Higher Ed Must Not Let AI Write Its Own Argument

To the editor,

Ray Schroeder’s “What Do We Teach Now?” (April 1, 2026) asks an urgent question. But the column grants too much authority to the systems and companies that most need scrutiny. It moves from OpenAI’s GDPval benchmark to a Gemini-generated interpretation of “the reality so far in 2026” before turning again to Gemini for recommendations about what colleges should teach. Moving from a vendor benchmark to AI-generated analysis and curricular prescription is not critical inquiry. It is epistemic outsourcing: allowing the systems under scrutiny to narrate their own necessity.

The issue is not that the column takes AI seriously; higher education should take AI seriously. The issue is that it mistakes machine-generated prescription for human judgment and acceleration for destiny. When Gemini is asked both to describe present conditions and prescribe the curriculum, the column does more than report on AI’s rise; it lets the technology argue for its own centrality.

The column also quotes Gemini-generated figures about enterprise job redesign and the unemployment rate without identifying the sources or methods behind them. Numeric specificity lends such claims borrowed authority. At the point where the argument most needs source criticism and methodological transparency, readers are asked to accept machine-generated figures as if they were settled evidence.

Even on the benchmark’s own terms, GDPval is narrower than the column allows. As Schroeder notes, OpenAI presents GDPval as a benchmark for economically meaningful tasks, while acknowledging its limits and future iteration. A benchmark may inform debate; it cannot determine what institutions owe students, what labor should remain human or what losses are acceptable in the name of efficiency—especially when the column notes that automation’s harms will not be borne equally.

Higher education’s task is not simply to produce “AI-proof” graduates who can orchestrate tools and verify outputs. It must also ask harder human questions: What should remain human? What forms of judgment, care, interpretation and trust should not be optimized away? Teaching is not mere content delivery. Writing is not just text production. Advising is not routing. Librarianship is not retrieval. These are not marginal add-ons; they are practices through which students learn judgment, accountability and responsibility to others.

Colleges should teach students to interrogate AI, verify its claims and understand its limits. But they should also reserve the right to limit or refuse AI’s use where human judgment is part of the work itself: advising, feedback on student writing, research consultation and other relational forms of educational labor. The question is not only what we teach now but also what we choose not to automate away. Higher education must not let AI write the argument for its own inevitability.