Chance Discovery Using Dialectical Argumentation

The authors propose the use of a dialectical argumentation formalism for chance discovery in domains where knowledge is distributed across a number of distinct knowledge-bases, as in a system of autonomous software agents. Each agent may have only a partial view of a problem, and may have insufficient knowledge to prove particular hypotheses; our formalism provides a means to aggregate across these partial views in a consistent manner. We identify a novel type of dialogue, which we call a discovery dialogue, and propose a formal model for its conduct. Moreover, we present locutions and rules for the implementation of these dialogues as dialogue-games. In exploring the question of whether this dialogue model may be automated, we consider a genetic algorithm to generate and test hypotheses in the space common to all the agents.