‘How can we check the evidence presented by GenAI’s response?’: students’ epistemic vigilance of their co-constructed scientific explanations with GenAI

This study describes how middle school science students enacted epistemic vigilance while co-constructing scientific explanations with generative artificial intelligence (GenAI). Using a qualitative research design, the study investigated the interactions of eight middle school students with Canva and ChatGPT during two science lessons on plant growth and the classification of fungi. To support their epistemic vigilance, two explicit epistemic scaffolds were implemented: the criteria for good scientific explanations as epistemic criteria and epistemic question prompts as reflective guides. Data including transcribed and translated video recordings of student-GenAI interactions, student-generated digital artifacts, and semi-structured interviews were analyzed using a multi-layered analytical framework. Findings indicate that students enacted epistemic vigilance through three complementary evaluative practices: evaluating the scientific quality of GenAI-generated explanations, verifying the credibility of information sources, and reflecting on their own understanding and assumptions while interpreting AI outputs. The epistemic criteria primarily scaffolded students' content evaluation while indirectly supporting source and receiver evaluation. Conversely, the epistemic question prompts explicitly scaffolded source and receiver evaluation while remaining grounded in evaluating their co-constructed scientific explanation quality. These findings suggest that within their cognitive levels, middle school students may engage in epistemic vigilance with explicit epistemic scaffolding. Limitations are further discussed.