Satisficing Coordination and Social Welfare for Robotic Societies

The design of robotic systems that are capable of sophisticated social behavior such as cooperation, compromise, negotiation, and altruism, requires more complex mathematical models than are afforded by the classical mechanisms for making value judgments and decisions. A new concept of multi-agent satisficing, defined in terms of relative effectiveness and efficiency, is an alternative to classical optimization-based decision making. Conditional utilities, which take into account the interests of others as well as the self, represent an alternative to the categorical utilities of classical decision theory. A multi-agent utility aggregation structure is developed that avoids the sure subjugation of the interests of any individual to the interests of the group. By expressing a society as a directed acyclic graph, Bayesian network theory is applied to artificial societies. A satisficing social welfare function accounts for the influence relationships among decision-making agents.