AI argues with intuition: Middle school students’ reasoning through AI-supported probability tasks

This study examines how students’ reasoning in probability tasks evolves through interaction with Artificial Intelligence (AI). Adopting an instrumental approach this qualitative study examines students’ pre-and post-test responses and AI-supported interactions to explore how AI supported changes in reasoning. The participants were 21 eighth-grade students from a public middle school, selected through convenience sampling. Data were collected through the prior knowledge test, probability tasks, and ChatGPT interaction transcripts. Prior knowledge responses were scored dichotomously, whereas task responses and interaction transcripts were analyzed using qualitative content analysis. Findings indicate that more than half of the students shifted toward more conceptually grounded reasoning. However, AI support was not consistently beneficial. While it often challenged intuitive assumptions through mathematically justified explanations, it sometimes reinforced misleading heuristics, particularly when students’ prompts shaped the interaction. This occasionally resulted in plausible but incorrect responses. Overall, AI support was shaped by students’ prompt framing and engagement with responses.