Sunday, September 10, 2006

Information and inference

I had an idea recently that I want to put out there in hopes of making it better at some point. The idea is that there seems to be a link between John Perry's information games and the Brandom-Sellars notion of material inference. Information games are transferring and applying informatoin to one's previously held or newly created information files. Learning new information about Bill let's you put that information into Bill's file. Perry explains more about the information games in several places, in particulatr Reference and Reflexivity. When someone makes an indexical utterance, say "I am hungry," that gives you information about the speaker, i.e. the person who said that is hungry. If you make the further connection that the speaker is Bill, then you are licensed to add that information Bill's file and infer that Bill is hungry. The connection should be clear. Information and files can give licenses to substitute identical singular terms, so from F(a) to F(b) where a=b. This is the model of inference that Brandom discusses. Going a bit further, since informatoin is what the world has to be like given some perceptions and constraints to which the agent is attuned, the proper constraints will allow the agent to infer that Bill hasn't eaten for a while, or from F(a) to G(a), substitution of frames. The description of information as the way the rest of the world must be in order for something to have happened sounds a lot like incompatabilities. Given that p, and p precludes q, if the constraint is accurate then not-q. If you get the information that Ed is a dog, and being a dog is compatible with being a mammal, then the world can contain Ed the mammal. However, this conclusion is weaker than the starting premise since one cannot go backwards from it. The constraint doesn't allow that.

At least, I think the direction from information to inference works. The direction from inference to information is a little trickier. The (in-)compatability and information connection works in both ways, since the sketch of an argumnt given above seems to reverse. I'm not sure how the argument from the symmeteric substitution of terms and and the asymmetric substitution of frames. In other words, I have an argument that info->infer, but I still need an argument that infer->info. Once I come up with that I can figure out what is wrong with this idea.

No comments: