Whenever you have a good idea...

Somebody else has come up with it...

So as I have been researching this more, I have discovered Maxent, and its applications in AI. Maxent, short for Maximum Entopy From wikipedia;

In physics the Maximum entropy school of thermodynamics (or more colloquially, the MaxEnt school of thermodynamics), initiated with two papers published in the Physical Review by Edwin T. Jaynes in 1957, views statistical mechanics as an inference process: a specific application of inference techniques rooted in information theory, which relate not just to equilibrium thermodynamics, but are general to all problems requiring prediction from incomplete or insufficient data (such as for example image reconstruction, spectral analysis, or inverse problems).

So in other words, statistical mechanics is based on an inference processes which let us build incrementally more accurate models of intractibly complex systems...Which as far as I am concerned, is AI in a nutshell.

I'm not sure how Maxent stacks up to other systems, but it seems to have been very successful at Natural language processing. A great tutorial, here, gives translation of English to French as an example. The most widely cited Maxent software I have found comes from the OpenNLP (open natural language processing) project at sourceforge.

Anyway, its vaguely dissapointing to find that the path you were walking has been walked before, buts its also very encouraging to see that its been walked before and it worked. It means my intuition is good, and I'm not nutty. If my intuition continues to be good, entropy can be used as a broad unifying model to describe a bunch of superficially different systems in AI.