Causal Pathways of Rational Action

Planning and Scheduling

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 12

Planning: Representation

A proper characterization of a rational agent’s actions involves much more than simply recounting the changes in the world affected by the agent. It should also include an explanatory account connecting the upshots of an agent’s actions with the reasons behind those actions, where those upshots might represent actual changes (either intentional or unintentional) or merely counterfactual possibilities. The conventional view of action makes it difficult to distinguish, inter alia, cases of attempts, accidents, coercions, or failures - such distinctions useful to agents engaged in recognizing or assigning responsibility for actions. Such a view also makes the characterization of actions that do not involve physical change, such as maintenance events, difficult, as well as the proper representation of negative actions; the latter commonly appearing in explanations and as objects of an agent' s intentions. In this paper, I present a formal analysis of these sorts of actions in terms of the causal pathways joining an agent’s intentions with his actions.