AI at Hollins

Chat-GPT Generated Analysis

On 10/30/25 I asked ChatGPT: "Who is writing clearly about the problem of introducing AI, and specifically Generative AI, into university curricula."

The system produced a reply that covers four 'elements' and four 'themes' with 11 example references.

Core questions the field is grappling with
Consensus insights / current best thinking
Leading authors and representative papers
Actionable recommendations for university leaders, faculty, and policy makers

Theme	Core Focus	Examples
Institutional Policy & Governance	How universities should design rules, policies and infrastructure for AI	Ref[11]; Ref[9]
Pedagogy & Faculty Development	How instructors can adapt teaching and learning to use LLMs responsibly	Ref[6]; Ref[3]
Assessment & Academic Integrity	How to redesign exams, writing tasks, and grading in an AI-enabled world	Ref[5]; Ref[4]
Equity, Ethics & AI Literacy	Ensuring access, fairness, and critical understanding of AI among all students	Ref[7]; Ref[12]

Core questions

How should universities balance safety (integrity, privacy, IP) with pedagogical innovation when adopting LLMs?
Who owns institutional data, model outputs, and fine-tuning artifacts? What procurement, risk, and vendor-governance rules are needed?
What level of local control vs. centralized policy is optimal (systemwide rules vs. unit-level flexibility)?

Consensus insights / current best thinking

Most institutions are moving away from blanket bans toward governed use: explicit permissions, required citation/attribution, and instructor-level rules embedded in course syllabi. [9]
Policy work must treat AI across three domains: (1) teaching & assessment, (2) research & IP, and (3) administrative services (HR, admissions, help desks). A single policy rarely fits all use-cases. [11]
Effective governance pairs policy with infrastructure (safe internal models, secure data environments) and training for legal/compliance teams. [9]

Leading authors & representative resources

Elia Rasky / CIGI — Generative AI Policy in Higher Education: A Preliminary Survey (CIGI, 2024). [9]
Leitgeb & Leitgeb — systematic review synthesizing governance pillars (2025). [11]

Actionable recommendations (for leaders & administrators)

Create an AI governance council with representation from academic affairs, legal, IT, IRB/ethics, and student body.
Publish clear, differentiated policies for teaching, research, and admin use — require course-level AI use declarations and syllabus language. [9]
Invest in secure on-premises / enterprise AI access (model hosting, privacy-preserving inference) for sensitive data and research.
Require vendor contract clauses on data use, provenance, and the right to audit model behavior.
Treat policy as iterative — pilot, evaluate, then scale; require reporting of harms/incidents.

Core questions

How can instructors use LLMs to improve learning (feedback, scaffolding, personalization) without letting students outsource cognition?
What professional development do faculty need to design LLM-informed assignments and teach prompt literacy?

Consensus insights / current best thinking

LLMs are most effective when explicitly scaffolded into learning activities (drafting + critique cycles, revision with AI feedback, tutor-style Q&A), rather than being an unsupervised aid. [6]
Faculty need hands-on PD that covers (a) capabilities & limitations of LLMs, (b) prompt design and evaluation, (c) ways to assess process and reasoning rather than final text. [6]

Leading authors & representative resources

Y. Qian — Pedagogical Applications of Generative AI in Higher Education: Systematic Review (2025). [6]
Practical case collections and AAU/discipline-specific guidance are emerging in 2024–2025 literature. [6]

Actionable recommendations (for faculty development)

Run short, example-driven workshops: “Prompting for pedagogy,” “Using LLMs for formative feedback,” and “Designing AI-aware rubrics.”
Adopt in-course AI contracts: students disclose AI use and reflect on how it shaped their work (process logs, annotated drafts).
Shift some assessment weight to process artifacts: drafts, annotated revisions, oral defenses, or reflective writeups that reveal student reasoning. [5]
Seed model-assisted labs/TA support — let faculty pilot LLM tools for grading suggestions and feedback but maintain human oversight.
Maintain a central catalogue of vetted AI teaching tools and exemplar assignments for faculty reuse.

Core questions

If LLMs can produce plausible answers, how should assessment design change to reliably measure learning?
Should institutions rely on detection tools, redesign assessments, supervised exams, or a blend?

Consensus insights / current best thinking

Detection alone is increasingly unreliable; the field recommends assessment redesign (authentic tasks, oral components, staged submissions) and stress-testing of assessments. Regulatory bodies in multiple countries now advise moving toward secure/varied assessments. The Guardian
Frameworks that help instructors decide whether to ban, permit with rules, or redesign (and how) are useful — and many institutions adopt mixed strategies per course. [6]

Leading authors & representative resources

Khlaif et al. — Redesigning Assessments for AI-Enhanced Learning: A Framework (MDPI, 2025). [5]
Policy surveys and higher-ed reports (CIGI, national regulators) documenting the move toward stress-testing and secure assessments. [9]

Actionable recommendations (assessment redesign playbook)

Map assessments to learning outcomes; retain some AI-free assessments that test independent demonstration of core skills.[8]
Use authentic, open-ended tasks requiring local context, reflection, or datasets students uniquely produce (reduces off-the-shelf LLM advantage).
Include orals / viva voce or live practicals for high-stakes evaluation.
Require submission of process artifacts (timestamped drafts, annotated AI outputs, prompt logs).
If using detection tools, pair them with manual review and transparency about limits; don’t treat them as definitive proof. [5]

Core questions

How do we ensure equitable access to generative AI affordances so adoption doesn’t widen achievement gaps?
How do we teach critical evaluation of AI outputs and address biases/harms from models?

Consensus insights / current best thinking

Early studies show demographic differences in perceived readiness and trust toward GenAI; institutions must offer universal AI literacy training to avoid unequal advantage. [6], [7]
Ethical guidance emphasizes transparency, rights around student data, avoiding vendor lock-in, and actively monitoring for model bias. [11]

Leading authors & representative resources

Actionable recommendations (equity & literacy)

Offer required, scaffolded AI literacy modules for students (how LLMs work, hallucination risks, citation practices).
Provide equitable access: campus-hosted LLM interfaces, subsidized compute credits, or library-supported AI services so students without premium tools aren’t disadvantaged.
Build mechanisms to monitor disparate impacts (analytics on who uses AI, outcomes by demographic) and adjust supports accordingly.
Clarify data-privacy rules (what student inputs may be shared with third-party models) and prefer private/enterprise models for sensitive data. [9]
Integrate ethics into curricula — require students to critique AI outputs and reflect on limitations and bias.