LILaC
- paper abstract
An assessment of actions with indeterminate and indirect effects in some causal approaches
Herzig, Andreas and Varzinczak, Ivan José
Abstract:
In this work we investigate the behavior of the main existing fluent-indexed approaches to reasoning about actions in dealing with domains that have actions with both indeterminate and indirect effects. As they all fail in solving this kind of scenarios, we argue for an action-indexed causal notion in order to deal with the frame and ramification problems. This is achieved by defining a weak form of causality in terms of a dependence relation involving actions, literals and formulae. This relation allows the literals to change their truth value without forcing or causing it. Once integrated in the framework of a propositional logic close to PDL, it gives us a simple and powerful formalism to reasoning about actions and a decision procedure in terms of tableau methods. We also show how our approach can deal with scenarios involving indeterminate and indirect effects without being subjected to the same problems of other formalisms. In order to help the knowledge engineer in describing domains, we give an interactive algorithm for generating the required dependence information.
Keywords:
Reasoning about actions, causality, dependence relation.
Note: A
preliminary version
of this paper was presented at NMR'02.
Bibtex-entry:
https://www.irit.fr/~Andreas.Herzig