We can use the word "see" in two different ways: we can refer straightforwardly to things that we see, or we can "notice an aspect": I can also see something as something. Wittgenstein gives a picture that can be seen either as a duck or a rabbit. We can see it either as a duck or as a rabbit, but not as both simultaneously. If I have only ever seen it as a rabbit, I do not say, "I am seeing it as a picture-rabbit," but simply, "this is a picture-rabbit." Nonetheless, someone else might well say of me, "he is seeing it as a picture-rabbit."
Our temptation is to say that there is an immediate perception that consists of colors and shapes in my visual field, and then also an interpretation of that perception, where I see those colors and shapes as a particular object. But I do not say of a fork, "I am seeing this as a fork," except in unusual circumstances. And when of the duck-rabbit I say, "I see a rabbit," I am reporting, not interpreting, my perception. My perceptions—what I see—are of things and people, not of colors and shapes.
Seeing is too complicated to be reduced to visual sense impressions. The element of familiarity, for instance, helps us to recognize a portrait that is the right way up far more easily than one that is upside down.
If "what I see" is just an inner picture, I see the same thing when seeing the rabbit or the duck aspect of the duck-rabbit. In this case a picture of "what I see" is an incomplete description because it does not fully make clear what I am seeing. Is it a different visual impression when I see the duck and when I see the rabbit? There is no inner picture to justify expressions of what we see or how we see it: there is simply how we describe it. The concept we bring to a picture forces itself upon us without interpretation. We could call it the attitude with which we look at the picture.
The experience of seeing certain objects as certain other things often requires a kind of knowledge or "mastery of a technique" in order to experience them as such. For instance, I must be familiar with ducks and rabbits to see those two aspects of the duck-rabbit. What we can do, what we know, often determine the kinds of experience we have: to talk about a hesitant posture or a plaintive melody presupposes some experience of certain emotions and their common expressions. That one fails to experience a certain aspect is not a defect in one's eyesight.
We can be struck by an aspect of a picture for a certain length of time, and then no longer be struck by it. This does not mean that our visual experience changes, or even that our thoughts change. What dawns on me when I see an aspect of a picture is not so much something in the picture itself, but in its relationship to other pictures.
We are more inclined to say we see a different aspect, not that we interpret the picture differently, because seeing is a state, and interpreting is a thought. Interpretation implies a kind of hypothesis, and no such hypothesis exists when we see a duck-rabbit as a duck.
Wittgenstein does not direct us to any particular conclusion here, but rather asks us to consider a matter more closely. In discussing the many different ways we can talk about "seeing," he is trying to complicate certain notions that are most apparent in sense data theory.
The basic idea of sense-data theory is that what I see are not objects themselves, but intermediary "sense data." There are a number of arguments for this position. In different lights, a room may look different, but the room itself does not change; therefore, it must be my sense data that change. When I see a stick half-submerged in water, it appears bent, but it is not bent; the bent stick occurs only in the realm of sense data. When I hallucinate, I see all sorts of things that don't really exist; they are only sense data.
A theory of sense data raises immediate skeptical questions. If what I see are not the things themselves, but only sense data, then how can I know there is any world external to my sense data? I could be a brain in a vat, hooked up to wires that give me the appropriate sensations at certain times. On a less far-fetched level, scientific investigation needs a foundation that acknowledges sense data, and not things themselves, as its object of study. Bertrand Russell and Rudolph Carnap have both made notable attempts to set forth such a foundation.
Wittgenstein's argues that seeing is a far more complicated activity than the sense-data theorists assume. We misuse words like "see" and "interpret" when we assert that all we see are sense data, and that we interpret these sense data as certain kinds of objects in the world. I do not first see a silver sense data of a certain shape and then interpret it "as" a fork. The example of the duck-rabbit is, among other things, meant to bring out when we can correctly speak of seeing something "as" something.
There are two obvious flaws with this notion that we see sense data and then interpret them as certain objects. First, we do not see sense data, and second, no act of interpretation takes place. The first point can again be demonstrated through the example of the duck-rabbit. Someone who is unaware of the duck aspect is just as justified in saying, "I see a rabbit," as someone who says, "I see a fork." We could say the same for someone unaware of the rabbit aspect saying, "I see a duck." Both these people have the same "sense data," but they are seeing two different things.
Regarding the second point—that seeing involves an act of interpretation—Wittgenstein points out that an interpretation requires thought. I can interpret pictures, but I by no means always interpret them. We have no reason for claiming even that there is a different mental act in the person seeing the duck and the person seeing the rabbit. Imagine someone who has grown up in a town filled with ducks, but has never seen a rabbit before. There is no mental act of "seeing it as a duck," because he does not even have the capacity to see it as a rabbit. This is not a shortcoming in his visual or mental apparatus, but simply a fact about his experience.
H. P. Grice, among others, has criticized this line of criticism against sense data theory. Grice argues that this criticism stems from a failure to distinguish semantics from pragmatics. To argue that we do not talk about "seeing a fork as a fork" is simply a matter of linguistic convention, and should have no bearing on the matter at hand. Regardless of what conventions we use to talk about seeing, the fact remains that my visual impressions are not the same as the objects we meet in experience, and it is worthwhile to distinguish between the two.
The objection raised by Grice is a complicated issue, and continues to divide philosophers today. A Wittgensteinian response would take the form of pointing out that we cannot so readily talk about experience divorced from linguistic conventions. A Gricean would set up his sense data theory by using words like "see" and "interpret," and expect us to understand him because he is using these words in ordinary ways.