The Markov assumption: formalization and impact
Alexander Bochman · International Joint Conference on Artificial Intelligence · 2013
We provide both a semantic interpretation and logical (inferential) characterization of the Markov principle that underlies the main action theories in AI. This principle will be shown to constitute a nonmonotonic assumption that justifies the actual restrictions on action descriptions in these theories, as well as constraints on allowable queries. It will be shown also that the well-known regression principle is a consequence of the Markov assumption, and it is valid also for non-deterministic domains.