{"schemaVersion":"1.0","generatedFrom":"https://brightaifuture.com/discoveries/guarded-learning","record":{"id":"guarded-learning","headline":"When the answer arrives too soon.","canonicalUrl":"https://brightaifuture.com/discoveries/guarded-learning","datePublished":"2026-09-07","dateModified":null,"sourcePublicationDate":"2025-06-25","author":null,"publisher":{"name":"Bright AI Future","url":"https://brightaifuture.com/"},"topics":[],"summary":"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.","evidenceState":"Demonstrated","keyFacts":[{"label":"AI’s role","value":"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."},{"label":"Documented result","value":"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."},{"label":"Important limitation","value":"The study examined short-term exam outcomes in one high-school mathematics setting."}],"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."],"evidenceLinks":[{"title":"Generative AI without guardrails can harm learning: Evidence from high school mathematics","url":"https://doi.org/10.1073/pnas.2422633122","type":"paper"}],"evidencePackUrl":"https://brightaifuture.com/evidence-pack/guarded-learning","embedUrl":"https://brightaifuture.com/embed/story/guarded-learning","attribution":{"credit":"Bright AI Future","requirements":["Link to the canonical Bright record.","Keep material limitations with the claim they qualify.","Link to the original evidence when repeating a substantive claim.","Do not describe a source check or organization-reported result as independent verification."],"sourceRights":"Linked source material, quotations, trademarks and media remain subject to their owners’ terms. No reuse right is granted for third-party media."}},"claim":{"humanProblem":"A tool that makes practice easier can also let a learner bypass the thinking needed to solve a problem alone.","priorConstraint":"Widely available chat interfaces are designed to be helpful in the moment, not necessarily to protect learning.","aiRole":"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.","documentedResult":"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.","whyItMayMatter":"It makes the design of help visible: a system can improve assisted performance yet leave less learning when the assistance disappears.","unresolvedQuestions":["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?"]},"evidenceAssessment":{"state":"Demonstrated","claimConfidence":"unassessed","reviewState":"approved","reviewMethod":"ai-assisted","reviewNote":"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.","lastSourceReview":"2026-09-07","independentVerification":"not-established-by-this-source-review"},"sources":[{"id":"source-guarded-learning","title":"Generative AI without guardrails can harm learning: Evidence from high school mathematics","url":"https://doi.org/10.1073/pnas.2422633122","type":"paper"}],"revisions":[{"id":"revision:a3f56235a1d3df554891","recordedAt":"2026-09-07","summary":"It makes the design of help visible: a system can improve assisted performance yet leave less learning when the assistance disappears.","sourceIds":["source-guarded-learning"]}],"corrections":[]}