On the Elusive but Vital Difference Between Privileged and Optimal Viewpoints

Abstract

I argue that two theses, which get conflated tacitly but frequently in both the philosophical and the scientific literature on perception, must be distinguished. The first is that there are optimal viewpoints, viewpoints from which an object’s shape is more readily discernable than from others. The second is that there are privileged viewpoints, viewpoints that alone secure the veridicality of perception. I claim that phenomenology establishes the ubiquitousness of optimal viewpoints, but that the notion of privileged viewpoints is indefensible. It emerges when the empirical investigation of the mechanism of perception, and specifically of the role of retinal images, becomes the basis for the phenomenology of perception. Both the notion of a privileged viewpoint and the models it serves, such as the two-step model, are, I argue, untenable. To emphasize: the claims are phenomenological, not empirical, and so cannot be confirmed or refuted by empirical evidence. Optimal viewpoints are further explored by critically examining Husserl’s notion of a “sum of optima” and assessing it in the context of his claim that normal viewpoints are optimal. The paper ends with some thoughts on what the relationship between the science and the phenomenology of vision ought to be.

1. Introduction

A prevalent paradigm in empirical research, and specifically in research of shape perception, is Bayesianism. I use this popular framework to discuss some conceptual issues which, I think, pertain to the study of perception in general. On this approach, given a retinal image, the visual system calculates the probabilities that the retinal image is the product of a projection of a certain shape for all the shapes that can be the source of the image, and after integrating other factors—distance, angle, light, etc.—generates an experience of the shape that scored highest. The reason such an apparatus is required is due to the ambiguity afflicting retinal images: one and the same shape, e.g., a trapezoid, can be registered on the retina as numerous quadrilaterals, depending on the angle from which it is viewed, and conversely, numerous quadrilaterals can yield the same retinal image, again, if the angle and distance from which they are viewed are chosen appropriately (Figure 1). Hence, the visual system has to “guess” which of the many possible shapes before it is the actual one. If the system guesses correctly then, assuming one is looking at, for example, a coin, an elliptic projection on the retina will lead to an experience of a circular object1.

But here’s the rub: for circularity to be chosen as the most probable shape from the many that can produce the same ellipse, the system needs to possess some knowledge of circularity (as well as of other, under the circumstances perhaps less likely, alternatives, such as ellipses). The question immediately arises: how does the system get this knowledge? Without it, a Bayesian calculation cannot lead to an experience of circularity. Therefore, this information is crucial. But what is its origin?

The discussion starts in the next section with a brief exposition of two standard responses to this question—nativism and past experience. The following Section 3 argues that appeal to past experience implies the existence of privileged viewpoints, privileged in that they, and they alone, ground veridicality. This notion is then shown to be untenable. Section 4 contrasts it with that of optimal viewpoints. The contrast between privileged and optimal viewpoints is further examined in Section 5 by discussing two notions of optimality that figure in Husserl’s work. The final Section 6 offers some thoughts on the relationship between the science of vision and phenomenology.

2. Two Responses to the Missing Information Problem

This problem has been discussed before. Fodor [1] (p. 2) presents the Poverty of the Stimulus Argument: “[a] Poverty of The Stimulus Argument alleges that there is typically more information in a perceptual response than there is in the proximal stimulus that prompts the response; hence perceptual integration must somehow involve the contribution of information by the perceiving organism”2. Marr [3] (p. 265-6) is also troubled by the issue: “In each case the surface structure is strictly underdetermined from the information in images alone, and the secret of formulating the processes accurately lies in discovering precisely what additional information can safely be assumed about the world that provides powerful enough constraints for the process to run”. The problem is obvious: the retinal image alone does not contain all the information required for producing the appropriate perceptual experience. The “perceiving organism” has to provide further data.

One approach to this challenge is nativism. Put briefly, according to the nativist, shapes, and with them the algorithms that transform them into their appearances from different perspectives (or vice versa), are innate. I confess that I do not find the theory that when looking at, e.g., a coin, an innate concept of circularity pops up and guides the visual system to the correct experience plausible. But even if it were, it does not address the problem at hand: how is this mechanism supposed to know which of the many innate shapes at its disposal is the right one for the occasion in question? Say one is looking at a coin. There are, in addition to round coins, elliptic coins of varying forms of ellipticity. The perceiving organism is supposed to select from its stock of shapes the ellipse matching the coin’s actual shape. But which? So, the original problem remains—there is not enough data for performing the correction from a distorted retinal image to a veridical experience of the object’s shape.

Here is what is probably the most common answer to this challenge (and specifically for Bayesianism): past experience. We have handled and seen coins before, and we have done so from a multitude of viewpoints. This rich past experience has endowed us with much knowledge and, when fresh visual data, such as the current stimulation of the retina, enters the visual system, the new data is integrated with the knowledge that is stored in the system to yield the experience of roundness. The elliptic projection, together with what we already know about the shapes of coins, integrate to yield the most suitable representation for that particular occasion.

However, two conceptual difficulties remain. As with nativism, given that in the actual world there are coins of many shapes, the question of which past experience should be invoked on a given occasion stands. But there is a further, more profound, difficulty: how is a veridical past-experience, which supposedly provides the missing data for the current experience, so much as possible? Supposedly, for an experience of shape to be veridical, retinal images have to be augmented with further information, the source of which is past experience. But if so, we are caught in an infinite regress. Every veridical experience presupposes a former veridical experience. One’s biography, all the way back, consists of a series of distorted projections on one’s retina, and distortions, numerous as they may be, do not combine to something veridical. The only way this regress is not infinite is if there is a first experience that does not rely on prior knowledge but is nevertheless veridical, in other words, an experience in which the retinal image alone secures veridicality.

3. Privileged as Foundational

We thus arrive at the notion of a privileged viewpoint. Much past experience indeed involved distorted projections whose shapes were different from the shape of the object being perceived. However, supposedly, there were occasions in which, for example, round things were viewed frontally, that is, from a viewpoint from which the thing’s actual shape and the shape of the projection on the retina coincide. These are unique experiences in which there are no distortions, and no corrections to make. One exposure of this kind and the subject has in its possession that in reference to which she can later make the corrections that are required when the vantage point provides only a distorted image. There is a vast amount of empirical research on canonical viewpoints. And there is no dispute that some viewpoints on objects are superior, in a sense to be specified, to others. I will return to this fact later. But here I must emphasize that the proposal being considered is not that a certain viewpoint, e.g., the frontal viewpoint, is better than others. It is that it has a foundational role as the grounding for other perceptual experiences. It is qualitatively distinguished in that it does not require prior information. It is the foundation that supports the rest but is itself unsupported.

Unfortunately, the idea of canonical views as foundational runs into serious problems as well. The main difficulty is that the perceiver does not know, and cannot know, not even tacitly, that a given perspective is foundational. For the frontal viewpoint on a coin to answer the missing information problem it needs to be recognized as privileged. But for this recognition to occur the perceiver already needs to know what the real shape of the coin is. In short, to know what something looks like we need to know what the privileged vantage point for viewing it is, but to find a privileged vantage point we already need to know the shape we are looking at.

Take a photo of a coin. Call the shape in the photo S, and the actual shape of the coin A. If the angle from which the coin is photographed is 90° (with respect to the plane the coin is on) then S = A. Otherwise S ≠ A. Supposedly, this gives a precise sense in which 90° is a privileged viewpoint. Here is a vantage point from which the apparent and the actual shape coincide—no distortions, no corrections to make. This experience then becomes the additional information that is tapped into when the coin is seen from other, “distortive” angles, providing the circle to which ellipses, seen from other perspectives, are corrected. The problem is, of course, that S itself does not convey the angle from which the object’s shape was captured. For any elliptical object O there is an angle Z such that a photo of O from Z would yield a circle, that is, a projection identical in shape to that of a coin photographed frontally. For the recovery of the actual shape additional information is required—Z. This information is not included in the photograph, or in the image on the retina. Again, a privileged viewpoint is needed to fill in missing information, but this information is required to know that the given viewpoint is privileged.

This conundrum is specifically troublesome for Bayesian accounts of perception. A key ingredient in the Bayesian selection of the highest probability option is “the prior probability of the scene description” which includes “variables representing surface shape, material reflectivity, illumination direction, and viewpoint” [4] (p. 4). Given the ambiguity resulting from there being a multitude of shapes compatible with a given retinal image, information regarding the angle in which the shape is tilted with respect to the retina and its distance from the retina is essential for veridical perception. But how can this information become available to the perceiver? Like shape perception, distance estimation is also being vigorously researched. Numerous parameters are thought to be relevant: blur, accommodation cues, vergence and disparity, head orientation, etc.3. However, the basic conceptual predicament afflicting the notion of a privileged viewpoint recurs here. The information that is recorded by the system regarding these parameters is not unequivocal; it is compatible with a variety of circumstances, and the system is tasked with establishing correct values from multivariate data, it is again asked to ascertain actual values from distorted appearances. The general problem is this: the scene needs to be taken in correctly for accurate information regarding reflectivity, illumination, distance, angle, etc., to become available to the perceptual mechanism, but this information is required for the scene to be perceived correctly. This seems conceptually disastrous4.

Warren [7] (p. 353-4) raises a similar worry in the course of his critical examination of the two-step model: “A standard indirect account is that, with experience, we learn to associate the real shape of the table with the various projected forms of which we are immediately aware. Perception is thus a two-step process, in which we first see the apparent shape (the projected form) and then infer the real shape based on prior knowledge. But how do we acquire these associations, without having perceptual access to the real shape?”5. Indeed, the difficulties presented above pose a problem for any theory (not just Bayesianism) in which retinal images, i.e., projections, are regarded as the fundamental input with which the process of perception commences. In the last section of the paper, I discuss why realizing this presents us with the need to clearly distinguish phenomenological studies of perception from empirical research into the mechanism of perception. As I argue elsewhere [9], projections have no place in the phenomenology of perception, but the phenomenology of perception must be part of the background against which the role of projections in the mechanism of vision is investigated.

4. Privileged vs. Optimal

In practically all everyday situations, we correctly recognize shapes, and we do so from a multitude of distances and angles. We readily and correctly see that the dozens of tiles that make up the floor are all identical. But imagine that, without a change in their location with respect to us, and without first seeing them all together forming a floor, we were shown these tiles one at a time. The shape of those under our nose would be recognized immediately. But, with respect to the more distant ones we may become unsure, and the farther away they are the more likely mistakes will become. Even if they are relatively rare, some viewpoints are disadvantaged.

This qualitative difference can easily be quantified, e.g., by attaching to each viewpoint (distance and angle) a number indicating the measurable likelihood that a normal observer, viewing the shape in question under normal conditions, will misrecognize it6. There will necessarily be viewpoints that will be identified as least likely to generate a misperception of shape. However, this fact must not be taken as establishing the existence of privileged viewpoints. We must distinguish here between privileged and optimal viewpoints. Optimal viewpoints are those that practically guarantee correct identification of shape or that provide the best possibility for such identification. Privileged viewpoints are those which have a foundational role as discussed above—they offer veridical experiences without tapping into further data, and are, furthermore, the source of the additional data required for veridicality when the viewpoint is not privileged. Viewpoints from which misrecognition of shape is extremely unlikely are not privileged but optimal. And they abound—viewpoints that are less than optimal, for example in the case of looking at tiles making up a floor, are the exception.

Viewpoints come to be regarded as better than others, as “canonical”, in the course of experience. It is by viewing a coin from many angles that we learn that the frontal view is superior to one that is almost sideways on. We also learn that many oblique viewpoints are just as good as the frontal one, and that at some angle there is a shift from slanted viewpoints that still afford veridical perception to those which are unreliable7. All these observations presuppose perceptual constancy: it is because numerous viewpoints show the coin’s actual shape that they can be compared with each other for quality, and that it can be established that some viewpoints are not reliable (more on perceptual constancy soon).

Optimal viewpoints then are those that are the farthest from the (minority of) viewpoints which actually do not enable the perceiver to correctly discern an object’s shape. For flat objects, these include but are not restricted to the frontal viewpoint. Again, it is the fact that from many other viewpoints a coin is just as readily seen to be round that makes it possible to comparatively assess the quality of viewpoints and mark some as better than others, and perhaps even as optimal in the sense that, unless something goes wrong, when an object is viewed from them mistakes simply do not occur. Optimal viewpoints are actual and ubiquitous, in contrast with privileged viewpoints, which are a myth.

5. Husserl’s Two Notions of Optimality

Here again is the reflection on the mechanics of perception that underpins this myth (as well as current vision science and scientifically inspired philosophical approaches to perception, such as the aforementioned two-step model). When making a realistic drawing of the rim of a mug, that is, of the rim as we see it, we plot an ellipse (Figure 2). This fact is taken as evidence that the rim looks elliptic—we have drawn what we see, have we not? And from this observation it is just a small step to the conclusion that perception is mediated by projections, and to the further conclusion that some viewpoints are privileged in that from them the projection is identical to the actual shape being projected, a fact in light of which these viewpoints, and they alone, reveal the object’s actual shape. In all other cases the actual shape must be recovered from a skewed projection8.

A different train of thought, that on the face of it seems to give rise to a somewhat related worry, comes not from vision science but from phenomenology. It is this: no matter how many times we have observed a given object, from how many angles and distances we have seen it, or under how many lighting conditions, there is always more to see, there are always new viewpoints from which it can be looked at, and these new, hitherto unoccupied perspectives expose the object in novel ways, adding visual information that was previously unavailable. Hence, no matter how thoroughly an object has been studied visually, visual acquaintance with it is incomplete and partial. There is always more to learn about it. The inevitable incompleteness of any actual acquaintance with an object is dramatically accentuated when the options of studying it under a magnifying glass, or UV light, or dissecting it so as to see its insides, etc., are also taken into account.

However, there is a significant difference between the contention that normal visual input is skewed and deformed, and the fact that, in some sense, normal perception is always partial. Partial does not mean non-veridical, nor does it entail that perception is mediated, e.g., via retinal projections. But the two worries do have in common the implication that, alongside normal viewpoints, there are also “special viewpoints”. Deeming perception partial immediately hints at there being, if only ideally, a viewpoint that is not partial. And so, although phenomenological considerations do not introduce the untenable notion of a privileged-as-foundational viewpoint, nevertheless they too suggest an intriguing distinction between two kinds of viewpoints, actual and ideal.

To examine it, we turn to consider two notions of optimality found in Husserl9. But first—a disclaimer: there are much-discussed tensions between claims Husserl makes in his different works, for example, between claims found in his early The Philosophy of Arithmetic and those in his Logical Investigations, or between those of the Investigations and criticisms of them elaborated in Ideas Pertaining to a Pure Phenomenology and to a Phenomenological Philosophy. I do not presume to engage in this interpretative effort. I discuss Husserl because ideas and concepts that he developed and explored can help clarify the theses of this paper, even if the way they are used here is not always faithful to his intentions at this or that stage of his lifelong preoccupation with perception. Specifically, even though it is suggested below that Husserl identified optimality with “the view from nowhere” and perhaps also with “the view from everywhere” (as he seems to have done in the Logical Investigations), it is debatable whether that is the case. The focus here, however, is on employing his penetrating insights as part of the analysis of the distinction between optimal and privileged viewpoints, rather than on advocating for a particular reading of his texts.

On to Husserl’s two notions. The first is very similar to the idea of optimality defended above. According to Husserl, regular, normal viewpoints are optimal, in that they leave nothing wanting insofar as our interest in the object on the given occasion in which it is being perceived is concerned. If my colleague tells me he left me coffee on my desk, when I return to my desk and find it there, my perception of it is optimal—there is nothing missing from it, nothing lacking. Since there is nothing to improve, it is proper to call perception under such circumstances optimal—when something cannot get better, it has reached its optimum. For Husserl perception is an intentional act the content of which is the intentional object, an object characterized by its significance or sense. When these are fully revealed in perception, the perception is optimal. The mundane, ordinary act of perceiving a coffee cup on one’s desk is a case in point.

Optimality, in this usage, is highly context dependent. For the same object, different conditions may count as optimal viewpoints on it, depending on the nature of one’s interest in it. A piece of rock will be looked at differently by a geologist, an archeologist, a sculptor, a photographer, a hiker, etc. One will use a magnifying glass to study it or a drill to inspect its insides, another a chemical substance to analyze it, the sculptor will scrutinize it from one perspective, the photographer from another. But all will see the rock fully10, as it is, and attain a visual grasp of it that will entirely satisfy their needs and interests. Normal perception is optimal in that it shows the perceived object completely, or even perfectly.

Optimality, completeness, and perfection are normatively charged terms. Indeed, optimality, as it is being discussed here, is tightly related to the normativity which for Husserl is constitutive of perception11. That they are normatively laden terms means that there is no such thing as “complete”, or “perfect”, simpliciter. Titling something as “perfect” is always done with some end, some purpose, in mind, it is always in relation to a goal, an interest. When I say that my perception of the coffee mug on my table is optimal, what this means is that, in relation to any interest I may have in this object on the given occasion, my perception of it is as adequate, as satisfying, as it can be. There is no thinkable manner, or need, to improve it. This, again, is very much akin to the claims regarding optimality made in the previous section.

However, in addition to this kind of optimality, Husserl invokes another kind of optimality, ideal optimality. As noted, one’s visual experience, restricted as it necessarily is to a finite set of viewpoints and therefore to an infinitesimal portion of all available viewpoints, must, so it seems, be partial, very partial. Normal perspectives, even if they are optimal in a practical sense, still do not offer the object in its fullness, in its “complete givenness” [17] (§143 p. 299). For that, something more is required. If the problem with normal perception is that it leaves out what would be seen from other viewpoints, the correction must be perception which comprises all viewpoints. Indeed, that is what Husserl proposes: a “viewpoint” that results from the integration of all possible viewpoints12. Since, supposedly, only such a viewpoint leaves nothing out, only it could reveal the object as it really is, the object thought of apart from any particular significance or sense it may have, apart from any manner of appearing, and specifically, independently of any perspective. It is of the intended object simpliciter, as opposed to an intentional object, to a noema, and it is conceived of not in the context of normal vision but in correlation to the integral of all viewpoints13. As Doyon comments, the latter pertains to “an ideal unity, the noematic correlate of this infinite synthetic process oriented toward the idea of perfection” [12] (p. 173).

Ideal perception is obviously not one that can actually be achieved, certainly not by a “limited finite consciousness” [17] (§143 p. 299). Its object is given “as a connexion of endless processes of continuous appearing… a continuum of appearances” [17] (§143 p. 299). And the question immediately presents itself: in what sense is this ideal optimality, optimality of perception? To discuss it, we cross over briefly to analytic territory, and consider Dummett’s characterization of what he calls “a complete description” of something: “I can make drawings of a rock from various angles, but if I am asked to say what the real shape of the rock is, I can give a description of it as in three-dimensional space which is independent of the angle from which it is looked at. The description of what is really there, as it really is, must be independent of any particular point of view” [18] (p. 356). Dummett then adds: “I personally feel very strongly inclined to believe that there must be a complete description of reality; more properly, that of anything which is real, there must be a complete, that is, observer-independent-description” [18] (p. 356).

For Dummett, a complete description is an observer-independent description, that is, as his reference to angles from the object is looked at clarifies, a perspective-independent description. But every time an object is perceived, it is perceived from somewhere. A perspective-independent description, insofar as it is gleaned from perception—what else could it be derived from?—can only correspond to a view from nowhere. And, as Merleau-Ponty points out, “to say that the house itself is seen from nowhere is surely to say that it is invisible!” [19] (p. 77). A view from nowhere is no view at all. Underpinning the notion of the view from nowhere is, again, the supposition that perception, being irremediably perspectival, cannot present things as they are. And this idea goes hand in hand with a staunch form of metaphysical realism, one which begins with a postulated, mind/observer/subject independent reality, a reality that is somehow grasped before any grasping begins, and proceeds to explain how subjective, perspectival vision can facilitate observers’ acquaintance with it14.

There are numerous criticisms of metaphysical realism of this kind15, the gist of many of which is that objects cannot be thought of independently of how they are given to consciousness. Husserl’s identification of the perceived object with an intentional object, and of optimality with normality, is perhaps the most powerful rectification to this misguided thought. But his addition of ideal optimality introduces back into phenomenology the same kind of troubling metaphysical realism that separates the world as it is from the world as it is for us, giving the former metaphysical and epistemological priority, for it is only from this ideal viewpoint that we see, to use Dummett’s phrase, “what is really there”. The “intended object simpliciter” is precisely the viewed-from-nowhere object of this type of metaphysical realism.

This needs to be qualified. Husserl’s ideal viewpoint, his integral of all optima, is not the view from nowhere but the view from everywhere. But I think it is not hard to appreciate that the similarities between the view from nowhere and the view from everywhere outweigh the differences. The view from everywhere, just like its counterpart, is no view at all. Only God (and perhaps not even God) views things from everywhere (or from nowhere). And it casts a shadow on real viewpoints for it introduces a bar in relation to which they are inferior. In particular, the notion of the view from everywhere undermines Husserl’s important claim that normal vision is optimal because it tacitly entails that the real object is never completely and perfectly presented in ordinary perception16.

Indeed, there seems to be a tension in Husserl between the claim that normal vision is optimal and the notion of a superior type of optimality associated with the sum of all viewpoints. Husserl himself, who is aware of the tension, proposes to explain away the “illusory sense of contradiction” [17] (§143, p. 299) by clarifying that “a thing in the real world, a Being in this sense, can within the finite limit of appearance appear only ‘inadequately’“ [17] (§138, p. 289). The “complete givenness” to consciousness of the object requires ideally optimal perception. This special viewpoint, “the integral of all optima”, is only an ideal, and the object correlated with it is meant “as Idea” (in the Kantian sense)” [17] (§143, p. 299), that is, as a regulative idea. The semblance of contradiction is thus eliminated by distinguishing “finite givenness” and “givenness in the form of an Idea”. And Husserl explains that “In the one case Being is ‘Immanent Being’ … in the other case it is transcendent Being” [17] (§144, p. 300). The identification of normal with optimal, so it turns out, is limited to the immanent intentional object but not to the transcendent intended object.

Doyon follows Husserl and elaborates: “There is no contradiction here, however, for just as he thinks of optimality in different keys depending on the attitudes or interests, Husserl also mobilizes different concepts of ‘truth’. The kind of truth disclosed, in the perceptual mode, through the fulfillment of practical intentions is—just as optima themselves are—relative, or partial, as they are relational and dependent on the interest of the experience. In that sense, perceptual cognition never yields apodictic truth or cognition. But if we think of perceptual optima as constituting an infinite continuum of appearances yielding absolute or adequate givenness, then the very idea of such a series represents truth in the highest, apodictic sense” [12] (p. 180).

Supposedly, then, the thesis that normal perception is optimal and reveals the object fully, perfectly even, remains intact also after the introduction of the further notion of ideal optimality. Two notions of object, the intentional object and the intended object simpliciter, two notion of cognition, and two notions of truth, secure the diffusion of the tension. Still, Husserl’s ideal optimality is disturbing because it renders normal vision optimal only in a qualified way, only in relation to the perceiver’s interest in the object and not in relation to the object’s “appearance”. Husserl’s inclusion of ideal optimality in his analysis of perception betrays an unease with the idea that normal perception is fully adequate. It clearly implies that actual perception can always be improved. As Doyon remarks in his discussion of the teleological character of perception, “it is in principle always possible to optimize our experience and gain more determinate content” [12] (p. 182). This, I believe, does not sit well with the claim that normal perception is optimal. What is optimal cannot be further optimized.

The train of reasoning leading to the introduction of ideal optimality is, again, grounded in the observation that there are infinitely many perspectives from which to view an object. But I wish to now argue, in this instance, at least, the premise does not entail the conclusion, for more is not always tantamount to better. When reaching my desk and finding the coffee cup my friend left there for me, it is irrelevant for me to subject the cup’s outer skin to a visual inspection under a microscope or infrared light. If I attempt viewing the cup from somewhere within the liquid, I will lose my coffee. The vision I have of it is already as adequate as can be and affords me full grasp of it—to deny this is to forget that perfection is not an absolute term, and has meaning only against a background of interests and significance. That there are additional perspectives is utterly irrelevant here, for they would improve nothing. The existence of other viewpoints does not mean that actual perspective-bound perceptions of the object are incomplete or imperfect, or that there is anything specifiable to gain from occupying them. Different perspectives answer different needs. They do not show the object better, they show it in another way. Seeing an object from a new angle does not bring us closer to an ideal and does not always advance our grasp of it.

We should not let the observation that there are infinitely more potential perceptions intimidate us into thinking that we are seeing only a part of the object. We know how to integrate functions and perform infinite summations, but it is far from clear that in this instance the notion of “the sum of all points of view” so much as makes sense. Take a random number generator. It works and works, spewing more and more numbers, and it can continue doing so indefinitely. There is no sense in which a new number just generated brings us closer to anything. It is questionable whether, in the absence of any arranging principle, it is even possible to speak of the set of all these numbers, or of a totality of all numbers thusly generated17. And even if we do insist on speaking of this collection, it remains utterly indescribable and cannot be thought of as something to which we are getting closer with each new number.

Analogously, the fact that there are always infinitely more unfulfilled viewpoints does not necessarily entail the existence of an ideal, optimal viewpoint, or of an ideal object that is correlated to this (unachievable) viewpoint. To reach that conclusion more is needed. For example, if one already has the distinction between the intentional object, the noema, and the intended object simpliciter (for Husserl, the object as a Kantian Idea), then one can justify the notion of an ideal viewpoint as the viewpoint that reveals the intended object. But, unless one has separate, unconnected reasons for positing an object as it is independently of all perception of it, this move is viciously circular: the distinction between two kinds of viewpoints, normal and ideally optimal, is explained in terms of two conceptions of object, the perceived object as actually perceived and the intended object, and this latter distinction is derived from distinguishing normal viewpoints, which are perspectival and therefore partial, from ideal viewpoints. In short, normal vision is claimed to afford only a partial perspective on the object because it does not reveal the ideal object, but the very supposition that there is such an ideal object already assumes that normal vision is partial.

This circularity goes unnoticed for the following reason. We often do seek new perspectives in order to improve our visual grasp of an object, additional viewpoints frequently do impart new and important visual information. This fact underpins Husserl’s studies of the dynamic aspects of perception, of the kinesthetic activity perception is inextricably linked to, and of the role of protentional and retentional consciousness in perception. Reflection on this central feature of perception can easily lead to the thought that the more viewpoints an object is perceived from the better one’s visual acquaintance with it. The notion of an intended object, conceived of in relation to a view from all possible viewpoints, a view from everywhere, follows readily. But, again, this reasoning is flawed. That sometimes additional viewpoints make a contribution does not mean that every new perspective adds something. This further step requires a justification of its own. Phenomenologically, there are no obvious and straightforward grounds for this thesis. And basing it on the notion of an object as a Kantian Idea renders the reasoning circular18.

In sum, that there are always additional perspectives does not entail that normal perception is partial, even when it is acknowledged that perception is a dynamical affair that standardly requires movement and additional viewpoints. Freed from the worry that normal perception is irremediably partial, from the shadow cast by the notion of an ideal viewpoint constituted by the sum of all optima, we can fully embrace Husserl’s claim that normal perception is, indeed, optimal, without lapsing back to the oxymoronic idea of an optimum that can further be optimized.

6. Science and Phenomenology

There are no privileged viewpoints, be they views from nowhere, from everywhere, or “canonical” in the sense of being foundational. Could it be that the quest for such viewpoints is driven by the sense that for normal vision to actually be optimal would be too good to be true—how can finite, hopelessly perspectival creatures grasp things in a manner preserved for supreme beings that are not bound by space and time? Conjuring the view from everywhere or from nowhere, as a Kantian regulative idea in Husserl, or as an optimum in relation to which the quality of one’s actual perception is measured for Merleau-Ponty, is a way of stating that normal vision, impressive as it may be, is still not truly optimal, and does not reveal the object in all its reality. But then establishing that there are no privileged viewpoints of any kind strengthens the position that normal vision is indeed optimal, even perfect, for there is no superior viewpoint in relation to which normal vision would count as inferior. Husserl’s identification of optimal with normal says just that. Indeed, if the perceived object is an intentional object, then the quality of one’s perception of it can be measured meaningfully only in relation to the interest of the perceiver in the object’s significance, an interest that can be fully satisfied. Hence the optimality of perception.

This understanding of perception contrasts sharply with the one resulting when perception is analyzed from the viewpoint of the mechanism of perception. As was repeatedly stated above, because a key component of the mechanism are retinal images, this approach inevitably leads to some variant or another of the two-step model. When studying the mechanism, if the fact that the mechanism is one of vision is set aside, and if we attempt to build our understanding bottom-up, as it were, from retinal images to veridical perception, then we are already committed to a two-step model, to veridical perception being the outcome of a correction (step two) made after (step one) skewed input is imprinted on the system. My aim in Section 3 was to show that such a correction can only be made if the system already “knows”, so to speak, what the skewed input has to be corrected to, but that there is no way for it to possess this information. This, I argued, damns the two-step model, but more generally, it shows that the notion that perception can be studied by studying its mechanism independently of a phenomenology of perception is also hopeless.

Phenomenology precedes science. Thus, a study of perception should take as its point of departure the fact that, for example, all the tiles making up the floor are seen to be of the same shape, even though, if drawn, they would be represented by as many different rectangles as there are tiles. So-called perceptual shape constancy should be thought of as a constitutive feature of perception that crucially informs, rather than challenges, both the scientific and conceptual analysis of perception. Given that shape constancy is an experiential given, this starting point should be abandoned only in the face of decisive evidence against it. It is certainly not experientially given that perception is mediated by retinal images, and there is no reason to prefer this thesis as the investigation’s guiding premise. Failing to realize this, holding the empirical study of the mechanism of perception to be prior to phenomenology, turns constancy from an experiential datum illuminating the inquiry into a liability, for now it becomes necessary to explain how different, evolving retinal images underpin an experience of a stable, unchanging shape. Science cannot satisfy this obligation for it lacks the vocabulary with which to articulate it—veridicality is a normatively laden term that cannot be dealt with by a framework that knowns nothing of normativity.

It is, however, easy to get confused at this juncture, for there certainly are retinal images, and they do play a vital role in vision. Moreover, even rudimentary reflection on how change of perspective affects perception can lull us into granting them a role as mediators. If a coin is drawn from several perspectives, the different ellipses on the page will correspond, supposedly, to the retinal projections caused by photons arriving from the varying perspectives in question. It is a small step from this observation to the thesis that retinal images are indeed what perception begins with, and that perception is a two-step, mediated affair, in which, with the aid of previously attained information and neural computation, skewed input evolves into veridical visual experience. This, however, is precisely how notions belonging to analyses of the mechanism of perception get wrongfully introduced into the phenomenology of perception, e.g., in the form of assertions to the effect that it is the role of the visual system to “convert visual images into the perception of objects” [5] (p. 1). Such contentions, which are routinely recited in the scientific literature, carry terms such as “retinal images” or “retinal projections”, coined in the courses of empirical research, over into an altogether different domain of study, namely, that of phenomenologically characterizing what perception is. This happens subtly, and it is hard to detect. The contention sounds innocuous enough. But, albeit tacitly, it already gives rise to the idea that perception is a two-step process.

Again, retinal images are vital for perception, as is the process, a Bayesian process, perhaps, that bridges between them and visual experiences. However, retinal images are physical entities; like screws and wires, they are, at bottom, patterned arrays of interactions between photons and receptor cells, or, more fundamentally, between photons and atoms. They are crucial for perception, but we need to know what perception is before we discover, describe, and study them. And that, what perception is, we know first from experience, and then from phenomenology. Both precede and underpin empirical research, and in neither do retinal images play a role.

Vision science has made immense progress over the last few decades and it is tempting to set aside “traditional” phenomenology and let the ingenious scientific theories, those pertaining to the anatomy of the visual organs or of the brain, and those modeling neural computations, set the agenda for a new phenomenology, one structured around an understanding of perception as a multi-staged process that begins with flat “appearances”, embodied in 2-dimensional retinal images, that by means of neural computations are “converted” into 3-dimensional representations of objects. This temptation must be resisted. Letting terms that are introduced in the course of vision science guide phenomenology leads to conceptions of what perception is that are phenomenologically untenable, for example, because they do not square with normal vision being optimal.

It is important to stress that this alternate phenomenology is not forced upon us by science. It is not part of the science of vision, and it is not established or confirmed by empirical evidence. Actual empirical research does not assume it and does not require it. It emerges only when, alongside empirical research, a phenomenology grounded in science is articulated. That is when two-step models are born, and the point is that such models are no longer part of science. Yet it happens that they come to be regarded as a working hypothesis for scientific research, and as a result get treated matter-of-factly as expressing scientific findings, which they do not. It should also be noted that these models make no contribution to actual scientific practice. Scientifically, nothing would be lost if the two-step model were replaced by “traditional” phenomenology.

The seepage of terms belonging to science into phenomenology is not unavoidable. There is no reason for what happens on the retina, or for the hypothesis that perception consists of corrections of skewed projections, to be part of the phenomenological account of perception. Saying that we see things as they are and talk of retinal images reside on different planes of discourse, i.e., the phenomenological and the scientific. With this separation acknowledged and observed, we can marvel at the fact that when we look at X we already see it as it is, that there is no gap between how we actually see it and how we would see it had we had the eyes of angels19.