Georgia Institute of Technology College of Sciences

Multi-agent systems have been studied extensively through the lens of game theory. However, most game theoretic models make strong assumptions about agents accuracy of knowledge about their utility and the interactions of other players. We will show some progress at relaxing this assumption. In particular, we look at adversarial noise in specific potential games, and assess the effect of noise on the quality of outcomes. In some cases, very small noise can accumulate over the course of the dynamics and lead to much worse social welfare. We define the Price of Uncertainty to measure this, and we compute both upper and lower bounds on this quantity for particular games.