Early research suggests AI tutors can improve learning through personalized instruction, but questions about accuracy, privacy, equity and the role of teachers remain.
THE UNIVERSAL RECORD
Sourced reporting. No opinions.
Brad Socha | August 25, 2026 | 5:09 AM EST
A student struggling with a physics problem could ask for another explanation, request an easier example and work through the problem one step at a time, without waiting for a teacher to become available. That possibility is moving closer to reality as AI tutors become increasingly capable of adapting conversations to individual learners.
The potential is significant. Human tutoring can provide personalized attention that is difficult to reproduce in a classroom containing dozens of students. Generative AI could make some elements of that experience available on demand and at much greater scale.
Evidence is beginning to show what that might mean. A randomized controlled trial involving Harvard University students, published in Scientific Reports in 2025, found that students using a specially designed AI tutor achieved substantially greater learning gains in less time than students receiving an in-person active-learning lesson covering the same material.
But the findings do not establish that a general-purpose chatbot can replace a teacher, or that AI tutoring will work equally well across subjects, age groups and educational systems. The Harvard system was deliberately engineered around established teaching practices, and the experiment involved specific undergraduate physics lessons.
The distinction could determine whether AI becomes an effective educational tool or simply another source of answers.
How AI Tutors Could Personalize Learning
Traditional classroom instruction has an unavoidable constraint: one teacher must divide attention among many students.
An AI tutor can potentially respond differently to each learner. A student who understands a concept quickly could advance. Someone struggling could receive another explanation, additional examples or questions designed to identify where their understanding broke down.
Unlike simply generating an answer, a well-designed tutoring system can be instructed to guide students toward solving a problem themselves.
The Harvard experiment offers an early indication of the potential. Researchers studied 194 eligible students in an introductory physics course using a randomized crossover design. Students learned material either through an AI tutor or an active-learning classroom lesson.
Students using the AI tutor recorded more than twice the median learning gains relative to the pre-test baseline compared with the classroom group. Their median time using the tutor was 49 minutes, while researchers estimated 60 minutes of learning time for the classroom lesson. Students using AI also reported significantly higher engagement and motivation.
Evidence is emerging outside wealthy universities as well.
A World Bank randomized controlled trial examined an after-school program for secondary students in Edo State, Nigeria. Students used Microsoft Copilot, then powered by GPT-4, alongside teacher guidance to study English and develop AI and digital skills during a six-week program in 2024.
The World Bank researchers reported an overall improvement of about 0.31 standard deviations on the assessment, including an effect of approximately 0.23 standard deviations for English. The researchers estimated that the overall gains were comparable to roughly 1.5 to two years of typical learning progress in similar settings.
Importantly, teachers remained involved. They introduced topics, helped students interact productively with the AI and encouraged reasoning rather than simply obtaining answers.
That model, AI augmenting teachers rather than replacing them, may prove particularly important.
AI Education Still Has Major Limits
The same technology that allows an AI tutor to produce individualized explanations creates risks that conventional educational materials do not pose in quite the same way.
Generative AI systems can produce incorrect information that sounds convincing. In education, an authoritative-sounding mistake could teach a student the wrong concept, particularly if the learner does not yet have enough knowledge to recognize the error.
There is also a deeper educational question. Learning is not simply acquiring correct answers. Students need to develop reasoning, persistence, critical thinking and the ability to solve problems independently. An AI system that does too much of the intellectual work could undermine some of the abilities it is intended to develop.
UNESCO has called for a human-centred approach to generative AI in education. Its guidance emphasizes data privacy, age-appropriate use, validation for bias and pedagogical effectiveness, and continued human accountability. UNESCO also warns that unequal access to devices, connectivity and capable AI systems could deepen existing educational inequalities.
Children’s data require particular attention. Conversations with an AI tutor can potentially reveal detailed information about a student’s abilities, weaknesses, interests and behaviour. Decisions about what information is collected, where it is stored and how it can be used therefore become educational as well as privacy issues.
Teachers remain central for another reason: education is social.
A teacher can recognize frustration, encourage collaboration, understand circumstances outside a lesson and make professional judgments that extend beyond generating an appropriate response to a question. UNESCO has argued that effective education still depends on well-trained and supported teachers even as AI capabilities expand.
The strongest evidence so far therefore points toward a more nuanced future than classrooms run by machines. AI could give students something education systems have historically struggled to provide at scale: individualized assistance available when a learner needs it.
Whether that becomes transformative will depend less on whether AI can answer questions and more on whether educational systems can design it to help students think.
The early results are promising, but the evidence remains limited compared with the enormous diversity of classrooms worldwide. Longer studies will be needed to determine whether learning gains persist, how AI tutoring affects independent reasoning and which students benefit most.
AI tutors may ultimately change education substantially. The more immediate possibility, however, is not replacing teachers, it is giving teachers and students a new form of individualized support that previously could not be provided to everyone.
Sources:
Scientific Reports — AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting —
https://www.nature.com/articles/s41598-025-97652-6
World Bank — From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria —
https://documents.worldbank.org/en/publication/documents-reports/documentdetail/099548105192529324
World Bank — Addressing the Learning Crisis with Generative AI: Lessons from Edo State in Nigeria —
https://blogs.worldbank.org/en/developmenttalk/addressing-the-learning-crisis-with-generative-ai–lessons-from-
UNESCO — Guidance for Generative AI in Education and Research —
https://www.unesco.org/en/articles/guidance-generative-ai-education-and-research
UNESCO — AI and Education: Protecting the Rights of Learners —
https://www.unesco.org/en/articles/ai-and-education-protecting-rights-learners
U.S. Department of Education — Artificial Intelligence and the Future of Teaching and Learning: Insights and Recommendations —
https://www.ed.gov/sites/ed/files/documents/ai-report/ai-report.pdf
About the Author
Brad Socha is the founder of The Universal Record, focused on sourced, factual global reporting. Coverage includes international news, geopolitics, technology, and major developments.
