blog.kennypearce.net

The Nature of the Visible Space in Berkeley's New Theory of Vision

One of the main aims of Berkeley's Essay Toward a New Theory of Vision is to argue that the application of spatial vocabulary ('far', 'near', 'big', 'small', 'right', 'left', etc.) to how things look ("visible objects") is derived from the primary meaning of that vocabulary as applying to how things feel ("tangible objects"). A big object is one you can't fit your arms around. An object looks big when the way it looks makes you think that you probably wouldn't be able to fit your arms around it. It is only by experience that we learn that objects we can't fit our arms around look a certain way, and then begin to apply the term 'big' to that characteristic look. Note that looking big is a very complicated property since, for instance, big objects look very different seen from close up or from a distance.

There is a debate among interpreters as to just how radical the dependence of visual space on tangible space is supposed to be. According to some interpreters, nothing would look big to the 'Molyneux man' (the man who was born blind and just began to see for the first time) until he had learned how to correlate visual and tangible experience. His visual data would be a complete jumble or, as William James famously said, a "blooming, buzzing confusion." Other interpreters believe that, according to Berkeley, some of the things seen by the Molyneux man would have the property looking big, but the Molyneux man would not know that it was correct to apply the word 'big' to those things, since he had learned that word only by tangible examples.

The first interpretation is problematic because if the visual stimulus was completely unstructured then there would be nothing for perceivers to latch onto to correlate with tangible stimulus. The second interpretation, however, doesn't seem to be sufficiently radical to capture Berkeley's intentions. I think we can chart a middle course: the Molyneux man would see things that had the property looking big. However, it is not just that he wouldn't know that these things were properly described as 'big'; he wouldn't even be able to pick out looking big as an interesting property worthy of attention. This can be clarified by the following two examples.

Imagine you are learning to read a language. (Vision is, after all, a language according to Berkeley.) You do not know any of the symbols in the writing system. Nevertheless, some of the words you are seeing have the property being in the fifth sentence. The property being in the fifth sentence is the same property as occurring between the fourth and the fifth instance of the symbol '.', and under the latter description your visual stimulus as you look at the page is perfectly adequate to allow you to pick out those words. However, not only do you not know that the property occurring between the fourth and the fifth instance of the symbol '.' is the same property as being in the fifth sentence, you also don't know that the symbol '.' has special structural significance. As a result, if you were, say, trying to ask about the meaning of a particular collection of symbols, you wouldn't think of specifying its location as being between the fourth and fifth instances of '.' as being any more helpful or informative than saying that it was between the 18th and 19th instances of 'a'. All the structural information is right there in your visual field, but it's not just that you don't know what to call it or how it is associated with the spoken language; it's also that you don't know which of the features you are looking at are important structural features and which are not.

Here's the second example. Suppose you've got the data for a bitmap image. This is a series of numbers which represent sequentially the color of each pixel in the image. Suppose further that you know that it's a bitmap image, but you don't know the width or direction of flow - that is, you don't know whether it goes left to right or right to left, whether it goes top to bottom or bottom to top, and you don't know where to break and start a new line. Now suppose that, in fact, it is a top to bottom left to right bitmap (as most bitmaps are) and that it is 100 pixels wide. Now it turns out that the property being immediately above pixel n (for any n) is the same property as being pixel n-100. Now you knew about the latter property all along and, for any given pixel n, you could specify which pixel had it. However, you didn't know that the property being pixel n-100 was more significant to the interpretation of the data than, say, the property being pixel n+23.

In the same way, I think, it is Berkeley's view that the structures which, in ordinary perceivers, are interpreted as spatial properties, are already present in the Molyneux man's visual experience. However, it is not merely that he doesn't know to associate them with tangible spatial properties. Rather, he is completely unaware that these features of his experience are especially important to the interpretation of his experience, and he therefore has no reason to direct his attention to them. Ordinary perceivers, on the other hand, are so used to organizing their experience in this way that they can't think of it in any other way, any more than you can look at this blog post without automatically thinking of it as organized into sentences.

Here is an experiment: try to see this page as divided into units bounded by each occurrence of 'a'. I find that I really can't do it. If you think that's because '.' is small and therefore creates some spatial separation between letters, then try bounding the units by ',' instead; I don't seem to be able to do that either. This is why we ordinary perceivers are completely unaware of any other way of organizing our visual experience than by the spatial location of the objects seen.