Brave New Words

Salman Khan opens the book with an old and appealing idea: every student would learn better if they had a private tutor all to themselves. Someone who knew exactly where they were stuck, who could explain things a different way when the first try failed, and who was never in a hurry or embarrassed to repeat the lesson.

The problem is scale. No country can afford to pay a tutor for every child, and in underfunded public schools, with crowded classrooms and overworked teachers, one-on-one attention is a luxury. Families with money pay for private tutors or expensive test-prep courses, while those who cannot afford them fall behind.

The author's bet is that generative artificial intelligence, the kind that chats in natural language, can finally offer something close to that personal tutor to millions of people. The book champions this optimism, but spends much of its time discussing the risks and the conditions required for it to actually work.

To understand the argument, it helps to know the author's backstory. Khan was working as a financial analyst when he started tutoring his cousin in math remotely. He recorded short videos so more people could watch, posted them online, and watched the audience grow until it turned into a nonprofit organization, Khan Academy.

Over the years, the platform came to offer videos, practice exercises with instant feedback, and tracking dashboards for teachers, used in many countries. That experience showed two things: that instruction can be widely accessible, and that practice with rapid feedback makes a real difference.

What was missing, according to Khan, was conversation. A video explains and an exercise drills, but there is no one to ask a follow-up question and notice where the student gets stuck. The arrival of large language models looked like the missing piece to complete the puzzle.

The book mentions a classic 1980s study by educational psychologist Benjamin Bloom, which became known as the two-sigma problem. Bloom observed that students who received one-on-one tutoring performed dramatically better than students taught in a traditional classroom.

The gap was so wide that the average tutored student outperformed the vast majority of their peers. Ever since, the challenge has been to achieve that result without the cost of dedicating a human being to every single student. No one had found a practical solution at scale.

Khan sees AI as the first real chance to get close to that goal. Not because a computer is better than a good teacher, but because it can be available at any hour, with infinite patience, for students who would otherwise have no one else to turn to.