Multi-Objective Reinforcement Learning: A Tool for Pluralistic Alignment
Multi-objective Reinforcement Learning: A Tool for Pluralistic Alignment
AI summary
The paper reviews the role of multi-objective reinforcement learning (MORL) in building pluralistically aligned AI. It points out that when there are multiple conflicting values or stakeholders, reinforcement learning that relies only on scalar rewards may be difficult to align sufficiently; over the past decade, MORL centered on vector rewards has gradually become an alternative.