Impact of generative artificial intelligence on novices' problem categorization

It is an open question how generative artificial intelligence (genAI) impacts students' learning in higher education. GenAI tools' ability to do student-level tasks is rapidly evolving, and genAI adoption rates across young adults continues to rise. Recent studies suggest that learners who rely on genAI may bypass critical thought even if short-term task performance is enhanced. It is unclear if genAI can support the learning of expert-like behavior better than traditional approaches. Prior foundational cognitive science research on categorization of physics problems revealed experts' tendency to use principles, contrary to novices' tendency to use surface-level features. A seminal study by Docktor et al. 2012 found that an elaborate feedback intervention during a short similarity categorization task shifted physics novices' categorization strategies from surface features to principles without improving correctness of the categorization. This study extends the study design of Docktor et al. 2012 by providing new feedback conditions that use genAI tools. We analyze how access to genAI and guided prompting practices impact introductory physics students' categorization strategies, correctness, engagement time, and confidence.