A Bias towards Relevance: Recognizing Plans where Goal Minimization Fails

Proceedings of the AAAI Conference on Artificial Intelligence, 13

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 13

The Environment

Domains such as multiple trauma management, in which there are multiple interacting goals that change over time, are ones in which plan recognition’s standard inductive bias towards a single explanatory goal is inappropriate. In this paper we define and argue for an alternative bias based on identifying contextually "relevant" goals. We support this claim by showing how a complementary planning system in TraumAID 2.0, a decision-support system for the management of multiple trauma, allows us to define a four-level scale of relevance and therefore, of measurable deviations from relevance. This in turn allows definition of a bias towards relevance in the incremental recognition of physician plans by TraumAID’s critiquing interface, TraumaTIQ.