Future of Physics AI (2024)
Undergraduate: Ivan Fuentes
Faculty Advisor: Duane Deardorff
Department: Physics
Every student reaches a point in their learning where they begin to struggle and need guidance to proceed. Teachers and tutors help guide learners through confusion and mistakes. Today our teachers are human, but tomorrow we may use AI (Artificial Intelligence). The objective of our investigation is to see if ChatGPT 3.5T can tutor students in physics. To accomplish this task, we used an untrained model to create a baseline for what Large Language Models (LLMs) are capable of achieving. The AI tool was asked to produce step-by-step solutions for 25 problems from an algebra-based introductory physics course. The responses were graded on their ability to provide quality feedback based on the solution’s accuracy, clarity, approach, depth, and strategy. They were graded by two different reviewers, and the inter-rater reliability was calculated. The results from the investigation show that ChatGPT 3.5T has significant potential to function as a physics tutor as long as it is kept within its limitations. These limitations include the inability to analyze pictures and make numerical calculations. From this investigation, we conclude that ChatGPT 3.5T can provide meaningful guidance to students as they solve physics problems.
Link to Abstract