College instructor turns to typewriters to curb AI-written work

Assessment in the age of AI

• Many instructors are shifting back to in‑person, paper-based quizzes and exams to make AI‑assisted cheating harder.
• Some already had “AI‑proof” structures: heavy weight on proctored written exams, projects defended line‑by‑line in person, or handwritten coding exams.
• Others argue too much exam weight is unfair (high‑stakes, time‑pressured, artificial compared to real work where references and tools are allowed).

Exams vs. “real life”

• One side: real work often allows Googling/LLMs; exams and whiteboard interviews are unlike anything in adult life, so designing around them is misguided.
• Other side: many roles require fast recall and reasoning under pressure (incidents, exec meetings, interviews); exams are a proxy for this and for verifying individual competence.
• Several note the real problem is poorly designed exams, not exams per se.

Oral and in‑person evaluation

• Some report systems where oral exams determine most of the grade; cheating is rare but bias risk is high, especially when a single professor controls a mandatory course.
• Defenders say commissions and written records mitigate abuse; critics say power dynamics still make contesting bias risky.

AI: ban, ignore, or integrate?

• “Ban AI” camp: AI lets students skip the learning process, devalues degrees, and harms honest students (especially under curves).
• “Integrate AI” camp: like calculators or compilers, AI should be taught as a core tool; design assignments where using AI still requires understanding, or where AI output is only a starting point.
• Some propose splitting: early years focus on fundamentals without AI; later years focus on doing harder work with AI.

Tool analogies and equity concerns

• Frequent comparisons to calculators, tractors, gyms, and running water; disagreement over whether LLMs are comparable, since they’re non‑deterministic and usually subscription‑based.
• Requiring paid AI tools is seen as widening inequality; others note local/cheaper models exist but may not match top-tier systems.

Cheating, credentials, and labor market

• Widespread AI‑assisted cheating plus weak enforcement may push employers to rely more on their own high‑stakes screening.
• Some argue many white‑collar jobs demand little true competence anyway; others expect AI will expose and eliminate low‑value “text-shuffling” roles.