AI in Higher Education Newsletter
August 18, 2026 · Vol. 29
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 a Degree Certifies
Generative AI did not invent weak assessment. It has made weak assessment impossible to ignore.
The distinction that matters is no longer AI versus no AI. It is assurance versus appearance. A polished product can look like learning without showing whose capability produced it. A credible degree must certify two things: that graduates have important capabilities of their own, and that they can use powerful tools without surrendering judgment to them.
This week’s news points toward a single settlement: protect occasions for independent thought, permit AI where it reflects professional practice, and design evidence that tells the two apart.
1. Unsupervised Assessment Has Reached Its Credibility Limit
A Policy Exchange report by Swansea University’s Philip Newton, drawing on Freedom of Information responses from UK universities, found that 78 percent of responding institutions use online remote exams for summative assessment and 70 percent plan to continue. Only 10 percent invigilate all such exams online, and 67 percent of relevant policies do not mention generative AI. Newton calls for remote unsupervised testing to be replaced with secure or in-person assessment.
As Vol. 24 reported, Brown University supplies the illustration. Economics professor Roberto Serrano set a difficult, unlimited-time take-home midterm; the class averaged 96 percent against a historical range of 65 to 80. After noticing answers that shared ChatGPT’s plausible-but-convoluted style, he moved the final into the classroom. Eighteen students withdrew and nine did not sit it. The 59 who did averaged 48.6 percent — a collapse that survived the attrition. Newton’s report gives us a reason to revisit the case: Brown’s anomaly now looks like an example of a wider assessment problem.
No single result proves how a student produced an earlier answer, and that is the institutional trap. Integrity procedures test individual allegations; this pattern is visible only in aggregate. Detection software does not close the gap. Redesigned evidence does.
None of this makes closed-book exams good assessment by default. Professional work is collaborative, iterative, and tool-assisted. But an outcome worth certifying should be demonstrated at least once under conditions where authorship is reasonably clear — through short oral defenses, live problem solving, staged submissions, or paired independent and assisted work. Any move toward timed or in-person formats must preserve approved accommodations.
Why it matters for Isenberg: Secure individual assessment belongs where independent fluency is essential: core concepts, quantitative foundations, first-pass analysis, and explaining a recommendation without machine support. AI-assisted assignments should name the different capability they assess.
2. Students Need to Know What Writing Is For
Research reported by Times Higher Education finds that students often treat writing as a product to submit. AI makes that reading rational: if the goal is fluent prose, a tool that produces fluent prose appears to solve the assignment. The same research describes students judging their own drafts against machine output and concluding the machine writes better — though polish can conceal shallow reasoning, invented support, or no position at all.
In management education, writing is a laboratory for judgment. A strategy memo forces a decision about which facts matter. A case analysis forces trade-offs among plausible interpretations. A recommendation forces commitment under uncertainty. Delegate those decisions before the student can make them and the assignment stays attractive while the education quietly vanishes.
Why it matters for Isenberg: Instructions should name the intellectual work, not merely the artifact. “Write a market-entry analysis” specifies a product. “Choose among competing explanations, expose the assumptions behind your recommendation, and defend the trade-offs” specifies a capability.
3. Responsible Use Has to Be Designed Into the Task
Alex Fenton and Andrew Firr at the University of Chester offer a working model. Students investigate a real campus process and propose a redesign. AI may help with planning, analytical questions, and prose; it may not invent observations, measurements, screenshots, citations, or a process the student never examined. Students submit a short statement covering the tool’s role, how outputs were checked, and which decisions were their own.
The boundary is operational: AI may help a student work around the evidence, but it may not manufacture the evidence. That is more useful than telling students to “use AI responsibly,” and it turns disclosure from a confession into evidence of a sound workflow.
Why it matters for Isenberg: AI may help formulate interview questions, organize notes, test an argument, or edit a draft. It may not fabricate a customer interview, financial figure, market observation, source, or consulting deliverable. Provenance is a clearer boundary than whether a chatbot was touched.
4. AI Literacy Is Becoming Core Preparation
The National University of Singapore will require all first-year students to take Applied Generative AI: From Prompting to Evaluation beginning in 2026-27, alongside expanded ChatGPT Edu access. The title gets the sequence right: prompting generates an output; evaluation determines whether it deserves to influence a decision.
A common AI foundation can reduce inequity and set shared expectations for privacy, disclosure, and verification. But a single course cannot carry the whole burden. Students must practice these judgments in the disciplines where the consequences become real.
Why it matters for Isenberg: Evaluating a marketing claim differs from evaluating a valuation, an employment recommendation, a supply-chain forecast, or a market-entry analysis. AI literacy becomes professional competence only inside the discipline.
One Practical Move for Fall
For one major assignment, complete this assurance map:
| Row | Question | What to assess |
|---|---|---|
| Independent capability | What must the student be able to do without AI? | Observe or test that capability under clear conditions. |
| Assisted capability | What may or should the student do with AI? | Assess the workflow, source checks, and quality of tool use. |
| Retained judgment | What decisions must remain the student’s own? | Require rationale, process evidence, or a defense of the result. |
Then align the conditions with the claim. If independent analysis matters, observe some independent analysis. If professional AI use matters, assess more than the polished output. This is a modest redesign but not a costless one — oral defenses and in-class work require faculty time, room planning, and accommodation support. Starting with a single assignment keeps the load proportionate.
What to Watch
Whether institutions move from broad AI policies to assessment-level assurance: which capabilities must be demonstrated independently, which may be tool-assisted, and what evidence supports each claim. And whether required AI-literacy courses stay generic or begin to connect with disciplinary assessment.
The durable question is no longer whether students used AI. It is whether the institution can say what they learned, what they can do, and how it knows.
Sources cited: Philip Newton, Evidence of Learning Through Assessment: Protecting the Value of a Degree in the Age of AI, Policy Exchange, Aug. 17, 2026; Times Higher Education, “Ban All Remote Unsupervised Tests ‘Immediately,’ Urges Report,” Aug. 2026; Johanna Alonso, “Brown Professor Suspects Majority of His Class Used AI to Cheat,” Inside Higher Ed, July 8, 2026; Times Higher Education, “Students ‘Confused About Assessment Purpose’ in Age of AI,” Aug. 2026; Alex Fenton and Andrew Firr, “Here’s How to Show Students What Responsible AI Use Looks Like,” Times Higher Education, Aug. 12, 2026; National University of Singapore, announcement of compulsory Applied Generative AI course and OpenAI collaboration, Aug. 11, 2026.
Questions or topics for next week? Reply to mlangenkamp@umass.edu. Prepared in collaboration with Thea.