Legible Behavior

Legible behavior in artificial intelligence focuses on designing AI systems, particularly robots and language models, whose actions and communications are easily understandable and predictable by human observers. Current research investigates how to improve the legibility of both robot movements and language model outputs, employing techniques like reinforcement learning and policy regularization to optimize for clarity and avoid unintended interpretations. This work is crucial for safe and effective human-AI collaboration, addressing concerns about adversarial attacks on language models and enabling more natural and intuitive interactions between humans and robots in shared environments.