Reasoning about knowledge and action
This report deals with the problem of making a computer reason about the interactions between knowledge and action. In particular, we want to be able to reason about what knowledge a person must have in order to perform an action, and what knowledge a person may gain by performing an action. The first problem we face in achieving this goal is that the basic facts about knowledge we need to use are most naturally expressed as a modal logic. There are, however, no known techniques for efficiently doing automatic deduction directly in modal logics. We solve this problem by taking the possible-world semantics for a modal logic of knowledge and axiomatizing it directly in first-order logic. This means that we reason not about what facts someone knows, but rather what possible worlds are compatible with what he knows. We integrate this theory with a logic of actions by identifying possible worlds with the situations before and after an action is performed. We use these notions to express what knowledge a person must have in order to perform a given action and what knowledge a person acquires by carrying out a given action. Finally, we consider some domain-specific control heuristics that are useful for doing deductions in this formalism, and we present several examples of deductions produced by applying these heuristics.
- Research Organization:
- SRI International, Menlo Park, CA (USA). Artificial Intelligence Center
- OSTI ID:
- 6755363
- Report Number(s):
- AD-A-126244/3
- Country of Publication:
- United States
- Language:
- English
Similar Records
Reasoning about change and exceptions in automated process planning
Summary of Glue Tests 1993