standard nomic informational view, would be a paradigm case of a nomic informational relation. However, it is intuitively obvious that we never do perceive photons of light impinging on the eye, nor acquire information about them on the basis of such cases of sensory stimulation.

The current functionalist view has a ready explanation for this failure of a nomic connection to be relevant to perceptual information, namely that, though the photons were one of the causal links in the chain of antecedent causes that produced a perceptual state, that state did not result in any dispositions being acquired or activated toward those photons of light themselves. Hence no perceiver-related information about them, or representation of them, was generated as part of the perceiver's current acquired or activated perceptual dispositional state.

To generalize this point, the reflexive functionalist explanation of information has the significant advantage that it can explain why only the desired property F of object X is perceived, rather than arbitrary other intervening items in the causal chain leading from property F to the perceiver's sensory stimulation pattern P--again, it is because only specific F-related dispositions were acquired or activated, rather than other dispositions which would have instead been relevant to such intervening items.

In view of the above distinctions, it will be useful to briefly revisit the section 2

issue of whether, in an epistemically perfect world, with completely nomic connections between properties F of objects and sensory stimulations P, a nomic informational theory would be just as adequate, and simpler, than a corresponding functionalist view. Potentially there is still a fundamental logical difference between the informational theories, in that on the functionalist theory there are logically possible worlds in which raw data P is nomically related to worldly property F, but nevertheless the perceiver Z acquires no F-related dispositions and hence does not qualify as perceiving property F, in spite of that nomic relation.

However, if a broadly naturalistic evolutionary theory is assumed, for possible worlds having laws similar to those of our own world, the evolutionary advantages of epistemically exploiting the assumed nomic relation between property F and raw data P would presumably ensure that sufficiently evolved perceivers Z would normally tend to acquire F-related dispositions whenever the relevant nomic connections held, so that the pre-theoretic intuition that the two kinds of informational theory would have roughly equivalent consequences in a nomically ideal world which is otherwise relevantly similar to our own can be upheld. Nevertheless, it should now be clear that the apparent parallelism of the two theories under such conditions is explicable more by evolutionary convergence plus epistemic utility than by any substantive theoretical similarities between them.

6. The Functional Role of Aspectual Information

More will now be said about aspectual information, i.e., of the functional role of aspect or aspectual representation in the present account. Recall that the concept was introduced because of the de facto radical failure of nomic relations to hold between raw perceptual data patterns P and relevant properties F of worldly objects X. But in spite of that failure, it turns out that at least a reliable, if not a genuinely nomic, relation does hold between particular raw data patterns P and a set of correlated aspectual versus object-related property pairs (Ai, Fi), so that non-nomic information can be extracted from raw data P if a cognitive system can inferentially estimate at least one of the factors Ai and Fi, since if both P and one of Ai and Fi is known then the other can be deduced.

The ontological basis of the epistemic points just made is that strictly speaking, a property F of a worldly object X cannot by itself cause perception of itself by a perceiver, because of the lack of a nomic relation between F and raw data P. However, this does not prevent the combined powers of both particular aspectual factors A and that property F from causing simultaneous perception of both A and F, when the joint causal factors provided by both A and F appropriately interact with broadly inferential dispositional structures in the relevant perceiver Z, so that she thereby acquires or activates F-related dispositions. For example, it is not the redness of an object surface alone or as such that causes perception of itself, but rather that redness as causally transmitted to a perceiver's eyes by the aspectual ambient light that impinges on the red surface, and which is, after being appropriately changed by the red surface, refracted by it into the perceiver's eyes. And a similar account could be provided for shape information also--shapes as such are causally inert, but a shape under particular aspectual conditions of lighting and perspective can cause perception both of itself and of the relevant aspectual conditions, in a perceiver having appropriately prepared dispositional structures.

Thus ontologically, there is of course only a single aspectual plus object-related property pair that causes perception of itself, even though epistemically, a perceiver must use her previously trained dispositional structure to inferentially estimate which is the most likely aspect-object pair (Ai, Fi), from the set of all possible such pairs, to be the cause of her current raw perceptual data P. And, as noted in section 1, the inferential process is potentially a fallible one, in that, for instance, raw yellowish retinal light stimulations might be the result of white light refracted from a yellow surface, or yellow light refracted from a white surface, and so on, and the perceiver might easily guess wrong as to which is the correct actual aspect-object pair, and thereby perceptually represent both aspect and property incorrectly via incorrect or inappropriate dispositions toward that actual pair.

At this point a likely objection to the introduction of aspectual factors should be considered. If both ontological aspectual factors, and perceptual representation of them, are as central to perception as I have claimed, how can one account for their almost universal neglect, both in previous epistemic and philosophical theories of perception and cognitive content, and in ordinary perceptual experience? As a broad initial answer, many aspectual factors such as lighting factors remain relatively constant, e.g. the roughly sixteen-hour cycle of daylight changes occur too slowly to be consciously perceived as such by a perceptual system that is functionally geared to giving epistemic priority to rapidly changing factors. Also, it is probably true, as an overall generalization, that the main function of perception is to extract information about the properties of worldly objects. Hence, though aspectual information must, on the present account, play an ineliminable role in any obtaining of that object-related information, aspectual information as such functions primarily only as a means to that object-related end, and so it typically is of little intrinsic interest to a perceiver.

Another factor may be that cognitive guesses as to the correct solution of the equation P=f(Ai, Fi), which require one aspectual or object-related factor to be guessed so that the other can be calculated, may tend to initially involve aspectual estimates rather than object-related estimates, perhaps because, again, many aspectual conditions such as lighting conditions remain relatively constant, and so are easier to guess or to have prior knowledge of. Thus often, aspectual factors are primarily part of the intermediate estimating processing leading to the desired object-related solution, rather than providing perceptual information that is sought for its own sake.

Nevertheless, some noticeably or quickly changing aspectual factors arguably are a familiar feature of normal conscious perception, such as the varying appearance of objects when seen from changing points of view as one walks by or near to them. In such cases, one simultaneously sees both the current aspectual appearance of the relevant object X, and the object X itself--one's perceptual system represents both the aspectual and object-related features. However, though this fact is obvious once pointed out, it is such a familiar feature of experience, and we become so expert at perceiving and learning about objects via such perceptual experiences of aspects of them, that the double representational content--involving both aspectual and object-related content--present in such experiences is hardly noticed as such, until attention is drawn to it, because of its omnipresence in normal perception.

Finally, a further likely theoretical reason for the neglect of aspects in perception is that of course in a rigorous, purely ontological description of the universe, aspects would not appear as a separate ontological category distinct from that of objects. Typical aspectual factors involve items such as light radiation, which enables color and shape information to be transmitted to perceivers, and perspectival factors concerning the geometry of objects relative to particular viewpoints in space, plus other relational factors such as the relative size, brightness etc. of a given object in relation to other objects located around it. But ontologically speaking, such factors make up a highly miscellaneous category having no theoretical unity.

More importantly, arguably the aspect versus object distinction is not, strictly speaking, an ontological distinction at all, but instead it is a perceiver-related, broadly functional or purpose-related distinction that classifies items of the universe in ways that are relative to our current perceptual interests and projects. Further, those interests are in addition arguably strictly temporary ones, in that ontologically speaking the very same physical items--such as fields of directed light radiation--may at one moment function to provide aspectual information to a perceiver insofar as they illuminate objects, and at a later moment function as 'objects' in their own right when their sources--such as a fluorescent panel--are directly observed. Thus it is not surprising that aspectual matters have been generally neglected in philosophically oriented perceptual theories, in that they are fugitive and temporary even within the study of perception itself. Nevertheless, the aspectual versus object-related distinction remains an indispensable one for the broadly epistemic and semantic purposes of the study of perceptual information processing and representational content, as has been argued here.

7. The Possible Dispositional Relevance of Aspects

The previous section acknowledged the functional indispensability of aspects, while nevertheless explaining their theoretic and folk-psychological neglect as being due to their often having supporting, rather than center-stage, roles in the overall purposes or functions of perception. For example, one's perceptual system might use an initial estimate of the aspectual light level Ai to calculate the likely color Fi of an object X in solving the equation P=f(A, F), for incoming raw visual data P, with the primary purpose of the perception being to acquire information about that color Fi rather than about the light level Ai.

However, it seems likely that if there are possible additional functions that could be carried out by aspectual content, which would contribute to biological fitness by providing processing efficiencies that would otherwise be unavailable, then evolutionary forces would have led to their implementation in currently surviving species. Also, in terms of an intuitive conception of prior probabilities or likelihood, given the great utility of object-related information in perception, it seems likely that aspect-related information could also be of significant utility in its own right, as well as merely serving as a necessary preliminary to the identification of object-related information.

One speculative suggestion along such lines is the following. There may be some complex properties of creatures such as biological predators, which properties are functionally of great importance to perceivers of them who are potential prey, and which would involve both aspectual as well as object-related features of an object. For example, it seems likely that dispositions to flee from a predator, or to approach closer to food, are perceptually prompted, not simply by perception that an object X is a predator, or food, but by more complex, possibly aspectually involving properties such as by the closeness of a predator, or by the ready availability of food, when the aspectual situation is that the food is on the ground rather than in a tree, or in a prone rather than an upright position so that it is more vulnerable. Or in other words, aspectual content conditions might themselves play a directly relevant causal role in the formation or activation of dispositions to flee or to feed, rather than just being dispensable preliminaries in the initial identification of a predator or food as such.

To be sure, those aspectual properties could, under other perceptual conditions or perceiver purposes, be functionally relevant directly as object-related rather than aspectual conditions, such as if a creature were to perceptually scan its neighborhood to find out which objects are close to it, in which case the relational property of closeness to the perceiver is the main focus of attention and current perceptual activity. But the current suggestion is that evolutionary pressures would confer advantages on species that could simultaneously use both aspectual and object-related information for effective dispositional perceptual reactions, without requiring such additional shifts in attention or purpose.

More broadly, such possibilities show that the current dual reflexive functionalist theory has the potential to provide an extremely nuanced and sophisticated account of perceptual activity, which closely integrates two functionally indispensable kinds of non-nomic perceptual content with a dispositional view that allows those kinds of content to be causally effective, yet which also allows for the broader possibility of more or less representation-free dispositional perceptual states--as briefly suggested in the introduction--that are closely integrated with worldly exploration and interaction, as defended by current 'enactive' perceptual views.[15]

Western Michigan University
Notes