Bright
← Living questionsRECORD / Education · Access

When the answer arrives too soon.

A preregistered randomized trial in a Turkish high school compared a standard GPT-4 chat interface, a teacher-informed GPT tutor with guardrails, and no generative-AI resource during mathematics study sessions.

Original sources ↓ · Revision history ↓

Demonstrated · source published 2025-06-25

The human problem

A tool that makes practice easier can also let a learner bypass the thinking needed to solve a problem alone.

The prior constraint

Widely available chat interfaces are designed to be helpful in the moment, not necessarily to protect learning.

AI’s actual role

The study compared GPT-4 assistance that resembled ordinary chat with a version whose prompts and boundaries were designed with teacher input to support learning.

The documented result

The paper's preregistered primary analysis of unassisted exams found that generative AI without guardrails could harm learning. The teacher-informed guardrailed tutor largely mitigated that negative effect.

Why it may matter

It makes the design of help visible: a system can improve assisted performance yet leave less learning when the assistance disappears.

Limitations

The study examined short-term exam outcomes in one high-school mathematics setting.

The paper says the guardrailed tutor mitigated harm; it does not establish a complete or universally effective learning design.

Long-term retention and transfer were not measured.

Unresolved questions

Which guardrails work across subjects and ages?

Can a tutor proactively find misconceptions without taking over the work?

How should educators assess learning when students have broad access to general chat tools?

Source history & evidence assessment
Maturity
Demonstrated
Claim confidence
unassessed
Event date
Not recorded
Source published
2025-06-25
Captured
2026-09-07
Last source review
2026-09-07
Editorial method
AI-assisted source review
Place / relevance
Not recorded

AI-assisted editorial comparison with the cited primary source; result, setting, source date and limitations retained. Independently checked within the research team. Publication authorized by the site owner; no human source review is claimed.

Maturity describes the tested or operational setting. Confidence describes support for the particular claim; one does not determine the other.

Original sources

Generative AI without guardrails can harm learning: Evidence from high school mathematics · paper

Institutions: University of Pennsylvania · Budapest British International School

Explore the underlying question

Revision & correction history

2026-09-07 · It makes the design of help visible: a system can improve assisted performance yet leave less learning when the assistance disappears.

No corrections recorded.