HN Debrief

Professor denounces mass AI fraud on an exam at Brown

  • Education
  • AI
  • Hiring
  • Developer Tools

The article says a Brown professor teaching an economics course switched to take-home, closed-book exams and concluded that mass AI cheating followed. He points to a sharp gap between the midterm and an in-person final as evidence, and says the episode threatens academic integrity at elite universities. Commenters largely accepted the core premise that AI has made cheating easier, but they did not treat Brown as the interesting part. The stronger read was that a take-home closed-book exam was already an unstable format before large language models, and AI just turned a bad incentive structure into a rout.

If you run training, hiring, or certification, stop assuming remote unsupervised assessments still measure individual ability. Move foundational evaluation toward proctored, oral, or process-verified formats, and treat any credential earned through easy-to-game workflows as a weaker signal.

Discussion mood

Mostly alarmed and unsentimental. People saw the professor's complaint as real, but also self-inflicted. The dominant mood was that AI cheating is widespread, expected, and now powerful enough that any remote trust-based exam format should be presumed compromised.

Key insights

  1. 01

    Design courses as adversarial systems

    The useful framing is to treat assessment the way you would treat a gameable product metric. Students optimize for grade per unit effort, so course design has to make the cheapest winning strategy pass through the learning objectives. The concrete tactics were paper exams and short one-on-one interviews about submitted code, where students must explain library calls, design choices, and failed approaches. That does not force original work every time, but it does force ownership of the work.

    Audit your own evaluation systems for the shortest path to a high score. If that path bypasses understanding, change the workflow rather than relying on norms or detection.

      Attribution:
    • bkallus #1
  2. 02

    High-frequency controlled testing already exists

    There is already an institutional model for computer-based assessment that does not depend on student-owned devices or honor-code assumptions. Illinois was cited as running weekly computer-based quizzes worth most of the grade, backed by dedicated testing infrastructure. The point is not nostalgia for blue books. It is that reliable digital assessment needs physical operations, staffing, and budget, not just software.

    If you want secure digital exams, plan for facilities and operations, not browser plugins. A real testing center is a product and infrastructure decision, not a policy memo.

      Attribution:
    • gchallen #1
    • recursivedoubts #1
  3. 03

    Handwriting is not a universal fix

    Paper exams stop one cheating vector, but they also create accessibility problems for students who cannot write for long periods, rely on screen readers, or need typed output for legibility. Several people described existing accommodation setups that already use locked-down university machines as a parallel path. That makes the stronger position clear: proctored in-person assessment is the requirement, while handwriting is only one implementation and often the wrong one.

    Separate the security goal from the medium. Require supervised individual work, then offer paper or locked-down institutional devices depending on the task and the student.

      Attribution:
    • tialaramex #1
    • vcf #1
    • glimshe #1
    • Waterluvian #1
  4. 04

    Curves and credential pressure fuel cheating

    The strongest structural argument was that cheating rises when students believe they are competing against each other for scarce outcomes. Curved grading, crowded hiring funnels, and degree inflation make integrity feel like unilateral disarmament. One former teacher said the result now looks like an arms race in advanced high school and college STEM classes. In that environment, moral appeals land weakly because the system keeps telling students the credential matters more than the learning.

    If you control evaluation, avoid forced ranking when absolute mastery standards will do. If you hire, assume a rising share of top-line academic signals were earned under distorted incentives and verify fundamentals directly.

      Attribution:
    • pants2 #1
    • rawgabbit #1
    • rsa4046 #1
    • shermantanktop #1
  5. 05

    AI-friendly assessment works only after fundamentals

    Some argued schools should stop fighting the tool and let students use AI freely, then grade the quality and correctness of the result. That idea was not dismissed outright. It was narrowed. People thought AI-assisted assessment can make sense for creative, project, or workplace-like tasks, but not for foundational courses where the institution is supposed to certify unaided competence first. The distinction was between learning the tool and outsourcing the skill.

    Split curricula into 'must do alone' and 'may do with AI' segments. Make the boundary explicit so students and employers know what the credential actually certifies.

      Attribution:
    • tmsh #1
    • aneesh #1
    • hackermailman #1

Against the grain

  1. 01

    One-on-one oral exams could scale more than assumed

    The standard objection to oral evaluation is staffing cost, but one commenter challenged the premise. If a class already demands well over 100 hours of student time, spending 20 to 30 minutes per student on a final competency check may be justified, especially in smaller courses. That does not make oral exams universal, but it weakens the reflexive claim that they are impossible.

    For courses where independent competence really matters, price oral checks against the cost of unreliable grading. A short viva may be cheaper than rebuilding trust after a compromised course.

      Attribution:
    • Lerc #1
    • JoshTriplett #1
    • panicinducer #1
  2. 02

    Grades still serve internal academic functions

    Against the 'grades are just free screening for employers' line, others pointed out that grades still gate progress through sequenced courses and allocate scarce seats, sections, and instructors. Even if employers overuse them, universities still need some mechanism to decide who is ready for Calc II or who gets one of a limited number of spots. That does not defend current grading practices, but it does block the idea that grading has become pointless.

    If you want to reduce the weight of grades, replace each function they serve one by one. You still need progression checks and seat allocation rules even in a less grade-centric system.

      Attribution:
    • nlawalker #1
    • shepherdjerred #1
    • voxl #1
    • catlikesshrimp #1
  3. 03

    The article already rules out pure paper exams

    A few people pushed back on the thread's rush to 'just handwrite everything' by noting the article itself describes why that specific professor cannot simply do that for everyone. Brown already uses accommodation infrastructure for students who need alternatives, and commenters with similar needs said typed exams under closer supervision work fine. The better lesson is not 'return to paper' but 'return to controlled environments.'

    Do not copy the most nostalgic solution from the story. Copy the security property you need, then implement it in a way your institution can defend on accessibility grounds.

      Attribution:
    • Yossarrian22 #1
    • Waterluvian #1
    • recursivedoubts #1
    • woodruffw #1

In plain english

AI
Artificial intelligence, a broad term for computer systems designed to perform tasks that seem to require human intelligence.
proctored
Supervised during an exam by a person or system that monitors for rule-breaking.
STEM
Science, technology, engineering, and mathematics.
viva
A spoken exam or oral defense where a student must explain and defend their work aloud.

Reference links

Teaching and assessment writeups

Testing infrastructure and tools

  • Illinois Computer-Based Testing Facility
    Example of institution-scale controlled computer-based assessment infrastructure.
  • Scantron
    Raised as a traditional exam technology reference point, mostly to question whether multiple choice can assess higher-level learning.

Articles and examples about anti-cheating formats

Research on handwriting claims

Analogies about cheating norms