Faculty Orientations Shape Adoption of AI in Research and Teaching

Synthesis: A mixed-methods survey of 90 STEM faculty in the RCSA Cottrell community identified a coherent latent construct — AI pedagogical orientation — that strongly predicts AI adoption across research, teaching, and professional activities. This orientation reflects deep beliefs about AI's role in disciplinary thinking, learning, and expertise development — not merely positive or negative sentiment.

Why This Matters

Current technology-adoption models (Scaffolding-like frameworks such as DoI and UTAUT) emphasize external conditions, perceived usefulness, and social influence. This study demonstrates that for AI in higher education, these factors are surprisingly weak predictors. Instead, a faculty member's epistemic interpretation of AI — their stance on what AI means for disciplinary knowledge production — is the primary driver.

Study Details

• Population: 90 STEM faculty (from 572 RCSA Cottrell awardees, 16% response rate), Fall 2025
• Method: Mixed-methods survey → Exploratory Factor Analysis (36 variables) + qualitative coding
• Key result: One dominant factor (AI pedagogical orientation, 9 items) consistently predicted AI use
• Weak predictors: Institutional initiatives, demographics (discipline, career stage, institution type), information sources
• Information flow: Department colleagues (64%) and discipline-specific news were top sources; institutional support mechanisms were less used

Connections to Knowledge Base

Faculty Development

This paper challenges the implicit theory behind many Educational Development programs: providing tools, workshops, and institutional support may be insufficient if faculty have not developed a coherent pedagogical orientation toward AI. The Cottrell community is already pedagogically motivated — yet 22% used no AI at all. Faculty development must help instructors articulate what AI means for their discipline, not just how to use it.

Teacher Adoption

The finding that concerns do not moderate adoption contrasts with AI Adoption Among Teachers: Insights on Concerns, Support, Confidence, and Attitudes, where institutional support → confidence → attitudes. This paper suggests a different mechanism: orientation shapes adoption directly, without being mediated by confidence or moderated by concerns. The Teaching literature may need to incorporate epistemic dimensions alongside attitudinal ones.

Institutional Change

The weak association between institutional initiatives and AI use is a cautionary note for A Framework for Institutional Change in the Age of AI and Change Management: top-down strategic plans and degree programs may have limited impact if they don't engage with faculty pedagogical orientations. Bottom-up, colleague-driven information flow (64% cited department colleagues) suggests peer networks are more influential than central initiatives.

AI Literacy

The orientation construct connects to AI Literacy at a deeper level: it's not just about knowing what AI can do, but having a coherent philosophy about what AI should do in one's discipline. This aligns with the distinction between instrumental and critical AI literacy.

Comparison to Computation Integration

The authors draw a deliberate parallel to the integration of computation into physics education — a decades-long process that succeeded only when faculty communities (e.g., PICUP) developed shared pedagogical frameworks. AI may follow a similar trajectory, where disciplinary communities — not institutions — drive adoption through shared epistemic norms.

Qualitative Dimensions of Orientation

Faculty views clustered around three questions:

1. Disciplinary thinking: Should students learn to think with AI, or without it?
2. Learning: Is AI a prosthetic that extends learning, or a replacement that short-circuits it?
3. Expertise development: What does expertise mean when AI performs tasks once central to mastery?

These are not resolvable by more information or better tools — they require disciplinary conversation and pedagogical judgment.

Open Questions

• How stable is AI pedagogical orientation over time? Does it change with experience?
• Are there discipline-specific differences in the content of orientations (physics vs. chemistry vs. biology)?
• Can faculty development interventions shift orientation, or is it a stable trait?
• How does orientation relate to actual classroom practice (not just self-reported use)?

What this means for practice

• Instructors. State where AI belongs in your discipline's thinking before adopting any tool: write down which tasks students should still do unaided and which AI can scaffold, then let that stance drive course decisions. AI pedagogical orientation — not sentiment, access, or institutional support — was the strongest predictor of use.
• Faculty developers. Replace tool-training workshops with disciplinary conversations about what AI means for knowledge production. Institutional initiatives showed weak association with use, while orientation strongly predicted research use (odds ratio 5.00), teaching use (4.26), and other professional use (2.80).
• Administrators. Invest in department-level peer networks rather than central initiatives alone: 64% of respondents named department colleagues as a top AI information source, and the 58 who reported institutional strategic initiatives still showed little relationship between those initiatives and their AI use.
• Instructors. Make verification a graded part of AI-supported work — one respondent had students "look for inaccuracies in the LLMs" — to counter the risk of what another called "overconfident incorrectness."
• Researchers. Treat orientation as a measurable construct rather than a proxy for enthusiasm: the nine-item AI pedagogical orientation factor predicted use across all three domains measured.

Limitations

• The sample is 90 STEM faculty drawn from 572 RCSA Cottrell awardees, a 16% response rate; the authors call the overall sample size modest and note the population is not representative of STEM faculty generally, since contemporary pedagogical practices are more common in this group than among faculty at large.
• All data are self-reported from a single Fall 2025 survey, so the authors cannot show that orientation actually shapes classroom practice and call for direct observation of AI-supported teaching and student reasoning.
• The results are a snapshot of a fast-moving technology: the authors note that AI tools and institutional initiatives have moved on since data collection, so the weak role of institutional context may not hold as adoption becomes more widespread and formally supported.
• The exploratory factor analysis ran 36 variables against n = 90, and the authors report poor model fit in initial solutions that required iterative removal of low-communality items before the single dominant factor stabilized.

Citation

Atherton, T. J., Descamps, I., Holmes, T. R., Vizcarra, C. L., Sui, N., Webel, M., & Foley, J. J., IV. (2026). Faculty orientations shape adoption of AI in research and teaching.