How preservice teachers evaluated the accuracy of ChatGPT-generated solutions of curriculum-based mathematical word problems in a self-directed context
Résumé
Introduction The increasing availability of ChatGPT and its capacity to generate solutions to mathematical problems has raised important questions for mathematics education, particularly regarding how preservice teachers engage in evaluating ChatGPT-generated mathematical solutions. Although prior research has examined the pedagogical use of generative AI in mathematics teaching and learning, less is known about how preservice mathematics teachers engage with evaluating the accuracy of AI-generated solutions to contextualised mathematical word problems under self-directed and unstandardised prompting. Methods Grounded in metacognition as a theoretical lens, this study adopts an exploratory, naturalistic design to examine how preservice teachers engaged with and evaluated ChatGPT-generated mathematical solutions. The study was conducted within a third-year mathematics methodology module in which AI tools were integrated into teaching, learning, and assessment. A qualitative document analysis was used to analyse 30 preservice teachers' written assignments, focusing on their evaluations of AI-generated solutions produced through individually constructed prompts in an uncontrolled interaction environment. Results The findings indicated that participants' evaluations were predominantly informed by metacognitive knowledge, particularly prior mathematical content knowledge, rather than metacognitive experiences involving real-time monitoring and regulation, which may be attributed to the task design of the problems involved in the study. However, variability in the AI-generated outputs, resulting from non-standardised prompting practices, meant that participants did not evaluate identical solutions, which constrains direct comparability across cases. Within these authentic engagement conditions, participants also tended to show higher levels of trust in ChatGPT-generated solutions when they were unable to independently solve the problems to correct the ChatGPT-generated solutions. Additional findings revealed evidence of both effective and ineffective application and regulation of metacognitive knowledge: some PSTs successfully applied conditional, declarative, and procedural knowledge to reject incorrect ChatGPT-generated solutions and justify accurate corrections, while others demonstrated weak regulation, superficial claims, or confirmation bias that led to perpetuated errors. Similarly, metacognitive experiences varied, with some students engaging in active monitoring and self-correction, while others failed to critically evaluate their own reasoning. Discussion This study contributes to research on the integration of generative AI in mathematics teacher education by offering exploratory insights into preservice teachers' metacognitive engagement with AI-generated mathematical solutions under naturalistic conditions. It highlights not only the need to develop evaluative and metacognitive competencies for AI-mediated environments but also the importance of strengthening both the application and regulation of metacognitive knowledge and experiences. In doing so, the study emphasises the methodological importance of distinguishing between controlled and self-directed AI interaction designs and contributes to understanding how metacognitive processes shape preservice teachers' ability to critically engage with AI-generated mathematical solutions.
Citer ce document
Accès au document
Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter
Voir l'article sur le site de la revueAuteur(s)
Statistiques
Consultations : 1
Téléchargements : 0