| |
| | [No title] |
 | | This paper suggests non-monotonic reasoning about the utilities of outcome states in decision analysis, for two reasons: first, it allows decision theory to be used on the kind of problem spaces to which AI planning is accustomed; second, it allows incremental refinement of the decision models. |
 | | Assigning utility to a set of possible worlds, W, described by a sentence, p, is also no problem, so long as we understand it as the weighted average of the utilities of each possible world, u(w(i)), for w(i) in W, where the weight of u(w(i)) is prob(w(i)). |
 | | Or it is supposed to be an arbitrary convention adopted by the specifier of utilities for the sake of convenience, as a shorthand. |
| www.cs.wustl.edu /~loui/deflities.text (3634 words) |
|