When Good Agents Go Bad: The Fragility of Cooperation in Shared-Policy Learning

New research reveals that increasing exploration can unexpectedly undermine cooperative behavior in multi-agent systems using shared policies, even with well-designed learning algorithms.
