Accenture’s 2024 TechVision report features our recent work on learning from facial reactions. “Similarly, a 2023 research paper from Accenture and Cornell University describes a way human-robot collaborations can benefit when robots can identify when they’ve made an error based on implicit reactions from humans they interact with – much like how people use social cues to recognize their own mistakes. The authors built a dataset of bystander responses to human and robot errors and used it as input to a deep-learning model to predict failures. By creating these systems that are sensitive to human social signals, they are efficiently using the expertise of human perception and action as a marker for mitigating robotic errors.”