Hybrid human-AI tutoring boosts learning by 61% with differentiated roles
Proactive tutors for low-performers and reactive for high-performers drove huge gains.
A new study from Carnegie Mellon University and Stanford, accepted at EDM'26, shows that differentiating human tutor roles in hybrid AI tutoring dramatically improves student outcomes. Researchers assigned 635 students in grades 5-8 based on their median state test scores: lower-performing students got proactive human tutor support (initiated by the tutor), while higher-performing students received reactive on-demand help. Using a regression discontinuity design across fall (AI-only) and spring (human-AI) periods, the team isolated the effect of the differentiated approach.
The results were striking: compared to AI-only tutoring, the hybrid model increased time-on-task by 25%, skill proficiency by 36%, and overall academic growth (on the MAP test) by 61%. Proactive tutoring specifically showed marginally higher growth (75%, p=0.065) for students farthest below the cutoff, helping narrow achievement gaps. Both groups benefited comparably in time-on-task and proficiency, suggesting a cost-effective, scalable strategy for hybrid instruction.
- 635 students (grades 5-8) assigned to proactive (< median) or reactive (≥ median) tutoring based on state test scores
- 61% increase in MAP academic growth and 36% boost in skill proficiency compared to AI-only tutoring
- Proactive tutoring yielded 75% growth for low-performers, narrowing achievement gaps (p=0.065)
Why It Matters
Personalized human-AI tutoring can scale equitably, giving low performers proactive help without slowing high performers.