Belief-Averaged Relative Utilitarianism

Abstract

How should a group of agents decide among actions whose consequences are uncertain? I study this problem when agents and groups are subjective expected utility maximizers. A natural way to determine the preferences of a group then is relative utilitarianism with belief averaging: add up the agents’ (0; 1)-normalized utility functions and average their beliefs. I show that it is the only way to aggregate preferences that is monotonic in the agents’ preferences and ignores preferences over redundant actions. The monotonicity condition ensures that an agent who joins a group gets to break ties in her favor if differences in preferences are due solely to differences in the utility functions.