A new study from the UK finds that AI-powered math tutors — when used in tandem with some human oversight — can deliver “extremely personalized” instruction without hallucinating or producing unsafe or misleading content.
The hybrid model works like this: the AI handles adaptive feedback, hints, and personalized problem sets, while human instructors intervene when needed — offering guidance, corrective support, or emotional encouragement. This combination helps preserve the strengths of human tutoring (judgment, empathy, context-awareness) while leveraging the scalability and consistency of AI.
In trials, students using these AI-plus-human tutoring setups showed improved learning outcomes compared to AI-only tutoring. The human-AI groups tended to stay more engaged, and gains grew over time as students spent more time interacting with the AI tools under human supervision.
The findings suggest that rather than replacing teachers, AI tutors are likely most helpful when treated as assistants — enhancing and scaling tutor effectiveness, especially where resources are limited. This hybrid approach could help democratize access to skilled instruction without sacrificing quality.
