The Semantics of Event Prevention

In planning tasks an agent may often find himself in a situation demanding that he choose an action that would prevent some unwanted event from occurring. Similarly, in tasks involving the generation of descriptions or explanations of sequences of events, it is often useful to draw as many informative connections as possible between events in the sequence; often, this means explaining why certain events are not possible. In this paper, I consider the semantics of event prevention and argue that a naive semantics which equates prevention with the elimination of all future possibility of the event in question is often difficult, if not impossible, to implement. I argue for a more useful semantics which falls out of some reasonable assumptions regarding restrictions on the set of potential actions available to an agent: (1) those actions about which the agent has formed intentions, (2) those actions consistent with the agent's attitudes (including its other intentions), and (3) the set of actions evoked by the type of situation in which the agent is embedded.