Learning in Multi-Agent Information Systems - A Survey from IS Perspective

Multiagent systems (MAS), long studied in artificial intelligence, have recently become popular in mainstream IS research. This resurgence in MAS research can be attributed to two phenomena: the spread of concurrent and distributed computing with the advent of the web; and a deeper integration of computing into organizations and the lives of people, which has led to increasing collaborations among large collections of interacting people and large groups of interacting machines. However, it is next to impossible to correctly and completely specify these systems a priori, especially in complex environments. The only feasible way of coping with this problem is to endow the agents with learning, i.e., an ability to improve their individual and/or system performance with time. Learning in MAS has therefore become one of the important areas of research within MAS. In this paper we present a survey of important contributions made by IS researchers to the field of learning in MAS, and present directions for future research in this area.