From the Student Success edition of October 25, 2026
UC Berkeley Fields medalist stops grading homework for correctness because of student AI use; campus judges AI detectors inadequate to standardize
University of California, Berkeley (Department of Mathematics; Office of Undergraduate Education) · Higher education assessment design and academic integrity · Berkeley, California, United States
- Publisher
- The Daily Californian
- Original publication
- The Daily Californian, 23 October 2026 (picked up by the New York Post on 25 October 2026)
- Source retrieved
- 2026-10-26
What happened
Richard Borcherds, a UC Berkeley mathematics professor and Fields medalist, told the Daily Californian that after seeing many students use AI to consistently receive top homework scores he saw "no point" in paying readers to grade homework for correctness. He now grades homework on completion, which he called "not a very good system", saying he has no alternative. An assistant vice provost for undergraduate education said the Academic Senate asks faculty to include AI-use expectations in syllabi, that policies vary by course, and that the campus has tried AI detection tools including Turnitin and Rumi but judged them not adequate to standardize. Course AI policy remains at faculty discretion, supported by sample statements in the campus AI Hub.
Why it matters
Shows a concrete assessment consequence of student AI use: graded homework, a common formative practice and early-warning signal, losing its value as a measure in at least one course, while detection is not seen as a campus-wide fix.
Evidence and measured results
One faculty interview plus an administrator email statement in student reporting; no data on grade distributions, homework scores or student outcomes.
Limitations and uncertainty
Single instructor and discipline; reported claims, not measured results. A New York Post re-report (25 Oct) was not inspected in full and is not relied on.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-26; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Student AI use makes graded homework an unreliable measure of learning, and detectors are not trusted as a standard control.
- Stakeholders
- Vice provost for undergraduate education
- Academic Senate teaching committee
- Department chairs in gateway STEM courses
- Learning analytics or institutional research
- Discovery questions
- Which courses still grade homework for correctness, and how has score distribution changed?
- Do early-alert models use homework grades as inputs?
- What replaced detection as the campus integrity approach?
- Value hypothesis
- Interpretation: redesigning formative assessment and checking analytics inputs could preserve reliable early signals of student difficulty.
- Potential engagement
- An assessment and early-alert input audit across a small set of gateway courses.
- Unsupported claims
- That AI use has lowered learning outcomes at Berkeley
- That completion grading improves or harms results
- That any detector is reliable enough to standardize
Pre-sales engineering
Role takeaway
- Fit
- Institutions running early-alert or learning-analytics models that use gradebook homework scores.
- Architecture and integration
- Gradebook and LMS data feeds into early-alert models; grading-policy metadata would need to be captured alongside scores.
- Prerequisites
- Inventory of model features drawn from homework grades
- Course-level grading-scheme metadata
- Historical score distributions
- Constraints
- Single-instructor evidence
- Policy varies course by course
- No outcome data in the source
- Security
- Keep any detector or third-party analysis of student work out of scope unless privacy review approves de-identified submission.
- Proof of value
- Measure feature drift in homework-score predictors
- Compare early-alert accuracy before and after grading changes
- Test alternative signals such as in-class checks
Delivery
Role takeaway
- Work
- Help departments redesign formative assessment where AI has weakened homework grading, and flag changes for analytics teams.
- Dependencies
- Faculty willingness to change grading
- Academic Senate syllabus guidance
- Analytics team capacity
- Ownership
- Office of undergraduate education with department chairs.
- Skills and adoption
- Faculty development on in-class and process-based assessment; analytics staff trained to handle grading-scheme changes.
- Governance checkpoints
- Syllabus statement review each term
- Analytics model review after grading changes
- Student feedback review
- Acceptance criteria
- Proposed: every gateway course records its homework grading scheme
- Proposed: early-alert model accuracy is re-validated after scheme changes
- Proposed: at least one reliable in-class formative measure per course
- Risks
- Loss of feedback for students who need it
- Silent degradation of early-alert models
- Inconsistent policies across sections
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
If homework scores no longer measure learning, early-alert models and dashboards that use homework grades as an input may lose signal.
Governance
Who approves, reviews and stays accountable for outcomes?
Policy is set course by course via syllabus statements; the campus declined to standardize AI detectors.
Security and privacy
What data, permissions and controls need testing?
Not discussed in the source; detector use generally raises questions about submitting student work to third parties.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Completion grading reduces paid grader work; impact on students who relied on homework feedback is not discussed.
Procurement
What should contracts, pricing and exit terms secure?
Campus testing of Turnitin and Rumi did not lead to a standard detector, a caution against detector procurement as the main integrity control.
Operating model
Which teams own the service once it runs?
Formative assessment and feedback may need redesign (for example, in-class checks) rather than more detection.
Publication history
- 2026-10-25Student Success · Issue 465 resources
Stable resource ID: dailycal-uc-berkeley-borcherds-completion-grading-homework-ai-detection-tools-20261023