A Tutor for Every Child: Sal Khan on How AI Could Transform Education

Sal Khan says his background was in tech and business school when he took a job as an analyst at a small hedge fund in Boston. A cousin struggling with math had been placed in a lower-track class, and when Khan offered to tutor her remotely, she caught up and was later moved into the advanced track. Word spread through the family, and soon he was tutoring about fifteen cousins.

Because many of the cousins struggled for the same reason, gaps in earlier knowledge such as decimals and negative numbers, he began writing software that generated problems, gave instant feedback and hints, and tracked what each student got right and wrong. He called it Khan Academy. A friend in Silicon Valley asked whether he planned to scale it, and suggested YouTube.

Khan first thought the idea was terrible, since YouTube then seemed to be skateboarding and cats playing piano. He learned to screen-capture, and his cousins joked that they preferred him on YouTube because they could pause, rewatch and avoid any embarrassment. When strangers began finding the videos and the web app, he made the videos public, opened the app to everyone and eventually left finance to run Khan Academy as a nonprofit offering free world-class education.

Khan says his early approach was intuitive, and he later learned it matched what educators call mastery learning, advocated by Benjamin Bloom. As a strong student himself, he believed academic success was less about being special than about truly understanding why methods work. He also saw peers who were good at chess or puzzles but struggled in class, which convinced him the issue was not raw cognitive ability.

The core problem is that students move on with fluency gaps. Khan recounts visiting a Bronx middle school where a seventh grader reached for a calculator to compute 3 times 7, and says he gently slapped away the calculator and watched the student count on fingers. In his view the child understands algebra concepts but is overloaded by weak basics, which mastery learning addresses by letting everyone work at their own pace until they have mastered a topic.

Khan states the mission as world-class education for anyone, anywhere, for free, covering the foundational academic material through the start of college. The platform spans early childhood through math, science, humanities, and areas such as finance and computer science, with practice and instant feedback and tools for teachers to assign and track work. He also describes the Khan Lab School he began so that his own children would attend a school that practiced mastery learning, and partnerships with Arizona State University and the University of Hawaii.

Khan describes the AI vision as a tutor for every student and a teaching assistant for every teacher. A tutor can answer questions and motivate learners, guide them, let them debate or talk with historical figures, and help teachers write lesson plans and rubrics. He calls it the biggest positive transformation education has seen, and calls it a game changer over the long term.

He explains why through Bloom's 1984 "2 sigma problem." Personal tutoring, the gold standard since the days of Aristotle, lifted the average student to the top of the distribution by two standard deviations, but classrooms could not afford it. Mass schooling forced fixed pacing and fixed schedules, so some students moved on with partial understanding and fell behind.

Earlier technology could approximate parts of tutoring, including on-demand video, practice with instant feedback and data for teachers to form small groups. Khan did not expect generative AI to arrive in his lifetime, but when he played with GPT-4 he was struck that it acted like a good, Socratic tutor. It had real problems with math and hallucinations, which his team worked hard to reduce, and early school pilots showed more engagement and less getting stuck.

Teachers in the pilots, Khan says, liked the assistant features that cut the time spent on lesson plans and rubrics and made teaching feel less lonely. He also describes a partnership with Instructure, whose learning management system many institutions use, so AI could help create assignments and report on the process behind a student's work, not just the final result. Looking ahead he expects tutors with memory that connect lessons to interests such as women's soccer, inspired by the AI primer in Neal Stephenson's novel *The Diamond Age*.

Asked how school structures might change, Khan says that in environments where students are well supported with Khan Academy, he sees progress faster than the traditional pace. Where fluency problems are fewer, teachers get time back, since most spend about half their time on grading and lesson planning. That frees energy for the students themselves.

He pictures students putting on virtual reality gear to explore the circulatory system or travel to ancient Greece and debate Socrates. More time and space would also let middle schoolers pursue passions, such as producing studio-quality music or making fantasy films. For children whose families cannot send them to enrichment camps, he argues AI can bring the world to them without large resources.

On adoption, Khan expects nearly all teachers to be using generative AI for lesson plans, grading, progress reports and rubrics within about five years, perhaps beginning next year. Teachers gain a lot of time, which should speed acceptance, and cheating concerns create urgency for solutions that must themselves use AI. Visiting a classroom in five years, he says, you might not see big differences at first glance, but you would notice teachers with more time for struggling students.

Khan explains that during the pandemic Khan Academy usage surged, but he noticed students were isolated and needed human support. Tutoring is usually expensive, so he asked whether volunteers could provide it. That led to Schoolhouse.world, a sister organization whose mission is to connect the world through learning with free tutoring.

The volunteers include high school and college students, teachers from top private schools and retired professors. The program is smaller than Khan Academy, but Khan says it is powerful and that the social connection across borders is as valuable as the academic help. He also describes Socratic seminars at the Khan Lab School, run for an hour or two rather than for hours, as a way to blend asynchronous learning with real discussion.

Khan says he is increasingly convinced that collaboration, social connection and emotional growth may matter as much as test scores. New teachers at the Lab School often remark on how collaborative the students are compared with more competitive schools. He ties this to mastery learning, where the message is that everyone should learn the material rather than beat each other.

Khan recalls that in late summer 2022, Greg Brockman and Sam Altman reached out to talk about their newest model, which he did not yet know was GPT-4. They said they wanted to apply it in areas such as education and healthcare with organizations that would use it responsibly, and he believes Khan Academy was among the first they contacted. Part of the reason for the call, he says, was that Bill Gates had said he would be impressed if the model could pass an AP Biology exam, and Khan Academy had many such questions.

In a demo with his chief learning officer, Khan asked it an AP-style question, then asked it to explain why the answer was correct and why the others were wrong. The explanations were strong, and the model could play a tutor role and hold a Socratic dialogue. They also noticed math errors and gave OpenAI feedback, and Khan says that helping reduce those mistakes serves humanity beyond Khan Academy.

Safety work began immediately because Khan saw the cheating risk. The first guardrail is in the system prompt, which tells the AI not to give answers and to ask guiding questions so the student does the work. The second is transparency: for students under 18, conversations are visible to parents and teachers, and unhealthy conversations are flagged to them.

Khan says Khan Academy also offers digital literacy guidance on how students should use generative AI, and that conversations are not used to train OpenAI's models. He calls it a very sensitive new domain, and says the team deliberately took conservative positions.

Khan does not expect generative AI to change universities as dramatically as it might schools, because the return on investment of college varies widely. Elite institutions will probably be fine, while community colleges, which are flexible and can respond quickly to jobs and credentials, will also do well. He is most worried about institutions charging very high prices that leave many students with debt and weak job prospects.

His proposal is that colleges should bear some responsibility for a portion of student debt. That would push them to give students better information about careers and to check that graduates can repay. He also wants a world where credentials reflect what you know and can do, not how long you sat in a seat, and says Khan Academy is working on credentials that employers can trust regardless of age.

Asked what skills to encourage in children, Khan offers two metaphors. The first is the camera: when it arrived, portrait painters feared for their jobs, but art moved toward impressions and emotion, and creativity exploded. He sees AI the same way, freeing creators who know how to direct and integrate these tools.

The second metaphor is that everyone will soon have assistants, so a programmer becomes an architect and a writer becomes an editor. Managing assistants well requires deeper skills than before. That is why Khan urges that children master reading, writing and math at a high level rather than just adequately.

He rejects the idea that you need not learn anything because Google exists. The more concepts and information you can use quickly in your own head, the faster tools will accelerate you, and a child who needs a calculator for 3 times 7 or a search for basic history will be left behind.