Multi-Principal Assistance Games: Definition and Collegial Mechanisms
This work addresses the problem of aggregating diverse human preferences for an assisting AI agent, which is significant for designing ethical and effective AI systems that serve multiple stakeholders.
The paper defines multi-principal assistance games (MPAGs) where a single agent assists N human principals with diverse preferences. It shows that if humans are responsible for a sufficient fraction of rewards through their own actions, their preferences are straightforwardly revealed, circumventing Gibbard's theorem.
We introduce the concept of a multi-principal assistance game (MPAG), and circumvent an obstacle in social choice theory, Gibbard's theorem, by using a sufficiently collegial preference inference mechanism. In an MPAG, a single agent assists N human principals who may have widely different preferences. MPAGs generalize assistance games, also known as cooperative inverse reinforcement learning games. We analyze in particular a generalization of apprenticeship learning in which the humans first perform some work to obtain utility and demonstrate their preferences, and then the robot acts to further maximize the sum of human payoffs. We show in this setting that if the game is sufficiently collegial, i.e. if the humans are responsible for obtaining a sufficient fraction of the rewards through their own actions, then their preferences are straightforwardly revealed through their work. This revelation mechanism is non-dictatorial, does not limit the possible outcomes to two alternatives, and is dominant-strategy incentive-compatible.