Trust-utility gap in introductory physics education: Students’ adoption, domain-specific skepticism, and preferences for AI integration

We report results from a mixed-methods survey study ( <a:math xmlns:a="http://www.w3.org/1998/Math/MathML" display="inline"> <a:mi>N</a:mi> <a:mo>=</a:mo> <a:mn>81</a:mn> </a:math> ) examining how undergraduate students in introductory physics courses use, trust, and prefer to integrate artificial intelligence (AI) tools into their learning. Among respondents, 91% (95% confidence interval (CI): [83%, 96%]) reported using AI for physics coursework, yet only 41% (95% CI: [30%, 52%]) trusted AI-generated physics explanations—a 50-percentage-point trust-utility gap consistent with domain-calibrated skepticism rather than uncritical adoption. Thematic analysis of open-ended responses ( <c:math xmlns:c="http://www.w3.org/1998/Math/MathML" display="inline"> <c:mi>n</c:mi> <c:mo>=</c:mo> <c:mn>67</c:mn> </c:math> ; Cohen’s <e:math xmlns:e="http://www.w3.org/1998/Math/MathML" display="inline"> <e:mi>κ</e:mi> <e:mo>=</e:mo> <e:mn>0.78</e:mn> </e:math> ) identified eight themes; the most distinctly physics-specific finding was that 40% of qualitative respondents spontaneously articulated AI fails —visual-spatial reasoning, circuit analysis, and abstract physical reasoning—aligning with physics education research on student difficulties with multirepresentational tasks. This skepticism is also consistent with Kortemeyer’s benchmarking results, which show that AI performance is weakest on precisely the multirepresentational task types that students identified as failure-prone, suggesting that student trust calibration may track underlying AI competence boundaries in physics. A majority (65%) preferred optional over mandatory AI integration. Because high-performing students were overrepresented by 13.9 percentage points, the estimated preference for optional integration (and/or resistance to mandatory) may be inflated in this sample. These findings highlight a critical unresolved priority—the verification gap between self-reported and actual verification behavior—and inform recommendations for AI-integrated physics pedagogy.