The Incoherence Strategy: Why NYU is Abandoning the AI Policy Myth

The core idea here is that universities are struggling because they are trying to solve a wicked problem, the integration of AI, by seeking a single, rational policy. Clay Shirky, Vice Provost at NYU, argues that this approach is flawed. Instead of chasing one optimized AI strategy, institutions should embrace incoherence. By allowing departments and individual faculty to maintain conflicting approaches, universities can preserve the social and intellectual tension necessary for real learning. For leaders and educators, the advantage lies in resisting the urge to force uniformity, which often masks deeper systemic issues, and instead focusing on the social formation that AI threatens to dismantle.

The Hidden Cost of Fast Solutions

Early advice on AI integration, such as redesigning assignments to be more multimodal or process focused, failed because it treated AI as a tool to be managed rather than a fundamental shift in the educational landscape. Shirky notes that when faculty attempted these interventions, students simply opted out, choosing the path of least resistance when facing deadlines.

The systemic failure here is the belief that a spectrum exists where good AI use naturally displaces bad AI use. Shirky’s analysis reveals that these are two distinct, competing problems. A student can use AI for productive brainstorming while simultaneously using it to shortcut the emotionally difficult work of learning.

> I spent about a year and a half thinking that is one single problem... But I only assumed that moving them towards the good uses would also move them away from the bad uses. I see. Yeah. And it is not a spectrum. We have two different problems.
>
> -- Clay Shirky

Cognitive Surrender and the Erosion of Social Fabric

The most non obvious consequence of AI in the classroom is cognitive surrender, a state where students rely on AI even when they know it is actively degrading their learning. The downstream effect is not merely a loss of knowledge, but a weakening of the student baseline cognitive ability. When the tool is removed, performance drops below the level it would have been at had the tool never been introduced.

Even more critical is the emotional offloading occurring on campuses. Students are using AI to draft apologies to professors and to navigate group projects, effectively opting out of the human friction that defines the college experience. Shirky emphasizes that universities are not primarily information transfer businesses; they are sites of social formation. If students stop engaging with each other, the informal learning that happens in peer groups, the institution loses its primary value, regardless of how efficient the curriculum becomes.

The Goodhart’s Law Trap

The pressure to adopt AI heavy curricula is driven by an external demand for AI fluency, yet these efforts often fall into the trap of Goodhart’s Law: when a metric, like a grade or an AI certification, becomes a target, it ceases to be a good measure of quality. By pulling the anxiety of future employment into the present, universities have turned grades into high stakes targets, incentivizing students to use any means, including AI, to protect their GPA.

> The biggest single mistake we have made as an industry in the last generation was to start sharing GPA and transcripts with outsiders... a metric that becomes a target ceases being a good metric.
>
> -- Clay Shirky

Why Incoherence is a Competitive Advantage

Shirky suggests that the university historical strength has been its ability to host incompatible conversations. Forcing a uniform AI policy across a campus is not just impossible; it is an attempt to solve a problem that lacks a stopping rule. By maintaining incoherence, universities allow different disciplines to treat AI according to their specific pedagogical needs, some banning it to preserve the struggle of philosophy, others integrating it for technical production. This diversity of approach is not a bug; it is the only way to navigate a future where the businesses hiring these graduates cannot yet articulate what they actually need from AI.

Key Action Items

• Audit for Social Erosion: Shift focus from assignment redesign to social preservation. Over the next quarter, identify where peer to peer intellectual interaction is being replaced by AI and introduce in real life requirements, such as device free P set groups.
• Decouple Assessment from External Metrics: In the next 12 to 18 months, move toward internal only feedback loops for early stage coursework. Reducing the outside world visibility into granular GPA data can lower the anxiety that drives students to cheat.
• Abandon Prompt Engineering Curricula: Stop investing in AI fluency courses that focus on today specific tool interfaces. These will be obsolete within five years. Instead, double down on foundational critical thinking and decision making under pressure.
• Host, Don't Settle: Stop trying to mandate a single campus wide AI policy. In the next semester, create forums that explicitly host the conflict between pro AI and anti AI faculty. This discomfort is the necessary wicked problem work that creates long term institutional resilience.
• Prioritize Low Tech Resilience: Over the next year, encourage the return of blue book exams and oral assessments. While these favor quick thinkers, they provide a necessary baseline for verifying deep, independent cognitive development that AI cannot replicate.