Policy Learning with a Language Bottleneck

arXiv:2405.04118v1 Announce Type: new
Abstract: Modern AI systems such as self-driving cars and game-playing agents achieve superhuman performance, but often lack human-like features such as generalization, interpretability and human inter-operability. Inspired by the rich interactions between language and decision-making in humans, we introduce Policy Learning with a Language Bottleneck (PLLB), a framework enabling AI agents to generate linguistic rules that capture the strategies underlying their most rewarding behaviors. PLLB alternates between a rule generation step guided by language models, and …

abstract agents ai systems arxiv cars cs.ai cs.cl cs.lg decision driving enabling features framework game human human-like humans interactions interpretability language making modern modern ai operability performance playing policy self-driving superhuman systems type