LEARNING IN MULTI-AGENT COMMUNITIES

ABSTRACT: Our understanding of humans and especially human learning is still very limited; see e.g. http://www.gwu.edu/~tip/theories.html for 50 theories of human learning; and from other sources some others can be cited. Traditional computer science learning studies (Valiant etc) provide little help for understanding human learning. However, it seems that development of WWW provides also a phenomena, which can be called learning and sometimes it has been even argued, that WWW (with suitable interface) could soon pass the Turing test for intelligence. In multi-agent communities can occur improvement of collective behaviour (collective learning) even if individual members of the community are rather limited (e.g. have very limited memory) - it is sufficient that they behave just "a bit better" than totally randomly (see e.g. Norman L. Johnson. Collective Problem Solving: Functionality Beyond the Individual, http://ishi.lanl.gov/Documents1.html) and these processes can be modelled and studied.