Individual action and collective function: From sociology to multi-agent learning

Co-learning of multiple agents has been studied in co-learning settings, and how do they help, or many different disciplines under various guises. For hamper, learning and cooperation? example, the issue has been tackled by distributed • How do we characterize the process and the artificial intelligence, parallel and distributed comdynamics of co-learning, conceptually, matheputing, cognitive psychology, social psychology, matically, or computationally? game theory (and other areas of mathematical econ• how do social structures and relations interact omics), sociology, anthropology, and many other with co-learning of multiple agents? related disciplines. These studies are often disparate. Different disand so on. ciplines tend to ignore each other, although there has A key question, however, is the following. As been cross-disciplinary work, such as AI models and Adam Smith (1976) put it: cognitive studies using game theory (e.g., West & Lebiere, 2001), or sociological work incorporating He generally, indeed, neither intends to propsychological insights. mote the public interest, nor knows how much he We believe that interdisciplinary interaction and is promoting it . . . . He intends only his own gain, integration are important, and cross-disciplinary and he is led by an invisible hand to promote an communications can help to make better progress. end which was not part of his intention. Therefore, we want to take a close look at research on multi-agent learning, accentuating its interdiscipThis paradox have been troubling sociologists and linary nature. economists for many decades, and now computer Many questions concerning multi-agent learning scientists and psychologists as well. The issue may can be asked, in an interdisciplinary way: be formulated as the apparent gap between the individual intention in deciding his /her own action • How do agents learn to cooperate with each and the (possibly largely unintended) social function others, especially under bounded rationality? of his /her action. For example, how may self-inter• What is the minimum cognitive capacity necesested action benefit social welfare? Or, how may sary for an agent to learn to cooperate with cooperation be established through each individual others? maximizing his /her own gain (Axelrod, 1984)? As • What are the realistic cognitive constraints in Castelfranchi put it: ‘‘The real problem is modeling how we play our social roles, while being unaware of the functional effects of our actions, not only with E-mail addresses: rsun@cecs.missouri.edu (R. Sun), http: / /www.cecs.missouri.edu/ | rsun (R. Sun). our routine actions but even when doing something