System helps humans predict when self-driving cars will make mistakes
AI Summary: Researchers from MIT and Motional developed the Concept-Wrapper Network (CW-Net), a method that provides clear explanations of the decisions made by deep learning models controlling self-driving cars. CW-Net translates the internal reasoning process of these models into understandable concepts, such as "approaching stopped vehicle" or "close to cyclist," without altering the vehicle's driving performance. In road tests and simulation studies, CW-Net explanations helped safety drivers and nonexpert users more accurately predict vehicle behavior, and can provide important feedback for engineers troubleshooting in-vehicle artificial intelligence systems. This technique could boost the safety and transparency of autonomous vehicles while building trust in drivers and passengers.