Detecting Inappropriate AI Use in Physics Homework through Learning Behavior Patterns in Time–FOI Metrics

With the rapid raise of generative AI tools such as ChatGPT and Claude, concerns have grown regarding their inappropriate use in completing academic assignments. Directly detecting or verifying such misuse is inherently challenging—particularly for take-home assignments in quantitative disciplines like undergraduate physics—where asking students directly if they did so is neither practical nor reliable. In this talk, we introduce the Time-FOI metric as a novel and data-driven approach for identifying potential irregularities in student work patterns. By having students voluntarily record the time they spend and their self-perceived difficulty/stress level (“Freak-Out Index”) for each homework question, we construct Time–FOI diagrams that reveal distinctive statistical correlations characteristic of genuine human learning behaviors. Our preliminary results suggest that deviations from these patterns may indicate non-human behavior, including potential misuse of AI tools. Although this project began as a curiosity-driven exploration by a group of physicists and learning scientists, our findings provide a quantitative tool with broader potential for understanding authentic learning behaviors and student engagement in the era of generative AI.