Reasoning About Knowledge

[maxkw]

I had a brief chat with Joe Halpern (Cornell Computer Scientist) and he suggested we read Chapter 4 of his book "Reasoning about knowledge". Its actually available online through the MIT library: https://lib.mit.edu/search/bento?q=reasoning+about+knowledge just click "View online".

The title of the relevant chapter is: "Knowledge in Multi-Agent Systems" check it out and let me know if its relevant?

[polyguo]

Common knowledge is a special case. Our current math around common knowledge is more than sufficient. We would only start caring about this material if we have things like A observes B->C but either B or C don't know. But really, i would just run our model as is and I think it would still do well naïvely. If we care about having perfect explicit models of this we're in POMDP territory, where we need to infer observations and games. There might be other interesting imperfect models, or constrained models around known common sources of information that we might want to use. As an example if social networks are known to determine odds of interacting and observing but agents don't get explicit knowledge of others' observations we can have a simple model over which we can do inference. Even then, this math is not quite relevant. Another interesting case might be that A observes that B did something to C but not what, but that's still outside the scope of this text.

[polyguo]

From what I've read the relevant generalization is "Subjective Logic", I started reading a book to confirm it does deal with the kind of things we're thinking about, but I'm not sure we need it yet.