When You Wonder Vol. 16: Sal Khan
On AI, dinner parties, and the power of knowing the basics
Long before Khan Academy became a global learning phenomenon – one that now reaches millions of people in nearly 200 countries – it was a conversation between a kid and a caring adult.
Sal Kahn was fresh out of business school when he volunteered to tutor his 12-year-old cousin. She’d been struggling with math, and though she lived hundreds of miles away, Sal helped her out by phone. “Slowly but surely,” he remembers, “she got caught up with her class, even [pulled] ahead of her class.” She moved from the remedial track to the advanced track. Her younger brothers came looking for tutoring, too – and then so did a group of their cousins.
Each of their needs was different, but Sal spotted a pattern. “My cousins weren’t struggling because they weren’t bright or hard working, or [because they] didn’t have access to good schools,” he says. Rather, the problem was that they – like every learner – had specific gaps in their knowledge.
If you’re an algebra student, for example, “but you forgot how to divide decimals or you’re kind of shaky on it, and all of a sudden you have to divide decimals in that algebra equation, then you’re going to have trouble,” says Sal. “And it’s hard for the algebra teacher to address that need, because if some of your class of 31 students has a gap in dividing decimals – which isn’t even in the curriculum – and other students are having trouble with negative exponents, and other students are having trouble with logarithms, then how do you handle that?”
The answer, for Sal, came in the form of a dinner party.
He’d written some software to keep track of his cousins’ progress. He was talking about it with friends over dinner – “I’m a super-fun dinner party guest,” he jokes – when the host stepped in with an idea: What if he turned his tutoring lessons into videos? He could even put them on a new platform called YouTube.
“I initially thought it was a horrible idea,” Sal admits. “YouTube did not seem like a serious platform for learning at the time, but I gave it a shot anyway. And, you know . . . after I made about 50 videos, [my cousins] said they liked me better on YouTube than in person.”
Sal laughs at the memory. But even 20 years later, the lesson stays with him. “What [my cousins] were saying is they still valued me in their lives. In fact, because of the videos, we were able to go deeper into their motivations and what was going on in their lives when we got on the phone. We were able to connect more as human beings.”
Sal and his cousins had rediscovered what Fred Rogers knew so well: that technology, when deployed with children’s best interests in mind, can be more than a powerful tool for learning.
In the right hands, it can strengthen the bonds that make us human.
Today, our conversation with Sal Khan kicks off Season 10 of our podcast, Remaking Tomorrow. We hope you enjoy Sal’s company as much as we did. (We think he’d be an excellent dinner party guest.)
Highlights from our interview have been edited for brevity.
Can you tell us a bit about the evolution of Khan Academy?
I was going down a very iterative and intuitive path, but I wasn’t the first to think of these things. Alexander the Great was tutored by Aristotle. That was personalized learning. If young Alexander needed to speed up, I’m sure Aristotle would have sped up. If he had gaps, he would have slowed down. But very few people could get that type of education.
Then, 300 years ago, we introduced mass free public education, which has done huge, positive things for society. But there were compromises: We couldn’t give everyone their own Aristotle. We began to batch students together.
When you do that lock-step, students don’t struggle because they’re not destined to master algebra. It’s because they have gaps. And it’s been a source of inequity, because wealthier families could tutor their children or hire someone to do it. Benjamin Bloom famously wrote about [one-on-one tutoring] back in 1984. He called it a “2 sigma problem,” because even though he made a solid argument that you could get a two-standard-deviation improvement [in student learning], it’s a problem because we can’t give everyone their own personal tutor.
Even back then, he said maybe software could approximate some aspects of [one-on-one tutoring]. That’s essentially the path that Khan Academy has gone down. And multiple efficacy studies have shown that if students put in even 30 to 60 minutes a week, they were getting accelerated, depending on the study, someplace between 20 and 40 percent over the course of a year. And some recent studies have been even more profound than that.
Three years ago, you launched Khanmigo, an AI tool developed alongside OpenAI’s ChatGPT. Tell us about that.
In 2022, six months before ChatGPT comes out, I get an email from OpenAI. They have a new model coming out, which would eventually become GPT-4. They wanted to show it to us and see if we had any use for it, because they wanted to launch it with socially positive use cases. I saw that with a little bit of prompting, it could get even closer to emulating what Aristotle might have done for young Alexander.
But there were still a bunch of issues, especially at those early stages. AI was making math errors. It was making up facts. And what about safety and privacy? What about cheating? We had all these debates. You know, Khan Academy is not-for-profit. We attract a very passionate, mission-driven team who really cares about doing things right. I was like, “Look, these are legitimate fears. I have those same fears. But we can’t use those as a reason to run away from this technology. We, if anything, have a role. When people engage with our AI, let’s make it transparent to the teachers and administrators. Our AI should be Socratic – not one that gives solutions. It shouldn’t share data, etcetera.”
Khanmigo isn’t a solution by itself. The main learning still happens from deliberative practice using our traditional exercises. But if AI can be one extra layer of support – and something that gives insights to the teachers and helps them save time with things like lesson planning – then that could be great.
How do you see AI playing a role in student learning?
There has to be a real – and probably overdue – conversation in education about things like home assignments. If our goal is to make sure a student can construct their own thoughts from scratch and write an essay, then I think that’s going to have to happen in class now – which is healthy, because the student is supported [by a caring adult in a classroom].
If you want a student to be able to do independent work outside of class, AI can help make the process transparent to the teacher, which can have benefits even beyond policing cheating. It can also tell the teacher: “A lot of your students are still struggling with formulating a thesis statement. Why don’t we modify tomorrow’s lesson plan a little bit to really focus on that?”
And at the high school and college level, I think every class should have at least one assignment per term where the teacher says, “Go use AI. Go use whatever you can. But if you use AI, I want you to build something that blows my mind. Instead of writing the business plan, go start the business. Make a theme song and advertisements. Vibe-code your initial prototype and get some real users.”
When you think about how to prepare young people for this AI driven world, what does it mean to do so ethically and responsibly?
I’m not going to claim that I have the perfect crystal ball – things are changing so fast. But if we’re talking about preparing students for this future world, then I think there should still be a very strong emphasis on the basics.
I think some people have made the argument that in an AI world, AI can write for you. AI can do your research for you. AI can do your math for you, etcetera. Do we really have to learn those things anymore? And I’m a big believer that we do. If you actually have those skills yourself – if you have the content knowledge – you’re going to be in a better position to use the technology that’s available to you.
This isn’t the first time that we’ve seen this type of phenomena. Back when the calculator came out in the late ‘60s or early ‘70s, people said, “No one has to learn arithmetic anymore.” Well, it turns out that the people who used the calculator were actually the people who did know their arithmetic. When the internet came out, people said, “Oh, no one has to know their content anymore. They can just search for it.” But to use the internet well, you actually do have to have good content knowledge.
AI will raise the floor. It can come up with a first draft, it can do all of these things, but you’re not going to get the most out of it unless you can keep up with it – or ideally, be even better. No one wants an editor who can’t write as well as the staff writer, which is now going to be the AI. No one wants the coding architect who can’t code as well as the junior coder, which is now going to be the AI. So the basic skills, I think, are as important as ever. And that’s where Khan Academy is focused: on math and English and language arts and science and beyond.
On all the things that make us human.
For more on AI in education, check out Sal’s latest books, Teaching Students to Use AI Ethically & Responsibly: Exploring AI With Intentionality, Curiosity, and Care and Brave New Words: How AI Will Revolutionize Education (and Why That’s a Good Thing).