NestRL: A Nested Training Regime for Mutual Adaptation in Human-AI Teaming
Researchers have developed NestRL, a novel nested training regime designed to improve human-AI teaming by enabling mutual adaptation. This approach models human-AI interaction as an Interactive Partially Observable Markov Decision Process (I-POMDP) and trains agents against adaptive partners from lower levels of a nested hierarchy. NestRL aims to prevent agents from developing opaque, partner-specific strategies, leading to better generalization and adaptability with both AI and human teammates. AI
IMPACT Enhances AI adaptability in human-AI collaborations, potentially improving performance in complex interactive tasks.