Forming opinions within a group of partially reliable agents: an architecture and a simulation study

Abstract. This paper reports the first results of a simulation experiment. There are two databases, one containing true propositions and the other containing their correspective negations. Five agents in turn access one of them. Each agent has a “capacity” that will be used as the frequency with which the agent accesses (unconscously) the database with the correct knowledge. Agents randomly exchange information with the others. Since they have limited degrees of capacity, their “cognitive state” quickly become inconsistent. Each agent is equipped with the same belief revision mechanism to detect and solve these contradictions. This embodies the Dempster-Shafer theory of evidence to evaluate the credibility of the various pieces of information and estimate the reliability of the other agents. The purpose of the experiments was that of evaluating on a statistical basis, the emergent cognitive behaviour of the group. For the moment we found that interaction increases the amount of correct information in each agent’s cognitive state and flattens their degree of correctness (ratio: correct information in the cognitive state/total information in the cognitive state).