How to design good learning agents in organization

摘要

This paper categorizes four types of multiagent learning in terms of both goals and evaluations in agents, and investigates the characteristics of each categorization to find an effective type for designing learning agents. Since the characteristics in this categorization are affected by the learning mechanisms of agents, the characteristics are investigated by referring to organizational learning in organization and management science as one of methods. Through intensive simulations on a complex domain problem, the following implication has been revealed: agents that pursue their own goals and are evaluated according to their total results show high performance in comparison with other types of agents when all learning mechanisms in the organizational learning are employed.