Team-solvability: a model-theoretic perspective
Alessandro Agostini · 2000
At present, the extension of formal learning theory to the multi-agent case considers "teams" of agents sharing a common end. Success is achieved if one or more of the agents is successful, and cooperation is not involved in the team formation. Unfortunately, this is rarely the idea of "successful team" we have in mind. One generally expects agents' behavior to influence each other in a way that is not captured by the present paradigms. A real problem in extending single agent learning methods to multi-agent setting is thus determining paradigms of cooperation. This paper makes a contribution to the solution of this problem. First, we advance a paradigm of cooperation as a kind of two-person repeated game and compare it to a major paradigm of solvability for isolated agents. Second, we pay attention to a subset of unsuccessful agents who take advantage from teamwork. For these agents, cooperation is proved to be a key of success. The formal results are raised within the model-theoretic tradition of formal learning theory. 1