Insights
How AI Tutors Drive Massive Learning Gains
In rural Sierra Leone, fewer than one in four junior secondary students reach basic math proficiency. Most of their teachers never received formal training in mathematics.
Randomized controlled trial across twelve schools
Last month, Google DeepMind published results from a randomized controlled trial across twelve schools in Port Loko District. Over 1,700 students used an AI tutor for an average of fifteen hours over eight weeks.
The result: learning gains equivalent to 1.2 to 1.7 years of typical progress.
Students who completed twelve hours moved from the middle of their class into the top third.
These weren’t students with private tutors and fiber internet. They were teenagers in one of the world’s poorest countries, learning fractions and exponents on basic devices — with a tutor that met each of them exactly where they were.
The people with the fewest advantages consistently gain the
A separate NBER study found a similar pattern at the other end of the pipeline: generative AI closes three-quarters of the productivity gap between workers with and without formal education. The people with the fewest advantages consistently gain the most.
Whether AI will take jobs
We spend enormous energy debating whether AI will take jobs. Perhaps the more urgent question is whether we’ll let it give people the preparation those jobs require.
Fifteen hours. That’s all it took to show how thin the wall between potential and contribution really is.