Social Interactions under Uncertainty in Multi Agent Systems

Multi-agent Systems (MAS) deal with environments in which there are several agents that may interact. The field of multi-agent systems began its rapid advancement with the development of distributed, interconnected computer systems, such as the Internet and multi-robot teams. Such interconnected settings, where one agent interacts with another, involve studying interactions such as coordination, cooperation and collective decision making. Many researches in the field of multi-agent systems have investigated these social interactions, but with the assumption of complete information. However, agents must also be able to make good decisions in situations that involve a substantial degree of uncertainty. In our work, we provide the foundation for building such agents. Specifically, we have investigated the computational aspects of two common social interactions, collective decision making by voting and collaborative search. However, we use probabilistic models that shed a new light on these known settings. The first part of our research investigates computational aspects of voting procedures, with the presence of uncertainty. We begin by considering the winner determination problem, which is termed “evaluation” in the probabilistic knowledge setting. In the evaluation problem a probabilistic model of voter preferences and a particular voting rule are given and the probability of a particular candidate winning needs to be computed. We provide a polynomial algorithm to solve this evaluation problem when the number of candidates is a constant, and we present experimental results illustrating the algorithm’s performance in practice. However, when the number of candidates is not bounded, we prove that the problem becomes hard for many prominent voting rules. We further show that even evaluating whether a candidate has any chance of winning is hard in many cases, and we proffer an approximation algorithm for both problems. We then consider another probabilistic model, where only the probability