Oregon State University researchers say everyday use of generative AI is quietly wearing down the thinking habits that students need to thrive in high demand STEM fields. Their analysis ties routine reliance on AI to lower levels of reflection, a weaker drive to truly understand material and softer critical thinking. For classrooms already juggling new tools, tight curricula and employer expectations, it is one more pressure point.
The team published its results in a paper on arXiv, analyzing survey data from 299 STEM students at five North American universities and using structural equation modeling to test how trust and routine generative AI use affect cognitive engagement. Their model found that routine AI use significantly predicted lower scores on reflection, critical thinking and a student’s reported need for understanding, and they framed the pattern as a self-reinforcing "cognitive debt." The authors include Rudrajit Choudhuri, Christopher Sanchez, Margaret Burnett and Anita Sarma.
“The most surprising finding was that the ‘AI-savvy’ ones…are actually spinning themselves deeper into the AI dependence spiral,” lead author Rudrajit Choudhuri told KOIN. According to the researchers, students who routinely swap in AI for their own effort start to build weaker intellectual habits, which in turn makes them more likely to lean on tools instead of their own reasoning. That cycle, they warn, could leave graduates less prepared for complex problem-solving roles that require independent judgment.
What researchers mean by "cognitive debt"
The paper describes “cognitive debt” as a downward spiral where trust in AI fuels routine offloading, which then weakens habits like reflection, debugging and seeking conceptual understanding, making students more likely to rely on AI again the next time. Those mechanisms and the supporting statistics are detailed in the arXiv paper, which also finds that the students most at risk were, somewhat ironically, those with higher technical confidence and risk tolerance. The concern is not a sudden loss of ability but a gradual erosion of the productive struggles that typically build durable expertise.
How classrooms can respond
To push back on that trend, the authors urge concrete instructional changes: design assignments for exploration, such as storytelling, games or puzzles, alternate AI-assisted tasks with independent work, and require students to critique or debug AI outputs instead of simply accepting them. Those fixes are highlighted in the arXiv paper and in reporting by KOIN. The researchers also recommend that AI interfaces nudge users to justify their choices so that the tools augment thinking instead of quietly substituting for it.
Why Oregon colleges should pay attention
The debate is part of a broader national conversation. A recent faculty survey reported by Elon University found that large majorities of instructors worry about student overreliance on AI. For Oregon programs that train engineers and scientists, the risk is highly practical. Employers expect graduates who can think through unfamiliar problems without a machine doing the reasoning. The OSU authors say the next steps are to evaluate which classroom interventions actually restore reflection and critical thinking over time, rather than simply adding one more tech rule to the syllabus.
