Estimating Central, Peripheral, and Temporal Visual Contributions to Human Decision Making in Atari Games
Researchers have developed a new framework to analyze how different visual cues influence human decision-making in Atari games. By training action-prediction networks on a dataset with synchronized eye-tracking, they found that peripheral vision contributes significantly more than direct gaze or past game states. The study also identified distinct behavioral patterns, such as focus-dominated or periphery-dominated decision-making, based on the information sources utilized. AI
IMPACT Provides insights into human visual processing and decision-making, potentially informing AI agent design.