LectureAgentSocieties.pptVIP

Lecture 12: Agent Societies Two perspectives Coordination in MAS by norms and social laws (textbook, Ch. 9.6.4) Using MAS for simulation of social phenomena Coordination in MAS by norms and social laws Conventions, Norms: an established, expected pattern of behavior e.g. queuing in the end of the queue, using a certain language to communicate with others Social laws: norm associated with authority enforcing them and punishment in case of violation e.g. laws protecting private property Provide template behaviors which save reasoning power, and behavioral constraints that assure that agents have a chance to achieve their goals Conceptualizations of norms within game theory Norms as solutions to problems of co-ordination Norms as solutions of conflict of utility Norms as solutions to problems of inequality Two origins of norms and social laws in MAS Embedded in the design on each agent. Social laws and norms are designed offline and hardwired in each agent. Examples: Tennenholtz, Conte Castelfranchi Emerging from within the interactions of self-interested but cooperative agents. Examples: Shoham Tennenholtz Embedded norms and laws Engineering multi-agent organizations Usually hierarchical design: who reports to whom; predefined authority structures Modeled after human organizations Defines the flow of information / control and the agent interactions Much like Software Engineering or like Protocol design for negotiations Design of a social law: example One possible social law Each robot has to move constantly Direction of motion fixed on alternating rows Move-up when at right-most column Move-down when either On the left-most column of even rows On 2nd rightmost column of odd rows Emergent norms How can a norm or social law emerge in a society of agents? How can agents reach a global agreement by using only local information Each agent decides which convention to follow based only on its own experience? The T-shirt game (Shoham Tennenholtz, 1992): can all