Why an NYU Vice Provost Says AI Has No Single Campus Answer

Colleges and universities should stop searching for a single answer to generative AI — and the institution's historic messiness is exactly the right structure for the problem. That is the argument of an essay by a vice provost at New York University published in The Chronicle of Higher Education, which casts AI in education as what planners call a 'wicked problem': undefinable, contested and without a stopping point.

The essay opens with a familiar campus scene: colleagues asking who is 'doing this right' with AI in the classroom and getting no confident answer. The author describes a majority of students using generative AI, with AI decoupling output from effort and research attaching names to the resulting threats — cognitive offloading, automation bias, the illusion of competence. Early hopes that students would voluntarily adopt good AI habits over lazy ones, the essay argues, have not been borne out.

One detail captures how much has already shifted: during last fall's midterms at NYU, the university almost ran out of blue books because far more professors than expected had quietly moved exams back onto paper and into the classroom. The author presents this as the realistic alternative to top-down policy, which he says cannot reliably change student behavior — citing weakened honor codes, students' refusal to report one another, and coordinated resistance ranging from campus anti-AI clubs to political opposition to data centers.

The essay's conclusion will disappoint anyone hoping for a master plan: no unified strategy is needed, and none is possible. Instead, students are owed clarity about what each professor expects for each assignment, spelled out in the syllabus, while faculty are asked to form opinions grounded in their own expertise rather than in the latest AI marketing.

Advertisement

Blue Books, Wicked Problems and Who Sets AI Rules

The 'Wicked Problem' Framework and Why It Matters

The essay's intellectual backbone is a 1973 planning paper that coined the term 'wicked problems,' applied to AI in education by Australian researcher Thomas Corbin and co-authors in a paper titled 'The wicked problem of AI and assessment.' A wicked problem cannot be rigorously defined, looks different from every constituency's vantage point and has no definitive answer. For universities, the pointed implication is that the failure to produce a uniform AI policy is not an institutional failure — it is inherent to the challenge. The author explicitly rejects the hope that renewed coherence is coming, including historian Nils Gilman's argument that large language models are acting as a 'catalytic solvent' exposing the incoherence that was always there.

What the Blue Book Run at NYU Really Signals

The near-depletion of NYU's blue book supply during last fall's midterms is the most concrete evidence in the essay, and it underlines a wider shift. Professors did not coordinate a return to handwritten exams; they individually concluded that in-class, on-paper assessment was the most reliable way to reconnect output with effort. The author is blunter about the pre-AI era: assignments often worked because cheating was inconvenient, expensive or easy to detect, and bringing some of that inconvenience back is one of the few responses that has shown immediate traction. The episode suggests assessment practice will keep diverging by discipline rather than converging on a campus-wide standard.

Where This Leaves Faculty, Students and Administrators

The prescription is deliberately modest. Faculty do not need to become AI experts, the author says, but they do need informed opinions and explicit syllabi. Students face a system that will remain inconsistent and should be told plainly what each course expects. Administrators should treat perpetual disagreement as normal rather than as a failure of leadership. The essay also warns that AI's tendency to amalgamate existing works without attribution collides with academic norms about intellectual ownership, and it recommends prioritizing tools that cite and restrict their sources over generic assistants.

The Longer Argument: Universities Versus Firms

The essay closes with a provocative contrast aimed at employers pressing campuses to be more dynamic: universities, the author claims, have better survival characteristics than firms. His example — 'a century from now, New York University will still be a going concern, but OpenAI will be gone' — is explicitly speculative, but it anchors a broader claim that higher education's tolerance of incompatible commitments, which looks like inefficiency to outsiders, has carried it through five centuries of predicted obsolescence, from movable type to the internet.

What Faculty and Administrators Should Do Without a Unified AI Policy

  • For faculty: state your AI expectations for each assignment in the syllabus — the author argues this clarity is the minimum students are owed, since no campus-wide policy can settle the question for every course.
  • For departments: treat in-class, handwritten exams as one legitimate option among several; NYU's experience shows professors will adopt them independently, so plan blue book supplies and grading capacity accordingly.
  • For administrators: abandon the search for a unified AI strategy and instead create room for bottom-up experiments — including honest, public discussion of what does not work, which the essay identifies as a particular campus weakness.
  • For technology decision-makers: favor AI tools that provide citations, let instructors specify which materials to search and which to avoid, and report their sources — the specific priorities the author argues align with academic norms.
  • For campus leaders: expect the conflict to persist; the essay predicts constituencies will remain dissatisfied with any approach, so plan for sustained, structured debate rather than a settled outcome.