Can AI Really Do Physics? Investigating AI Agents' Conceptual and Mathematical Reasoning Through Textbook Physics Problems
While advances in AI have resulted in models that can earn gold-medals in the International Mathematical Olympiad, how well can these agents do physics? Physics problems create a new challenge, as they require conceptual reasoning in addition to mathematical calculations. This study explores the problem-solving ability of agentic AI systems on textbook-style mechanics and electromagnetism physics problems. Different agents are compared to assess their strengths and weaknesses, allowing for implications of applications to broader engineering problems.