AI in Higher Education - September 25, 2026

September 25, 2026 · Vol. 34

A weekly brief for the Management Department, Isenberg School of Management, UMass Amherst. By Matthew D. Langenkamp / 雷邁德, prepared in collaboration with Thea 🪻✨.

← Back to archive · ← Back to Dictionary

Overview: Judgment in the AI University

Cornell’s new report on the future of the university offers a more useful starting point than another inventory of AI tools. Its proposed purpose is a “human-centered, integrated learning ecosystem to cultivate sound judgment.” That formulation puts the educational question first: what should students become capable of deciding, explaining and taking responsibility for, with or without AI?

Cornell, Dartmouth and UT Austin illustrate different aspects of that question. Cornell connects judgment to teaching, relationships and research. Dartmouth’s debate over its provost’s AI-assisted writing tests the relationship between institutional principles and personal practice. UT Austin is embedding its response in a shared undergraduate curriculum. These are not interchangeable initiatives; their stages of implementation differ. Together, however, they move the discussion beyond whether universities should provide access to chatbots.

Executive Summary

Cornell: A newly endorsed report makes sound judgment central to the university’s purpose. Its recommendations emphasize applied learning, human relationships and intentional choices about AI use and non-use.
Dartmouth: A dispute over the provost’s AI-assisted writing brings disclosure and accountability into focus. His response and the limits of detection tools matter to a fair account.
UT Austin: A proposed common core includes technically grounded AI education; prompting alone will not satisfy its criteria. Course pilots are planned for fall 2027.
Student attitudes: A Canadian survey conducted in July–August 2026 offers a recent, limited comparison concerning trust in AI grading.
Life sciences: Anthropic’s reported biology finding raises a brief but important question about hypothesis selection and experimental judgment.

1. Cornell: Judgment develops through practice

On September 22, Cornell released its faculty committee’s report and vision statement, following a year of consultation with more than 6,000 stakeholders. Provost Kavita Bala explicitly accepted and endorsed both documents. Bala said Cornell would develop a strategic plan to turn the recommendations into action; endorsement should not be confused with completed implementation. (1)

The report gives “judgment” practical content: weighing evidence, testing ideas, engaging with disagreement, and being willing to revise a conclusion. Its undergraduate recommendations make a pointed claim: “Sound judgment develops through practice, not transmission.” Students should demonstrate, build, argue and solve problems, rather than only produce content that a machine can generate. Seminars, laboratories, studios, capstones and field-based courses are presented as different settings for that practice. The report also argues for regular access to classes small enough for students to be personally known, particularly in the first two years. (2, pp. 47–49)

This is not a blanket argument against AI. The committee identifies a tension: using AI well requires judgment, yet excessive reliance on it may inhibit the development of that judgment. Its vision statement calls for faculty to make intentional decisions about both technology use and non-use in classrooms, laboratories and studios. The educational purpose should determine the tool, not the other way around. (2, pp. 47–48; 3, p. 8)

Cornell already offers a concrete example: its library and Center for Teaching Innovation have developed an approximately two-hour AI Critical Literacy Program. The university says all first-year undergraduates will participate early in fall 2026 through a course or a college-provided enrollment link. The four sections cover what AI is, the ethical questions it raises, its effects on learning, and how to create a personal AI-use policy. The program’s stated goal is informed choice about how—and whether—to use AI, not simply proficiency in prompting. It is a program offering and a participation requirement, not yet evidence of improved learning. (4)

Why it matters for Isenberg: The report suggests a useful test of a case assignment: can students explain why they selected particular evidence, defend a recommendation against an alternative, and revise it when the situation changes? These are possible applications of Cornell’s argument, not a new Cornell assessment mandate.

2. Dartmouth: Human judgment, disclosure and institutional credibility

Dartmouth is preparing an October 29–31 conference titled “AI at 70: What Must Remain Human?” Its August announcement asks where judgment should never be delegated and who remains accountable when intelligent systems fail. It also promises preliminary findings from a multiyear study of student AI use, learning, confidence and judgment. Those findings are forthcoming, not available survey evidence for this issue. (5)

This week, that institutional framing acquired an uncomfortable counterpart. The Dartmouth’s September 21 investigation examined Provost Santiago Schnell’s published work and reported AI-detector classifications that prompted questions about authorship and disclosure. Schnell acknowledged using AI for language refinement, copyediting, clarity and organization, but disputed the characterization of his Washington Post essay as entirely AI-written. He said he developed the ideas, evaluated evidence and took responsibility for the final text. The newspaper’s corrected account also notes that two reviewed academic works acknowledged AI assistance. (6)

A detector score does not establish how a manuscript was produced or adjudicate misconduct. The substantive issue is what authors owe readers when assistance ranges from proofreading to generating prose, and whether those distinctions are communicated consistently. Dartmouth’s existing undergraduate coursework policy, last revised in September 2025, disallows generative AI in submitted work unless expressly permitted and requires acknowledgment when it is used. Its sample statements include one for AI proofreading assistance. That policy applies to specified students and coursework; it does not, by itself, establish a rule governing a provost’s outside publications. (7)

Why it matters for Isenberg: The question extends beyond one disputed article: what disclosure standard makes intellectual responsibility visible, and how should expectations differ—or remain consistent—across student, faculty and administrative work?

3. UT Austin: AI literacy within a common intellectual foundation

UT Austin’s proposed core framework, announced in August and back in the news amid September debate over course eligibility, links AI preparation to a broad general education. Its planned 36-credit structure comprises arts and humanities, social sciences, and STEM, including a Computer Science & AI requirement. The task-force report says students should understand how algorithms and data shape systems, use computational tools for defined tasks, and examine privacy, bias, reliability and organizational consequences. (8, 9)

One criterion is particularly noteworthy: courses must include a strong technical component, and “Courses that merely prompt AI systems are not sufficient for inclusion.” The proposal combines hands-on work with explicit reasoning about correctness, efficiency and limitations. This is a more demanding definition of AI literacy than familiarity with a conversational interface. (9, p. 18)

The university’s current timeline calls for curriculum development and staffing in fall 2026, pilots beginning in fall 2027, and all new core courses available in fall 2028. The full transition date remains to be determined in coordination with university and state bodies. This is not an AI requirement already completed by every undergraduate. (10)

The broader redesign is contested, including which courses retain core eligibility and how much discretion faculty have over content. The substantive question is what every graduate should understand—and what does not qualify as sufficient preparation. The framework was announced in August; September reporting concerns its implications for existing courses. (11)

4. A recent student-attitude comparison: Canada, July–August 2026

KPMG Canada surveyed students and educators between July 22 and August 5, 2026. KPMG reports that 67% of surveyed students would question a result if their instructor used AI to grade their assignment. That is an attitude toward a specific institutional use, not proof of inaccurate grading. The survey covered 1,009 adult respondents across students and educators after screening 1,864 Canadians; its student population spans university, college, postgraduate or part-time study, high school and vocational education. The release does not provide the student-only sample size or the item-level base for this percentage. (12)

This is a limited international comparison, not a U.S. college estimate. It raises a useful question: do universities distinguish students’ use of AI from students’ trust in institutional AI use?

5. Life sciences: A brief test of the same principle

Anthropic reports that Claude helped identify an uncharacterized enzyme system in genomic data, followed by experiments performed by human scientists. Its biological function remains unknown. For life-science researchers, the potential lies in searching data and prioritizing hypotheses—not bypassing experimental validation. The connection to Cornell’s argument is judgment: as candidate explanations become easier to generate, how should researchers decide which deserve scarce laboratory time and what evidence would establish their value? (13)

Implications for Isenberg

These examples suggest three distinct responsibilities: cultivate judgment through practice, make AI assistance and human accountability legible, and define AI literacy more carefully than tool access. The important institutional measure is not how much AI appears in university life, but whether students and researchers become better able to explain, test and take responsibility for their decisions.

Sources