Norman Carver, Zarko Cvetanovic, Victor Lesser
In the functionally accurate, cooperative (FA/C) distributed problem solving paradigm, agents exchange tentative and partial results in order to converge on correct solutions. The key questions for FA/C problem solving are: how should cooperation among agents be structured and what capabilities are required in the agents to support the desired cooperation. To date, the FA/C paradigm has been explored with agents that did not have sophisticated evidential reasoning capabilities. We have implemented a new framework in which agents maintain explicit representations of the reasons why their hypotheses are uncertain and explicit representations of the state of the actions being taken to meet their goals. In this paper, we will show that agents with more sophisticated models of their evidence and their problem solving states can support the complex, dynamic interactions between agents that are necessary to fully implement the FA/C paradigm. Our framework makes it possible for agents to have directed dialogues among agents for distributed differential diagnosis, make use of a variety of problem solving methods in response to changing situations, transmit information at different levels of detail, and drive local and global problem solving using the notion of the global consistency of local solutions. These capabilities have not been part of previous implementations of the FA/C paradigm.