Deep dive
Four readings
Chan (2023)
Higher EdA comprehensive AI policy education framework for university teaching and learning
International Journal of Educational Technology in Higher Education, 20, Article 38
Method
Mixed-methods survey · Hong Kong universities · 457 students, 180 teachers and staff
Central claim
Universities need an AI policy built from stakeholder input, with named owners.
Output
AI Ecological Education Policy Framework: 3 dimensions, 10 key areas
- ▸Data came within about four months of ChatGPT's release, framed as an academic-integrity crisis.
- ▸Low personal use (students 2.28, teachers 2.02) but high optimism; the strongest agreement: institutions need a plan for managing risks (4.50 / 4.54).
- ▸Teachers worry far more about over-reliance (4.24) than students do about themselves (3.11).
- ▸Neither group believes AI will replace teachers (2.14 / 2.26), and neither is confident teachers can detect AI use (3.02 / 2.72).
- ▸Ten key areas include misconduct, governance, monitoring, equity of access, attribution, AI literacy, rethinking assessment, balanced adoption, workplace readiness, and holistic competencies.
Pedagogical
Lead: Teachers
Rethinking assessment; holistic competencies; AI-driven workplace; balanced adoption
Governance
Lead: Senior management
Misconduct and ethics; privacy, transparency, accountability, security; attribution; equity of access
Operational
Lead: Teaching-and-learning and IT staff
Monitoring and evaluation; training and AI literacy
Dunnigan et al. (2023)
K-12“Can we just please slow it all down?” School leaders take on ChatGPT
TechTrends, 67, 878–884
Method
Column built on two focus groups (~3 hours) · 6 Michigan districts and 7 Pacific Northwest independent schools
Central claim
Leaders are deliberately waiting on policy and training; equity is barely on the radar.
Output
Four themes and an ecological “path forward”
- ▸“Generative AI doesn't come with a user manual.” Leaders wanted direction from universities and industry.
- ▸“Current acceptable use policies cover generative AI.” With one exception, no leader had an AI-specific policy. No one banned the tools: “It's a genie out of the bottle.”
- ▸“No immediate urgency for training teachers.” Post-COVID change fatigue: “Now, I gotta do AI?”
- ▸“Issues of equity and access are barely beginning to emerge.” Only one leader raised it; the authors warn subscription pricing is “opening a new chasm in the digital divide.”
- ▸Path forward (Zhao & Frank's ecological view): AI literacy as a new TPACK, updated teacher prep, educator standards, and state/federal guidance on copyright, privacy, and ethics.
Hughes et al. (2025)
Higher EdReimagining higher education: Navigating the challenges of generative AI adoption
Information Systems Frontiers
Method
29 interviews plus Interpretive Structural Modelling with 7 academic experts
Central claim
GenAI challenges are tightly interdependent, so piecemeal fixes fail.
Output
9 factors in a two-level model; 3 research propositions
- ▸Six base drivers (all max driving power): pace of GenAI development, reliance on GenAI for academic outputs, knowledge disparity, LLM training risks, output reliability, weak platform regulation.
- ▸Three outcomes at the top: lowering of academic standards, threat to the traditional academic role, validity of the HE degree.
- ▸Proposition 1: pace of development drives growing reliance in a self-reinforcing loop.
- ▸Proposition 2: interdependence demands holistic, connected strategies, not isolated fixes.
- ▸Proposition 3: reliance on GenAI could erode standards, critical thinking, and rigor.
Henriksen et al. (2026)
Teacher EdThe human side of human-AI creative collaboration: Expert perspectives on creativity, agency, and learning
Proceedings of SITE 2026, pp. 2091–2095
Method
Conceptual synthesis of 8 public-scholar interviews, 2023–2025
Central claim
Embodiment, struggle, foundational knowledge, and agency matter more with GenAI.
Output
4 dimensions of human creativity; 4 teacher-education implications
- ▸Embodiment and intentionality: “AI is not a weak form of human intelligence. It's an alien intelligence.” (Dede)
- ▸Process and struggle: “Struggle is not a barrier to learning—it is a condition of learning.”
- ▸Foundational knowledge: the Novice's Dilemma—“to drive it properly, you already need to know what you're doing.” (Furze)
- ▸Agency and voice: learners must be able to say, “That doesn't sound like me. I'm not going to use that.” (Beghetto)
- ▸Implication: “The teaching methods that are most invalidated by AI are the worst teaching methods.” (Mollick)
Cross-reading synthesis
Shared themes, reading by reading
| Theme | Chan (2023) | Dunnigan et al. (2023) | Hughes et al. (2025) | Henriksen et al. (2026) |
|---|---|---|---|---|
| Integrity & assessment | Clear rules, attribution, redesigned assessment | ELA teachers ask how to know work is authentic; leaders refuse to ban | Undetected misconduct is a named threat; strategy must be “multifaceted” | Redesign over policing |
| AI literacy & training | Key Area 6; owned by T&L and IT staff | No urgency; post-COVID fatigue; calls for new TPACK | Knowledge disparity is a top driver | Principled engagement over tool training |
| Equity & access | University buys tools so access is equal | Raised by one leader; “a new chasm in the digital divide” | Divide within cohorts and between countries | Framed as power: who shapes AI futures |
| Over-reliance | Teachers 4.24; students 3.11 | Not a focus | Reliance drives lower standards | “Cognitive deskilling is pretty insidious” |
| Who should act | Teachers, management, staff each own a dimension | The state, universities, industry | HE decision-makers and regulators | Teacher educators as “educator-designers” |
| Evidence base | Survey of 637 + open-ended | 2 focus groups, ~3 hours | 29 interviews + 7-expert ISM | 8 expert interviews |
Patterns
Converge · Diverge · Unanswered
Converge
A systems view
Chan calls her framework “ecological”; Dunnigan et al. borrow Zhao and Frank's ecosystem metaphor; Hughes et al. model every challenge as linked.
Complement, not replacement
Chan's respondents reject teacher replacement; Hughes et al. want “the human educator in the loop”; Henriksen et al. keep the locus of struggle human.
Detection is not enough
Neither students nor teachers trust detection (3.02 / 2.72). Both point toward assessment redesign.
Diverge
Opportunity vs. threat
Chan's respondents are optimistic; Hughes et al. report only threats by design; Henriksen et al. ask what is distinctly human.
Act now vs. wait
Chan's respondents want a plan now (4.50 / 4.54). Dunnigan et al.'s leaders deliberately wait.
Level of analysis
Institutional HE policy (Chan, Hughes), K-12 leadership (Dunnigan), pedagogy and human capacity (Henriksen).
Detectors vs. redesign
Chan lists detectors to procure for equal access; Henriksen et al., via Mollick, retire practices AI defeats.
Open questions
What happens to learning?
All four rely on perceptions or expert opinion. None measures student learning outcomes under AI use.
Where are students?
Chan surveys students but gives them no framework dimension. No reading gives students a role.
What works for equity?
Every reading names equity. None tests an intervention that narrows an access gap.