AI in Higher Education Newsletter
August 26, 2026 · Vol. 30
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
What Should Remain Human?
Bill Gates’s new essay begins with a blunt diagnosis: “There is no plan to ease the entry into the AI era.” His argument ranges across employment, national security, taxation, childhood, and international governance. For higher education, three parts matter most.
First, AI belongs in the category of technologies that require institutional governance, not merely individual rules. Second, the same tool that can expand access to knowledge may allow people to learn less, particularly when it substitutes for critical thought. Third, Gates proposes a domain he calls Human Reserved: work society deliberately keeps under human authority even when a machine could technically perform it.
The phrase is useful. It changes the question – what and how do we reserve for humans? Universities have spent three years asking where AI may be used. The harder question is now where a human must remain responsible, present, and answerable.
1. Gates Places Education Inside the Transition Problem
In The turbulent AI era is here. The choices we make are critical, Gates argues that AI will affect employment, education, taxation, energy, elections, public health, finance, law enforcement, and critical infrastructure at once. Existing agencies divide those responsibilities. AI crosses them.
That observation travels directly to universities. Campus AI policy is often distributed among academic affairs, information technology, libraries, research administration, legal counsel, career services, procurement, and individual instructors. Each office can manage its own risk while the institution still lacks a coherent account of what it is preparing students for.
Gates’s labor forecast is severe: entry-level and mid-level white-collar work will contract, and retraining alone will not absorb the transition. The timing and magnitude remain uncertain. Capability does not automatically become adoption; firms must redesign workflows, manage liability, and earn trust. But universities cannot wait for a settled forecast before reconsidering what a degree prepares a graduate to contribute.
Why it matters for Isenberg: AI policy, curriculum, and career preparation are becoming one problem. If entry-level analytical work changes, we need to identify the judgment, interpersonal authority, domain knowledge, and accountability that graduates must add beyond machine output.
2. “Human Reserved” Is a Design Principle, Not a Ban
Gates compares Human Reserved work to a nature reserve: society could build there, but chooses not to because the loss would be too great. He points to caregiving and consequential human encounters. In education and health care, his preferred arrangement is not prohibition. It is a mixed system with a human in charge, using AI to extend what the person can do.
For universities, this is more precise than saying education should remain human. Many tasks should not be reserved. AI can generate practice questions, translate instructions, summarize a meeting, test an explanation, or help a student rehearse. The reserved domain begins where relationship, evaluation, responsibility, or institutional legitimacy depends on a person standing behind a decision.
Likely examples include mentoring a student in difficulty, determining whether work demonstrates learning, giving consequential feedback, defending a curricular decision, and certifying that a graduate possesses a capability. AI may inform those activities. It should not become the unnamed authority inside them.
Why it matters for Isenberg: The useful boundary is not human-only versus AI-enabled. It is human authority versus automated authority. A faculty member may use tools while remaining responsible for the educational judgment.
3. The Critical-Thinking Claim Requires Care
Gates writes that the same tool that allows people to learn more could lead many to learn less. He cites a preliminary survey associating heavier AI use with lower critical thinking, with a stronger relationship among younger users.
That is a warning, not a causal conclusion. People who struggle with a task may use AI more; self-reported dependence is difficult to measure; and “AI use” covers activities ranging from asking for an answer to testing an argument. The educational variable is not exposure alone. It is what cognitive work the tool replaces and what cognitive work the student retains.
This distinction matters because an institutional response based on fear can be as intellectually thin as uncritical adoption. Requiring students to avoid AI does not by itself produce judgment. Requiring them to compare evidence, explain choices, test outputs, and defend a conclusion can.
Why it matters for Isenberg: We should evaluate the sequence of work. If AI enters before the student forms a question, reads the evidence, or attempts an analysis, it may replace learning. If it enters after an initial judgment, it can expose alternatives and improve revision.
4. A New Grading Study Shows Where Authority Still Matters
A study reported August 25 in Times Higher Education tested two versions of ChatGPT on 50 undergraduate bioscience essays, using seven criteria and four prompting conditions. The models did not reliably reproduce human judgment. In all but one condition, they assigned higher average marks than human graders. Lower-scoring essays tended to be inflated, stronger essays were sometimes marked down, and one individual AI-human difference reached 40 points on a 100-point scale.
The study is limited: one discipline, a modest sample, two model versions, and human marks that are themselves imperfect. It does not prove that AI can never assist assessment. It does show that consistency under repeated prompting is not the same as validity. A model can give a stable answer to the wrong evaluative question.
AI may help a faculty member check rubric coverage, locate passages for closer review, or draft low-stakes formative comments. Assigning a consequential grade is different. The decision affects progression, opportunity, and the meaning of the institution’s credential.
Why it matters for Isenberg: Grading extended, interpretive work should remain Human Reserved. Tools may support the process, but a named faculty member must exercise and own the judgment. Student work should not be submitted to an external model without appropriate institutional approval and consent.
5. AI Degrees Are Growing Faster Than the Field Can Stabilize
The Observer reports that at least 70 U.S. colleges now offer AI majors and at least 90 offer minors, compared with five institutions offering majors five years ago. The curricular problem is visible inside the growth: a course on current systems can age during the semester.
The strongest programs respond by emphasizing foundations that survive product cycles: mathematics, statistics, data, computing, evaluation, and the ability to learn a new system. That is also the answer for professional schools. Most business students do not need an AI major. They need durable disciplinary knowledge plus the ability to use changing tools without confusing fluent output with sound judgment.
Why it matters for Isenberg: We should teach current tools, but assess durable capabilities: framing a problem, understanding evidence, recognizing incentives, making a decision under uncertainty, and accepting responsibility for the result.
One Practical Move for Fall
For one course, assignment, or program outcome, create a Human-Reserved Map:
| Domain | Question | Example |
|---|---|---|
| Human authority | Which decisions require a named person to remain responsible? | Grades, consequential feedback, certification, exceptions, and academic-integrity findings. |
| Independent capability | What must a student demonstrate without AI assistance? | First-pass analysis, core concepts, oral defense, or live problem solving. |
| AI-extended work | Where may AI expand the student’s or instructor’s capability? | Practice, translation, comparison, revision, simulation, and low-stakes feedback. |
| Evidence and disclosure | What record makes the division of labor visible? | Process notes, source checks, version history, AI-use statement, or brief defense. |
The map should be small enough to use. Its purpose is not to reserve every familiar task for people. It is to identify where removing the person would change the meaning or legitimacy of the educational act.
What to Watch
Whether Human Reserved becomes a serious policy category or remains an evocative phrase. The difficult questions arrive immediately: who decides what is reserved, whether the boundary protects human dignity or merely preserves inefficient work, and whether institutions share productivity gains with the people carrying the remaining human burden.
For higher education, the near-term test is simpler. As universities adopt AI for instruction, advising, assessment, and administration, do they state clearly where a human remains in charge?
Sources cited: Bill Gates, “The turbulent AI era is here. The choices we make are critical,” Gates Notes, Aug. 26, 2026; Axios, “Bill Gates Wants to Keep Some Jobs Off-Limits to AI,” Aug. 26, 2026; The Guardian, “Bill Gates Calls for ‘Human-Reserved’ Jobs in Face of AI Takeover,” Aug. 26, 2026; Times Higher Education, “AI Tends to Mark Students’ Essays Higher Than Humans - Study,” Aug. 25, 2026; William Kay et al., “Can Generative Artificial Intelligence Mark Undergraduate Essays?”, Assessment & Evaluation in Higher Education, 2026; Observer, “As A.I. Majors Grow, Universities Race to Keep Curricula Relevant,” Aug. 2026.
Questions or topics for next week? Reply to mlangenkamp@umass.edu. Prepared in collaboration with Thea.