Inside the Mental Models: Instructor’s Conceptions on Generative Artificial Intelligence
Plain Language Summary
We look at how engineering instructors think about generative AI (GAI) tools like ChatGPT, and specifically at the moments in interviews where an instructor realizes a scenario doesn't fit what they already believed about the technology. We call these "aha" moments. Drawing on interviews with 169 instructors across 18 universities, we show that these realization moments cluster around a few recurring topics, and that they offer a window into where instructors' thoughts about AI remain unsettled.
Contribution
Rather than surveying instructors' GAI policies or attitudes directly, we operationalize belief change itself by identifying spoken "aha" moments in interview transcripts via a targeted cue-word screening procedure, then interpret their content through the Reasoned Action Approach (RAA), offering a method for surfacing where instructors' mental models of GAI are actively being revised rather than only where they currently stand.
Research Questions
- How, and around what topics, do engineering instructors' mental models of GAI become visibly challenged during interviews?
- How do the behavioral, normative, and control beliefs described in the Reasoned Action Approach help explain these moments of realization?
Methods
As part of a larger NSF CAREER project, we conducted semi-structured interviews with 169 engineering instructors (118 undergraduate-level, 51 graduate-level) across 18 universities selected by size of undergraduate degrees conferred, between January and December 2025. We screened transcripts for cue words (e.g., "haven't," "hadn't," "thought," "before," "yet") to locate "aha" moments, then read the surrounding context to confirm each cue reflected a genuine knowledge gap or insight, rather than treating the cue words themselves as sufficient evidence. We then read the content of these moments against the three belief components of the Reasoned Action Approach: behavioral, normative, and control beliefs.
Key Findings
32% of instructors produced at least one identifiable "aha" moment, most commonly in response to three protocol questions: whether they would use GAI to predict student performance, whether they would use it to provide feedback, and how they had used GAI tools in their own teaching. Instructors' behavioral beliefs ranged from cautious optimism about GAI as a starting point ("not something you're going to cite in a research paper") to firm rejection on grounds that assessment requires a customized, instructor-driven process; normative beliefs were shaped by colleague conversations and, in at least one case, an industry advisory board whose input did not move the instructor's stated intention. Perceived behavioral control varied with institutional access to GAI tools, though we note that access and technical literacy alone did not appear sufficient to drive adoption, since time and interest (particularly among, as one participant put it, "older faculty") also mattered.
Implications
Because "aha" moments can either destabilize or reaffirm an instructor's existing mental model, we suggest they mark a productive site for reflection rather than a marker of confusion alone, and that engaging with them intentionally could support more grounded GAI integration in engineering education. Our stated long-term goal is a reflective toolkit that surfaces these edge cases without the time burden of a full interview protocol, giving instructors a more scalable way to work through the conceptual tensions GAI introduces. We frame this as an argument for designing faculty development around belief-revision moments rather than static policy statements alone.