The glasses analogy: a radical proposal about the epistemology of basic modal knowledge.

In the paragraphs above I described a Quine-inspired problem for learning truths about our own application of language by applying induction to our past judgments. Perhaps we get to knowledge of what is (linguistically) possible by doing induction over what is actual. The problem is, that everything which is actually true is both metaphysically possible and permitted by the laws of physics. So this theory would have us applying induction to the same data points to get out on the one hand physical laws and on the other hand linguistic/metaphysical necessities. And this can’t be right since surely there are genuine possibilities which are ruled out by the laws of physics. So if we get knowledge of linguistic possibility through induction, how does the linguistically possible end up differing from the physically possible?

I propose that there are two factors, one major and one minor, which end up differentiating analytic from physical possibility. I’ll start with the major factor. Suppose that you wake up to find a pair of glasses attached securely to your head, and a orangeish blotch in the bottom left corner of your visual field. Most people would agree that it should be possible for you to figure out whether the orange thing you are seeing is on your glasses, or the glasses are clean and you are seeing it through your glasses. If, as you move your head around, the spot follows evenly it is a better hypothesis to assume that the spot is on your glasses than that it is roaming around outside and just happens to move whenever you do. On the other hand if it shifts its place in your visual field as you move your head and eventually disappears, it would be a better hypothesis to say that the orange thing is not attached to your glasses, then to say that it is on your glasses but it moves around and shrinks and grows in just the right way. Now note, that the hypothesis that an orange thing which isn’t attached to your glasses just happens to be floating around in just the right way is perfectly consistent with your observations. We can only reject this kind of hypothesis and thereby learn about whether the spot is on our glasses or not because we have ideas of scientific virtues which go beyond matching the data.

To put the point in more detail: Any sequence of observations could be accounted for by either a theory which posits that the orange material is on your glasses or that it is not. So it might seem like it is impossible to tell where the orange stuff really is. But this is not correct, because we also have cannons of scientific virtues like simplicity/we assign different a priori probabilties to empirically equivalent theories. Thus we can find out where the spot is by moving our head around until we get a sequence of experiences such that the explanation of them in terms of orange stuff on the glasses is a much better theory scientifically speaking than the explanation of them in terms of it being seen through the glasses or vice versa.

Now I want to suggest that our project of scientifically understanding ourselves in the world is like this discovery about the glasses. The ‘data’ we start with are the different sets of sentences we have judged to be true at various times. We then simultaneously embark on two projects. One is a logical/linguistic project of theorizing about how our words contribute to making sentences true or false (think for example about the kind of discovery involved when a person goes from being able to use the words ‘and’ and ‘not’ correctly to seeing that their contribution to the truth or falsity of sentences is well captured by standard propositional logic). The other is a more standardly scientific project of positing contingent laws about what we will or won’t find in the world. The facts about what sentences a person has judged and will judge to be true will be (ignoring errors on their part for the moment) jointly determined by these facts about how language determines the truth conditions for a sentence together with the physical laws (and maybe history) of the world.

Now it’s true that the fact that a person has always judged a particular sentence to be true could be explained either by a story about how the terms in that sentence determine its truth conditions which already suffices to ensure that it will always come out true or by a story on which these linguistic facts don’t determine the sentence’s truth value except in combination with contingent physical laws. But this doesn’t mean that both (linguistic theory, physical theory) pairs will be equal in scientific virtues or warrant the same a priori probability. Thus I think the cannons of scientific virtue together with induction can give us justification for sorting the sentences which we know to be true into those whose truth is ensured by language, and those which express consequences of physical law and those whose truth is a mere matter of chance, in one way rather than another. This is the major factor.

The minor factor, which I think also plays a role in distinguishing between what we take to be analytic/logical/metaphysical necessity and physical necessity is, roughly speaking fictional and (modulo some Kripkian concerns) mistaken speach. The idea would be that part of our learning to use certain terms was not only learning to apply them correctly but also to certain scenarios described as fictional or possible but not actual. So for example, it would be part of our training to make or respond to certain kinds of fiction but not others (e.g. ‘the magician waved his wand and the cup floated’ vs. ‘the magician waved his wand and the cup ceased being identical to itself). (Borges might be an exception here). This training in certain fictional examples would mean that we would have two distinct bodies of judgments to apply induction to. First there would be the facts about what you had actually judged as true, and applying induction to these would give you an idea of what is physically necessary. Then there would be the body of things you have actually judged to be true together with the things you have judged to be true of certain stories which you had been trained to consider as possible. The inclusion of these additional data points then could make a huge difference to which inductive generalizations are supported, and this different, weaker set of generalizations becomes your idea of what is linguistically/analytically nessicary