The greatest school in history isn’t Oxford, Cambridge, Harvard or any other university you know. And no matter how hard you try, your kids won’t get in. Why? Partly because it was so selective it only admitted one student — but mainly because it closed in 336 BC. For me, Aristotle’s seven-year tutelage of Alexander is the education against which all others should be judged (after all, more than 2,300 years later we still refer to the lone pupil as “The Great”). It’s the ultimate testament to the power of tutoring — a power that artificial intelligence is poised to unlock.
The problem with tutoring is it can’t scale. Or it couldn’t. Because even as we’re besieged by concerns that AI-aided plagiarism is destroying education, we’re starting to see evidence that AI-enabled tutoring might supercharge it. Getting the technology right, though, will require lots of real-life experimentation. While there’s a limit to how much our traditional public school system allows for this kind of test-and-learn approach, this need creates an opportunity for the country’s growing crop of charter schools to make a unique contribution to the future of education.
The wealthy’s appreciation of tutoring did not die with Alexander. I paid rent my first year out of college as a private math tutor and today there are a host of companies offering tutoring services, with those at the high end often charging more than $1,000 per hour.
But for every student who can afford tutoring, there are hundreds more who could benefit from it. A meta-analysis of dozens of experiments with K-12 tutoring, conducted with students of all socioeconomic statuses, found that the additional academic attention significantly boosts student performance. And let’s say you could overcome the cost issue — with more than 50 million students in US primary and secondary schools, there will never be enough tutors to work with them all.
Early experiments with AI-based tutoring suggest it might help fill the gap. In a study of three middle schools in Pennsylvania and California, researchers found that a hybrid human-AI tutoring model — where the technology supported human tutors, allowing them to work with many more pupils — generated significant improvements in math performance, with the biggest increases going to the lowest-performing students. And in a study of four high schools in Italy, researchers replaced traditional homework in English classes with interactive sessions with OpenAI’s ChatGPT-4 and found that all the AI-aided groups did at least as well as those engaged in traditional homework — with some performing significantly better.
It could help at a college level, too. In a Harvard University physics course, for example, professors trained an AI tutor to work with some students (replacing their normal class time) while others had a traditional instructor-guided class. Students with AI tutors performed better — in fact they learned twice as much — and were more engaged with the lessons than those in the normal class, even though they had less interaction with a human instructor.
The most impressive findings may come from the developing world. Rising Academies, a network of private schools with more than 250,000 students across Africa, has implemented Rori, an AI-based math tutor for students, and Tari, a support system for teachers, both powered by Anthropic’s Claude and accessible via WhatsApp. Students who used Rori for two 30-minute sessions twice a week for eight months showed an improvement in their math performance “equal or greater than a year of schooling.”
None of this means AI-aided tutoring is a panacea. But it does suggest that such tutors are, if well-designed and implemented, very likely to be helpful even if they remain inferior to the best human options. Since many families can’t access or afford traditional tutoring, what matters is if they are better than no tutors at all.
But “well-designed and implemented” is a crucial part of that sentence. We don’t yet know what the best practices are for AI tutors. Learning this will require extensive experimentation. And, much as it pains me to say this as a proud product of public schools, that kind of free-form experimentation is likely to be a struggle for public school bureaucracy.
Research by the Department of Education and the Center on Reinventing Public Education at Arizona State University suggests that charter schools, which operate with more freedom about how they staff and teach, are often more innovative than traditional public schools. And because charters are not private schools, they cannot charge tuition or be selective about who they admit. This lets them generate useful data about what does and doesn’t work.
Of course, this doesn’t mean that charter schools are better than their public counterparts. Most innovations fail. But however painful failure is for an individual school, it can actually benefit the system because even bad outcomes produce useful information. Successful AI-based tutoring programs pioneered at charters can and will be adopted by public schools, and failed ones avoided. Given the potentially revolutionary change in education AI is driving, learning should be our primary goal — and charters are likely to be our best instrument toward it.
Gautam Mukunda writes about corporate management and innovation. He teaches leadership at the Yale School of Management and is the author of “Indispensable: When Leaders Really Matter.” He wrote this column for Bloomberg Opinion.
