In 1984, Benjamin Bloom demonstrated empirically that one-on-one tutoring produces achievement two standard deviations higher than group instruction. Forty years later, is AI tutoring solving this once-"impossible" problem?
A 2025 Harvard physics study found that AI tutors produce twice the learning outcomes in less time compared to traditional classroom instruction. Yet the OECD 2026 Digital Education Outlook warns that generative AI can both help and harm learning. What makes the difference?