"Reconsideration, Cooperation and Negotiation in Open Environments" Speaker: Omer Tripp, Computer Science, Hebrew University Date: Thursday, 1 March 2007 Time: 11am Place: Ross 201 Abstract: Adaptive behavior is an essential quality of any autonomous agent that operates in an inaccessible, dynamic environment. Such an agent must be sensitive to changes in the environment and update its policies accordingly. The BDI architecture addresses this need by introducing a reconsideration function, which allows the agent to modify its course of action following changes in the environment. However, in most instances of the architecture, the rules for triggering the reconsideration function are hardwired into the agent at design time. We present a learning algorithm that remedies this situation by allowing the agent to strike the balance between deliberation and action dynamically, and thus improve its performance considerably. We then extend our dynamic adjustment algorithm to the domain of inter-agent cooperation, and demonstrate its viability in the context of the framework developed by Goldman and Rosenschein for cooperation on the basis of state-changing rules. This leads naturally to a more general strategy for inter-agent cooperation and negotiation in open environments.