In The Wall Street Journal, John Masko argues that AI “should be severely restricted for undergraduates” but is “vital for researchers.”
He sees the campus debate over AI as two different debates because a university does two different kinds of work. Research depends on the technology. Undergraduate learning may need protection from it.
I think he’s right about the split. I’d draw the line in a different place.
The real difference isn’t between researchers and students. It’s between work someone can already do independently and work they are still learning to do.
When you already know how to do the work, AI can save time.
When you’re still building the skill, the same help can skip the struggle that builds it.
A recent set of randomized trials shows why that distinction matters.
Participants worked on fraction problems and SAT-style reading questions with an AI assistant that could provide complete answers on request. After a brief period, the tool was removed.
While the AI was available, they did better. Once it was removed, they solved fewer problems than people who never had it. In the SAT-style reading study, the AI group also skipped 8 percent of the questions. The control group skipped 1 percent.
A field experiment with nearly 1,000 high school math students at one Turkish school found the same divide between assisted performance and independent learning.
Students who practiced with a custom GPT-4 assistant without learning-specific safeguards improved during practice, then scored 17 percent lower on the unassisted exam than students who never had it.
A second group used a guardrailed AI tutor built around teacher-designed hints rather than answers. On the unassisted exam, they matched the control group rather than outperforming it.
The line runs through every learner, and it depends on how the tool is used.
Expertise is task-specific. The same person may be ready to use AI for one kind of work and still need the struggle in another.
For parents, my practical rule is simple.
Have the student attempt the problem first.
Ask the tool for a hint, not the answer.
Then close it and ask the student to finish or explain the work unaided.
AI can amplify a skill someone has already built. When the skill is still being built, the tool has to support the thinking — not replace it.
Great learners are built, not born.