On the value of information in multi-agent decision theory
Abstract
Several agents, each with his own opinion (probability measure), make decisions in order to maximize their expected utility. A super partes person, ‘the chief', releases information with the goal of maximizing a social expected utility, which is an increasing function of the agents' utilities. Additional bits of information are individually beneficial to each agent, but might be socially detrimental if the social utility is concave and therefore diversification is valuable. In this paper, information is modeled by filtrations on a suitable probability space, and the problem of establishing how much information ought to be released is tackled. Two situations are examined, in which the chief either updates or does not update her opinion.