Learn: The AI education emergency

For teachers

The short answer

The strongest results so far come from AI built for teaching and used with a teacher: an AI that coached human tutors in real time raised student mastery, most for students of less experienced tutors, and a guardrailed AI tutor avoided the learning loss an unrestricted chatbot caused. Teachers decide how it is used.

AI that makes a tutor better

In a randomized trial with 900 tutors and 1,800 K-12 students in Title I schools, students whose tutors had an AI tool suggesting next moves in real time were 4 percentage points more likely to master topics, and students of lower-rated tutors gained 9 points. The tool cost about $20 per tutor per year arxiv.org. The AI supported a human tutor; it was measured on lesson exit tickets, not state tests.

Guardrails decide whether students learn

In a field experiment with nearly a thousand high school math students, an unrestricted ChatGPT-style tool raised practice grades but lowered unassisted exam grades by 17% compared with students who never had it. A version with prompts built to protect learning avoided most of that drop econpapers.repec.org.

In a Harvard physics course, a tutor built on the same teaching practices as the in-class lessons, with pre-written answers, produced learning gains over double those of in-class active learning; the authors state it should not replace in-person teaching pmc.ncbi.nlm.nih.gov.

In Nigeria, a six-week, teacher-supported after-school program using Microsoft Copilot raised English scores by 0.23 standard deviations in a World Bank randomized trial econpapers.repec.org.

The OECD's 2026 review: completing tasks with AI does not necessarily translate into learning, and advantages can disappear when AI is removed, as in exams; AI guided by clear teaching principles can support learning digital-skills-jobs.europa.eu.

The classroom you already have

About 1 in 8 U.S. teaching positions is either unfilled or held by a teacher not fully certified for the assignment, according to 2025 state data compiled by the Learning Policy Institute learningpolicyinstitute.org.

Sources

  1. Wang, Ribeiro, Robinson, Loeb and Demszky, "Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise," arXiv:2410.03017, 2025 (preprint). arxiv.org
  2. Bastani, Bastani, Sungu, Ge, Kabakci and Mariman, "Generative AI without guardrails can harm learning: Evidence from high school mathematics," PNAS 122(26), 2025 (abstract). econpapers.repec.org
  3. Kestin, Miller, Klales, Milbourne and Ponti, "AI tutoring outperforms in-class active learning," Scientific Reports 15:17458, June 3, 2025. pmc.ncbi.nlm.nih.gov
  4. De Simone, Tiberti, Barron Rodriguez, Manolio, Mosuro and Dikoru, "From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria," World Bank Policy Research Working Paper 11125, May 2025 (abstract). econpapers.repec.org
  5. OECD Digital Education Outlook 2026, as summarized by the European Commission Digital Skills and Jobs Platform, 2026. digital-skills-jobs.europa.eu
  6. Learning Policy Institute, "An Overview of Teacher Shortages: 2025," July 16, 2025. learningpolicyinstitute.org

See an error? Report a correction.