Back to blog

Why does homework still fail to prove learning in the AI era?

Thắng Đoàn
Thắng Đoàn

Your homework grades no longer tell you who understands the material. When AI can produce every answer, the grade measures access to a chatbot, not learning.

The signal is gone

Take-home work used to be a cheap proxy for student effort and understanding. That proxy is broken. A UC Berkeley professor, Fields Medalist Richard Borcherds, stopped grading homework for correctness because AI use is now widespread among students. Per the New York Post, he said grading for correctness "just adds a huge incentive for them to use AI for homework instead of doing it themselves, and penalizes students for honesty." When grading punishes honest students, the instrument is the problem.

Institutions are improvising

Borcherds now awards credit for completing assignments rather than for correct answers, the New York Post reports. He also calls paying graders to evaluate AI-generated answers a waste of time and money. CNET's Rachel Kane, a former professor, describes grading AI-era homework as "an infuriating exercise in futility," with discussion posts citing materials far outside course reading. VnExpress reports a professor at a world top-10 university making the same call. Nobody has found a replacement yet; they are just triaging.

The tools are not the referee

Some hope AI itself will fix this, but the tools lean the other way. The New York Times tested ChatGPT for Teens, OpenAI's mode designed to address education concerns, and found its study tool still does a user's homework. A children's safety nonprofit told the Times the chatbot flunked its tests. Gizbot lists features parents should know before letting teens use AI for homework, which shows the same gap: features exist, but supervision still falls on adults. Building assessment on top of the same technology that broke it is a fragile loop.

Work with it, not against it

The most pragmatic response accepts AI use and redirects assessment. Khaleej Times reports UAE teachers would rather guide students on proper AI use than fight it, since students will use these tools anyway. That shifts the teacher's job from verifying answers to observing process. It also means Homework becomes practice, while proof of learning moves to supervised settings like class discussions and in-person work.

What this does not solve

None of these fixes resolve the core measurement problem. Completion-based credit tells you nothing about mastery. Accepting AI use assumes students will use it well, but the Times testing shows teen-focused tools still hand over answers. Moving assessment into classrooms costs teacher time that schools may not have.

Which of your current take-home outputs still prove genuine capability, and which only prove access to AI?
If you moved proof of skill into supervised settings tomorrow, what would that cost you?

List your three most-used assessments and mark each one: still valid with AI, or broken.

Sources

CNET

Khaleej Times

New York Post

New York Times

Gizbot

Share:

Recommended for you

Enjoyed this article?

Subscribe for new articles. No spam. Unsubscribe anytime.

By subscribing you agree to receive the newsletter. No spam, and you can unsubscribe anytime.