Generative AI Availability, Grades, and Student Satisfaction at a Large University

URL SCAN: Generative AI Availability, Grades, and Student Satisfaction at a Large University
FIRST LINE: Computer Science > Computers and Society

The Dissection

This is a narrow institutional stress test, not a test of economic obsolescence. It examines whether GenAI availability after ChatGPT changed grades and student satisfaction more in courses classified as AI-susceptible. Its result is limited: effects are small, sensitive to COVID-modeling choices, and not detectably different by prior preparation. Instructors were already making courses less susceptible.

The paper measures the university’s visible dashboard while the underlying task architecture is being altered. Course redesign is adaptation. It may preserve grades and satisfaction by changing what students are asked to produce. That is not proof that AI failed; it may be evidence that the institution is hardening its assessments against substitution.

The Core Fallacy

The central error is treating the absence of detectable grade inflation or satisfaction decline as evidence against meaningful cognitive substitution.

The Discontinuity Thesis asks whether human cognition remains economically necessary. This paper asks whether one university’s grades and self-reported satisfaction visibly moved through 2025. Those are different variables. Stable grades can coexist with AI use, harder assessments, altered grading standards, or the relocation of human effort away from routine coursework.

The findings do not challenge P1, P2, or P3. They show, at most, that the academic system possesses a temporary lag defense: it can modify assessment formats before its metrics visibly collapse.

Hidden Assumptions

The analysis relies on several fragile assumptions:

• Course-level susceptibility extracted from syllabi accurately represents actual AI exposure and student use.
• ChatGPT’s release is a clean treatment boundary rather than one point in a rapidly changing capability curve.
• Grades, satisfaction, and the test-score-to-grade relationship are valid proxies for learning and future productive participation.
• “No detectable effect” means no meaningful effect rather than low power, offsetting effects, or heterogeneous effects.
• Instructor adaptation is cheap, scalable, and capable of preserving human-only work indefinitely.
• Results from one large university generalize across institutions, disciplines, and labor markets.

The unchanged relationship between pre-college scores and grades primarily shows that academic sorting remains intact. Credential sorting can survive after the economic value of the underlying cognitive work begins to decay.

Social Function

The result is best classified as a partial truth serving transition management. It gives universities a defensible near-term message: no measurable crisis, instructors are adapting, and student satisfaction has not collapsed.

That message becomes ideological anesthetic when inflated into “GenAI is not displacing cognition.” The study protects the legitimacy of the educational dashboard without answering whether the dashboard still measures scarce human capability. This describes the result’s social function, not necessarily the authors’ intent.

The Verdict

Useful as evidence that academic metrics had not visibly broken by 2025. Useless as a rebuttal to systemic obsolescence.

The paper finds a lag, not a reversal. The university may be keeping the credential carcass presentable while changing the work inside it. The decisive question—whether humans retain economically necessary cognitive labor—was not measured.