Saturday, August 13, 2005

The Content of Representation

It can be really difficult to know how to interpret the meaning of a representation. Consider again my example of the mouse who scurries away from hawk-shaped objects. Let's suppose the mouse's behaviour here involves some significant cognitive processing, which includes matching up perceptual data against his internal hawk-shape "search image". How should we interpret this search image? Does it represent hawks, or predatory birds, or something to scurry away from, or merely such and such a shape?

Suppose I create a hawk-shaped cardboard cutout, and hold it up above a mouse, which promptly scurries away upon noticing it. There are (at least) two interpretations of the mouse's cognitive processing: (1) It formed a false belief-like representation that there is a hawk; or (2) It formed a true belief-like representation that there is a hawk-shape. (This might then combine with the false belief that the hawk-shape posed a danger and so should be scurried away from.) How are we to pick between these two rival hypotheses?

Two strategies spring to mind. On what we might call the teleological account, we appeal to the idea that the mouse's internal search image has a specific biological function that led it to be selected for in the evolutionary history of mice. The intentional content of the representation is fixed by this biological function. That function, let us suppose, was to identify hawks. So the mouse's search image represents hawks, and my cutout tricked the mouse into forming a false representation. That is, interpretation (1).

Alternatively, we might prefer an instrumental account, according to which the content of a representation is fixed more holistically, by its place in our best (most predictive) theory of the agent's entire intentional set (i.e. all its beliefs, desires, etc.). The information given so far is simply insufficient to establish a fixed content for the hawkish representation. But if it turned out that our best holistic theory of the mouse's mind ascribed only content about hawk-shapes, and not about hawks per se, then that would yield interpretation (2). (That's a big "if" -- we might just as well end up with an entirely different interpretation. It all depends on the details of the 'big picture'.)

This problem arises for languages too. To adopt an example presented in (a different) class, imagine two humanoid species, called "Hubots" and "Rumans". They live in a land where yellowish metal is scattered about, which both species call "golper", and is their staple food. But suppose it turns out that this metal is not one common substance after all. Instead, half of the metallic clumps are actually gold, and the other half are copper. Further, it is only the gold that has nutritious benefits for hubots, and only copper is nutritious for rumans. Here's the question: What does the word "golper" mean? (Does it mean different things for hubots and rumans?)

On the teleological account, we're led to conclude that "golper" means gold in hubot-speak, but copper in ruman-talk. Assuming that they're two distinct linguistic communities, we can take the word 'golper' in each case to mean something like "that nutritious yellowish metal which feeds and sustains us" -- a description that picks out different substances for the two communities.

Alternatively, on an instrumentalist account we might instead say that "golper" is more broadly defined to include both gold and copper. This would be especially plausible if there was much communication between the two species. If they formed a gathering party to go out and find "golper" together, then they understood themselves to be gathering the same thing. It seems odd to claim that they're mistaken about this. Even worse, suppose a hubot holds up some yellow metal and proclaims triumphantly, "I have golper!", to which a ruman replies, "Yes, you have golper." It would seem exceedingly odd to say that their two statements have different truth values!

One way to understand the instrumentalist method is to see it as applying a Davidsonian 'principle of charity', interpreting representational content in such a way as to make most of the agent's beliefs come out true. I think Dennett too talks about the "assumption of rationality" that accompanies adoption of the intentional stance. (I need to read more about that though.)

A seemingly easy way to resolve the problem would be to go up to the hubots and rumans, and explain the gold/copper situation to them, in neutral terms. Then ask: "So, given these facts, what do you take the word 'golper' to refer to?" Let them decide whether it means just one of gold or copper, or whether it's the disjunction of the two. If there's a clear consensus as to the answer, then it seems that fixes the objective truth of the matter in a more principled fashion than instrumentalist interpretations do.

Unfortunately, we can't converse with animals, so this simple solution isn't going to work there. So how are we to determine the contents of their mental representations? Should we go with teleology, or instrumentalism, or is there some other alternative?


  1. 'suppose a hubot holds up some yellow metal and proclaims triumphantly, "I have golper!", to which a ruman replies, "Yes, you have golper." It would seem exceedingly odd to say that their two statements have different truth values!'

    It seems to me that, in cases like this and the mouse example, subjectivity is important. If neither the hubot nor the ruman can tell whom the bit of golper in question will nourish, then the truth values of the two statements cannot be known until one of them eats the golper (without positing a more knowledgeable external observer). A real subjectivist might claim that the statements actually do not have truth values until one of them eats it. If they can tell the difference, then yes, one of them is lying, but your example doesn't seem to be set up that way, unless I'm misunderstanding.

    In the mouse example this doesn't quite apply, since you know, even if the mouse doesn't, that the hawk-shape it runs away from is not really a hawk. However...

    I do think it's reasonable to assume that the mouse in your example does recognize a hawk-shape, because it scurried away. If it were identifying hawks solely through some other mechanism than shape (such as recognition of its motion, or the sound of air over feathers, or some such), it would not have run away. So I think it's certainly true that your interpretation 2) is accurate. I don't think, though, that this necessarily conflicts with interpretation 1). Why can't the mouse simultaneously believe (correctly) that there's a hawk-like shape, and (incorrectly) that it is a hawk?

    In fact, I think it's necessary that there are two parts to the mouse's cognition, which becomes clearer if you observe that the mouse's pattern-recognition is not instantaneous and does not produce certainties, but probabilities. So, the mouse sees a shape and begins to analyze it, the first bit of cognition. At first glance it seems likely to be a hawk, and the mouse has to make a decision: the second bit of cognition. It must either flee immediately, or observe longer to see if the probability that the shape is a hawk (or is dangerous, if you prefer; I don't think it changes the analysis much) rises or falls as more accurate observations are made and the pattern-recognition system has more time to work on them.

    In other words, your two interpretations of the mouse's behavior might be better written, from the mouse's perspective, as 1) I have a pretty good idea of what that shape is, and if I'm right it's dangerous; and 2) I have no idea what that shape is, but it might be dangerous.

    Which may or may not change your later analysis; I'm not at all sure I understood it properly. I do think, though, that it's important to realize that pattern-recognition is not instantaneous and does not produce certainty, but probability.

  2. "I do think it's reasonable to assume that the mouse in your example does recognize a hawk-shape, because it scurried away."

    I certainly agree that the mouse discriminated the hawk-shape. But what I'm interested in is what it saw it as. ('Intentionality', in philosophical jargon, or about-ness.) Did the mouse perceive the image as a hawk, or merely as such-and-such a shape? (Put another way: do mice actually have a concept of hawks?)

    One way to tell these hypotheses apart might be (to build on something you say) to look for similar behaviour caused by "some other mechanism than shape". If hawk-shapes, hawk-smells, and hawk-cries all elicted similar behaviour from the mouse, then we might be justified in unifying these behaviours under the single concept of 'hawk'. I'll pursue this idea further in a future post.


Visitors: check my comments policy first.
Non-Blogger users: If the comment form isn't working for you, email me your comment and I can post it on your behalf. (If your comment is too long, first try breaking it into two parts.)