AI shortcuts are hollowing out higher education

How Generative AI Is Undermining Learning and Teaching in Universities

There's a growing gap between what AI promises and what it delivers in higher education. The reality on the ground: generative AI often shortcuts learning, weakens critical thinking, and pulls students away from primary sources.

Calling AI a "life skill" misses the point. If students outsource reading, analysis, and writing to a tool that fabricates facts and smooths over nuance, their learning stalls.

Shortcuts Over Scholarship

Misuse is widespread. Many students run assessments through LLMs despite guidance, then polish the output just enough to pass.

The results are generic, dull, and frequently wrong. Example: after reading a short 1922 piece by Henry Ford, AI-assisted responses framed him as building a "sophisticated HR performance management function" and as a "transformational leader." That's a misread of both context and character.

Why LLMs Fail Core Academic Outcomes

• Weak reliability: hallucinations and confident errors pass as fact.
• Shallow synthesis: generic phrasing that blurs nuance and context.
• Atrophied skills: less reading of originals, less argumentation, less reflection.
• Source opacity: unverifiable citations or invented references.
• Creativity drain: templated ideas crowd out original thought.

Where AI Adds Little to No Value

In disciplines that depend on close reading, primary data, studio critique, fieldwork, or lab precision, LLM output often gets in the way. It produces filler where we need evidence, method, and judgment.

Practical Steps for Educators and Departments

• Recenter the process: require reading logs, annotated bibliographies with page numbers, proposal-to-draft-to-revision cycles, and brief viva-style defenses.
• Design AI-resistant tasks: use local data, primary texts, in-class writing, oral exams, studio critiques, and fieldwork reports tied to firsthand evidence.
• Demand verifiable sources: require citations with page or DOI, plus short notes on how each source informed the argument.
• Mandate disclosure: add an AI-use statement to each submission; collect versioned drafts to show development over time.
• Assess reading, not summaries: low-stakes quizzes on assigned texts, seminar leadership, and cold-call questions that probe understanding.
• Teach verification: show how to test claims against library databases and primary materials; explain why LLMs fabricate.
• Clarify policy and consequences: define allowed uses (if any), set expectations in rubrics, and align with integrity procedures.
• Protect formative work: use in-class analog tasks and timed responses to sample a student's authentic voice.
• Support staff: run workshops on assessment redesign and the limits of detection tools; share exemplars of AI-resistant assignments.

If You Allow AI, Contain It

• Limit use to brainstorming, outlining, or generating questions-never final text.
• Require side-by-side evidence of verification against primary sources.
• Grade the thinking: rationale, method, evidence, and revision notes carry weight.

Policy and Culture: Choose Rigor Over Hype

Convenience has a cost. If we accept mediocre, unverifiable text as learning, we lower the bar for scholarship and professional readiness.

Be sceptical. Keep primary texts, data, and original analysis at the center. Use tools where they help, but guard the core skills that define higher education.

Further Reading and Faculty Support

• UNESCO guidance on generative AI in education
• Jisc: Generative AI guidance for UK higher education
• Faculty-focused AI training by job function