Sal Khan CEO & Founder of Khan Academy: Personalized education, AI and the drive to learn

Now, I just think it's such a lovely story the way you went from a hedge fund to starting Khan Academy. Would you mind just recapping in a few words how you got started on this?

. I was a year out of business school. My original background was in tech, but I go to business school. I end up at a very small hedge fund in Boston. A year out of business school, I get married in New Jersey.

I was born and raised in New Orleans. My family comes for my wedding in New Jersey and then comes and spends time with me in Boston and just comes out of conversation that my 12-year-old cousin, Nadia, who was staying with us, was having trouble in math. So I offered to tutor her. when she went back to New Orleans. She agreed.

And the tutoring worked, long story short. And that same Nadia, who was struggling with unit conversion in seventh grade, by that summer, was taking calculus classes at the University of New Orleans. So I started tutoring her younger brothers. Word spreads in my family, free tutoring is going on. Before I know it, I'm tutoring 10, 15 cousins, family, friends, after work every day.

Obviously, this was just a hobby.

And I saw a common pattern that they were struggling because they had gaps in their knowledge. They needed to review things. They needed to refresh things. So in 2005, I started writing software for them. Once again, it's a family hobby, but I called it Khan Academy.

And that software would give them practice problems and for me, as their tutor, to keep track of it. I didn't have videos at the time. Then, in 2006,, a friend suggested that I make YouTube videos for my cousins to help scale my lessons, because I actually was having trouble scheduling with 10, 15 cousins and I felt like I was repeating the same thing over and over again. And so I started making those videos, and my cousins famously told me that they like me better on YouTube than in person. And what they were saying, I think, is they liked having an on-demand version.

There was no shame that they had to review something from a few years earlier. They could watch it in the middle of the night, if that's when they were working on the problems. They still appreciated having me check in on them, et cetera. And by 2008, 2009,, there were about 50,000 to 100,000 folks using the videos and the software. I actually had to turn off the access to the software because it was overwhelming my web hosting.

And that's when I, in 2008,, I set it up as a nonprofit. At the time, I didn't think I was going to quit my hedge fund job. I actually liked my hedge fund job, but I set it up as a nonprofit with a mission, free world-class education for anyone, anywhere. And then, by 2009,, I had trouble focusing on my hedge fund job because I was getting letters from all over the world. I was just waiting to get home to work on the next video or to write the next software module.

And so that's when I quit my day job. My wife and I looked at our finances and we said, okay, maybe I could do this for a year, live off of savings a bit. And that first year was tough. Whenever you try to start anything, you have maybe a delusionally optimistic view of maybe people will come and support it. It took a little while, but by fall of 2010, we got our first significant philanthropic support.

But whenever I read those letters from people all over the world, saying how, you know, some of them are just simple thank yous, like, hey, you helped me pass this exam or that exam. But some of them are pretty emotional, where they're talking about how they had no self-esteem in some topic, they, you know, people thought that they were dumb their whole life. Then they went some other career track, or they joined this or that, and then they found Khan Academy. And then they started working at their own time and pace, they discovered a love for this subject, and then they go back to college, or now they're getting a PhD in something. When you get letters like that from people you don't know all over the world, it's incredibly motivating.

And so that, you know, the hardest point was probably that year, from fall of 2009 to fall of 2010,, where I had quit my day job, we're digging into our savings. My wife was still in training to be a doctor. Our first child had been born. You could imagine, I had given up a very good job that was paying very well.

That was, you know, three or four months into that, it was very stressful. And whenever I was tempted to refresh my resume and apply for another finance job, I would look at these letters. I was like, no, this is something here. And I did tell myself, if I didn't get any real philanthropic funding in the next year, I wouldn't give up the project. I would go back to work so I could support myself and my family, but I would keep working on it as a hobby.

And maybe the timing just wasn't right. So I did want to make it unkillable, because I felt in my heart of hearts that there was. there's something there.

What should I do? And he's like, oh, you should work at a hedge fund. And I said, well, that sounds great. What's a hedge fund? And you know, we explained, it's an investment firm, but they have more flexibility and they can do more creative things than, say, a mutual fund.

And so I was drawn to that career because of the intellectual journey that it seemed to have. And I was so naive. I started asking some friends in finance. I said, hey, do hedge funds pay? well?

And they said, yeah. And I said, about how much? And once I learned that, and I had, I still had some debt from undergrad, and then I had about $120,000 of debt from business school. I said, you know, I did not grow up, I think, you know, my family, I grew up in a single mother household. at best.

I would call us a lower middle class. at best. I was like, okay, I need to pay off this debt, save some money, and, you know, get a down payment on a house eventually, maybe help support family, et cetera. And it was intellectually stimulating. So that's what drove me down the hedge fund path.

But my wife, at that time my fiancee, always used to give me a little bit of grief about it. She's like, you know, people like you, you have this education, you have all of these skills, and you're just using it to make investments. I was like, well, you know, the world needs hedge funds. It helps price discovery and liquidity and all of this kind of stuff.

. You have the financial crisis. And I was already making videos for my cousins on algebra and physics and SAT prep. And then I saw that people were not, did not understand the difference between insolvency and illiquidity. They did not understand what a credit default swap is or what a mortgage-backed security.

And these things that historically were kind of exotic things that people in high finance would know were now on the headlines in the newspaper or on the news. And so I started making videos on how does the Federal Reserve work? What does a financial contagion look like? What's a mortgage-backed security, credit default swap, et cetera, et cetera. And it turns out that the mainstream press, I started getting letters from the mainstream outlets saying, hey, we're watching your videos before we're reporting on the financial crisis.

And then they, the first national publicity or international publicity that Khan Academy got had nothing to do with academic subjects. It was. CNN brought me on to talk about the financial crisis for 15 minutes straight on the day that I think, you know, the market tanked by like 4% that day or 5%. But that was a signal that to me that, look, even though what I was, you know, now it's become mainstream for people to do something on YouTube and to get a lot of attention and even make a living. In 2008, 2009,.

that was not a mainstream thing, especially because I wasn't doing dance videos. I was doing fairly, you know, I was explaining credit default swaps. But that, you know, when you get that type of attention for it, I was like, okay, there's something here. And then that helped bring even more attention and word of mouth to what was happening on the math and the science and all of the other subjects we were doing.

And I am guessing that, if, you know, Aristotle said, okay, this kid's going to be emperor one day, if he's struggling with military strategy, I'm not just going to give him a C and move on, I'm going to work with him and make sure he understands military strategy. If he's struggling with the accounting for the nation or for the empire, he needs to understand that in order for him to govern for his future. So young Alexander was getting personalized tutoring, personalized teaching from Aristotle. You fast forward to about, let's call it 300 years ago, a very utopian idea of mass public education. It started in some of these first countries to industrialize, you know, what is now.

Germany, the UK, the United States, Japan were the first countries to say, hey, we should have mass public education, but they had to make a compromise. They said, well, we can't afford to give everyone an Aristotle. We're going to use the tools of the industrial revolution. We're going to batch students together, move them at a set pace. We're going to create some standards.

We're going to create assessments that measure. And, you know, we don't need everyone to be, to master quantum physics or calculus. They didn't know about quantum physics back then. We don't need everyone to be able to become a professor. So we'll just sort people, you know, we'll just measure the product.

And at certain decision points, certain product is really retaining the knowledge. And those are the people who are going to go into the knowledge economy. And then certain people are in the middle, although, you know, those could, people can kind of be the managers of the factories. And then there are certain people who are struggling, and those could be the less skilled labor that we need for the industrial revolution.

He wrote this paper. It's, it's probably the most cited paper on tutoring and mastery learning. And mastery learning is just this idea. If you haven't learned it yet, keep working on it. That's all.

mastery learning is. Don't, don't move on, or you can move on, but come back and fill in that gap. And, you know, he argues in that paper and he has some data to back it up that you could get a two standard deviation improvement by having one on one tutoring in a mastery framework. Now, since then, people have debated, is it two standard deviations? Is it less?

But generally speaking, there is. there's a large evidence pool that more personalization one on one is good for, is good for students. It has, it has an effect. But he framed that paper as a problem because, like, well, this is nice. But how are you going to scale that?

Well, if you think about what Khan Academy has been doing, even before we talk about things like generative AI, I was trying to scale myself as a one on one tutor. And the first thing I did is I created some of those personalized exercises that I created, dashboards for myself to monitor my, my students, my cousins, so that I could be more personalized with them. when I got on the phone. I started making on demand video. It's not the same as a one on one tutor, but it has aspects of it.

If my cousins were stuck in the middle of the night, they could get that on demand help. And I was making these short five, 10 minute videos. So it's almost like and there's a library of hundreds and now thousands of them that whatever your question is, you can click on that video. And now, with AI, we can potentially go, go, go even further. So that's always been a little bit of the motivation in hindsight, that how do we scale what a one on one tutor would do?

What I was doing with Nadia.

So why has that not happened?

You know, she. she wrote me a letter. I didn't know who she was. I assume she's middle class. But then she says that some volunteers were set up computer lab in her orphanage in Mongolia.

Speaking of Mongolia, and she got hooked on Khan Academy. And then that allowed her not just only to to get her degree, but she's. she then became the top creator of content in the Mongolian language for Khan Academy. Another young woman in Afghanistan, Taliban keeps her from going to school. She's actually this.

Her name is. her name is Sola. She just wrote a book about it. But she, when she couldn't go to school, an uncle introduces her to Khan Academy. She was lucky enough to at least have an Internet connection.

She's middle class family. She uses that to self-educate herself, starting at age. I think she was about 12 years old by 16 or 17.. This is all she's doing, you know, for two hours a day when she's not doing her chores. And she decides she wants to become a quantum, a quantum researcher in the United States.

And so she she lies to her parents to go to Pakistan to take the SAT because it wasn't administered in in in the US. She does shockingly well. She's for the most part, educate only on Khan Academy. And that's what I found out about her. I try to figure out what colleges will take her without any formal education.

Nicholas Kristof of The New York Times writes about her. And then she was able to get a visa. ASU takes her. She rocks the physics department there. Now she's actually a quantum computing researcher at Tufts University.

So these stories are out there and we get a letter kind of like this on a weekly, monthly basis. I just we just got one last week of a of a young person in the UK. They had a child at a young age, dropped out, ended up becoming a construction worker, found Khan Academy, got hooked on it, just started talking to everyone they could find about physics and chemistry. And their manager said, hey, if you don't go to college, I'm going to fire you because you're just boring all of us talking about physics and chemistry all day. And this person is now, you know, first of all, met their their their, you know, life partner in college.

They graduated, you know, it sounds like straight A's and now they're entering a Ph.

D. program in biochemistry. So so this is out there. So that's one story that, for for those who are reasonably motivated and I'm not necessarily high, achieving because this last example was someone that the system was telling was was not achieving. But if they're motivated and that motivation might kick in at different stages of life, but if they're motivated and they at least have access to an Internet connection, a computer, an awareness of Khan Academy and other tools, there's MOOCs and other things we are seeing a little bit of a mini revolution of these people who would have otherwise been lost.

Now, for better or for worse, that that group of people that I just described is a small minority of the broader population. And so I think there's two tasks. One is how do we lower, how do we make it even easier and easier to do what these people just did, so that you don't have to have quite their determination to be able to pull it off. And then I think the other element is there are just a lot of learners. And I don't think they're inherently demotivated, but I think something has happened to them over the years where they have just given up.

And so I think there needs to be systems to reengage them. And I think that's primarily going to happen through the traditional school system. And so that's where we're working to do that.

We think this is going to be really interesting for a lot of people, but it also might be a little bit jarring, because it's going to be quite powerful. We want to launch with and partner with organizations that people trust that can show socially positive use cases of it. And, by their own words, Khan Academy was the first organization they thought of.

. I thought they were cool, super fascinating. But they were like. they were good at writing things that sounded thoughtful. But when you really tried to read it, you're like, this is this is meaningless or it's really not.

It doesn't really have a good handle on the information. It doesn't seem intelligent. But when that first, when they, when they, we, you know, they essentially shared their screen with us. And this was months before CHAT-GPT existed. And CHAT-GPT, when it launched, was based on GPT-3.5..

Here was GPT-4.

. And they showed me a biology question. And they said, Sal, what's the answer here? And I read it. I'm like, OK, the answer is C, osmosis.

And they said, OK. And then they, then they asked the AI, the AI said C. I'm like, OK, maybe got lucky. I said, ask it to explain why. And explained it.

Very good explanation. Perfect explanation. I'm like, OK, this is interesting. I said, explain why the other choices aren't right. Very good explanations.

Now I'm getting goosebumps. And I said, have it write another question like this. It did it. I'm like, uh, oh. And then I said, have it, write 10 more questions like this.

It did it. Now I'm like, OK, OK, this is a. this is like the world. I'm in a science fiction book. Am I dreaming right now?

And they said, hey, would you like access to it? I'm like, yes, please. And so, you know, we signed all the NDAs and everything. And so we got myself, our chief learning officer, our CTO, got access to it that weekend. We couldn't sleep.

And you could imagine it was. I have trouble keeping secrets. I'm like an open book on like everything. And like, I felt like the three of us had like the biggest secret on the planet that no one else knew about. You know, those were.

But we also realize there's a lot of fears here. How do you have a 12 year old? use an AI? And what if they have a weird conversation, inappropriate conversation, an unsafe conversation? What if they use it to cheat?

What if? how do we make sure there's data privacy safeguards, et cetera? So we were having these debates and we said, look, we can't shy away from this. Let's build this. But let's turn all of those fears and risks into features.

So let's make it so it's transparent to teachers. Let's make it so that another AI is monitoring. If it notices anything going funny, it'll shut down the conversation and actively notify teachers. You know, we have all sorts of other guardrails on it. It won't cheat.

It's more of a Socratic tutor. But it will nudge you forward with with leading questions. And so that's what we launched with Conmigo. We launched alongside the GPT-4 launch back in March of 2023.. And, you know, we launched it as a pilot and it still, in a lot of ways, is a pilot, but the response has been more positive and people are ramping into it much faster than I would have expected.

I could speak for an hour about all the features we're hoping to add in the next, in the next year. But it's a lot's happening there.

It's not even allowed for under 18 users, although I'm sure a lot of under 18 folks are said they're 18 and are using it. And so the important thing is to use these tools in an environment that is, one has all the right guardrails for education. And you're not using a tool for the sake of it. If we want to save teachers time on grading, on lesson planning, et cetera, they should be using a tool like like Conmigo. So it's it's.

it's really about embracing the underlying technology, but using it in a way that is thoughtful and purpose built for education.

Philanthropically, I remind my, I tell donors, hey, that's the budget of a large high school in the US. And, you know, we're educating a lot of folks and reaching a lot of folks. But I would probably take another 30 to 40 million a year and further accelerate a lot of the content and software and and product development we're doing. Our goal has always been, even before AI, to cover all of the core academic subjects and grades. We're close already, but I would go and that's from pre-K through the core of college.

So I would use some of that to accelerate into even more subject areas. We already have math. We already have a pretty comprehensive. I would go even further into college. Science were quite comprehensive.

I'd go even further into sophomore, junior level college. We're starting to have a pretty robust humanities and English language arts offering. I would try to accelerate that. And then, on the AI front, yeah, I would. I would use a lot of those resources to just accelerate it that much more.

But, you know, with that type of resources, I would also probably create national or international systems around credentialing, where the tools exist now, where you can learn a lot on your own or with support and those tools, and then you can even prove that you know it. But the the legacy systems don't fully recognize it yet. And so, you know, I've told people if I was secretary of education or minister of education for any country, the first thing I would do is create a competency based set of assessments that if anyone did it, these count as good or better than any university, any high school, any graduate program. And then you allow the world to create resources, organizations like Khan Academy to create resources to prepare people for those. Some people could use the Khan Academy.

Like, if you think about this young Afghani girl, some people like her could use the Khan Academy resources alone to prepare and then take these assessments, and then they have a degree that's better than any degree, but a lot of students would need more supports. Well, then that's where the traditional school system could use Khan Academy, plus all of the inhuman, the in-person supports to to actually support these students more. So that's where I would put a lot of resources. I still don't think that would get us fully to a billion dollars. But, you know, then, you know, maybe I would create a new type of university that is much lower cost and much more hands on and, you know, gives exposure to more kids.

we don't want to make this like just a. you got, you know, you took a MOOC course and you got a badge, and now you go from being a general computer scientist to a data scientist that exists already. And that's great that that exists, that that pathway. But we don't have a pathway right now. that says, hey, this is just as good as Harvard if you do this.

You know, right now in the U.

S. there's all of these debates about, you know, you have limited seats at these top universities. And so it's these debates between equity and merit. You know, do you just take the top two thousand kids at Harvard every year or do you make sure it represents the population, et cetera, et cetera? But then there's questions about, is it merit based or is it quota?

So people are getting, they're politically charged. And, you know, I've told many university presidents, I was like, look, this is a false tension. The real issue why you're, why you're having to do this game is because you cost too much and you don't have enough capacity. If, if you could bring the cost down, so it's almost free, like the actual cost is almost free. It doesn't have to have massive government subsidy.

And you make it so that any student who's capable of doing the work can do the work. Then anyone who wants to do it at any age can do it. And I think, you know,

if I had my druthers over the next 10 years, that will exist. And I am pitching that to folks every now and then.

And so in the short term, these kids and these families have a huge advantage. But what's happening, I would say, in the medium to long term, is that these, these families have always had a huge advantage. If their kids are struggling, they hire tutors for their kids and paid $50 an hour, $100 an hour. If they're, when their kids apply to college, they hire coaches for their kids. These families themselves are highly educated and oftentimes have more flexibility for travel and to tutor their kids themselves.

They know how the system works. I mean, these are all things, you know, I know how the system works. And so my kids are going to have that, that that advantage. Now, if you bring in an AI in, the AI overnight is not going to completely level that playing field, but it's going to allow, let's call it the high school version of me that did not have those advantages to be able to to, to tap in to more opportunity, to be able to learn things faster, to be able to have some. You know, my my guidance counselor in high school is great, but he had no experience.

There were 300 students for every guidance counselor. So I had like half an hour with him once a month or something. And he had no experience with, you know, I wanted to apply to MIT. No one in my school had ever gotten into MIT. So he had no context there.

Now we have an activity on Conmigo where you can talk to the AI. And we've had some of the top college coaches in the in the country help work with that AI to make sure that it's giving the best advice, it's really almost indistinguishable from what a top college advisor would now give. So that starts to level the playing field.

You're going to have to find a facility. You're going to have to maybe be flexible around. these kids might have other family responsibilities. Like it's. it was an expensive proposition and the world has been trying to do it for hundreds of years and frankly, hasn't been able to pull it off with tools like Khan Academy, especially as you start introducing the AI.

The problem becomes much more of, hey, can we get people, devices and Internet access at least to start? I'm not saying that that's equivalent to a full school, but at least it raises the floor pretty dramatically. And then, if you want to create a school, the problem is no longer you need to find someone who's a master of physics. You just. you just have to create a nice, safe environment, have access to resources.

The adults in the room just have to motivate, keep the kids motivated and on task, but then they could use some of these tools for some of the content expertise. It's great if those adults have it, but they don't have to have it. So I think it starts to level the playing field there. Once again, not going to completely level it, but it's going to be more level than it is today. The place where I'm most worried about AI, unleveling the playing field.

And we've we've to some degree seen this trend over the last 30, 40 years with with technology generally, is for those who know how to use these tools, it massively improves productivity. But what it does is it concentrates that productivity in a smaller, smaller group of people. What used to take. You know, 50 years ago, if I told you I have a even if you adjust for inflation, if you have the equivalent of a billion dollar company today, people would assume that company must have 100000 employees. When WhatsApp sold to Metta for 18 billion dollars, it had 18 employees.

And this is before. they weren't even using coding copilots to write that code. But 18 employees were able to create an 18 billion dollar company. I think over the next 10, 15 years, you're going to see five, 10 person companies using AI create billions of dollars in value. Now, that could be overall good for society.

It's going to create tools that increase productivity, or entertainment or whatever, but that a lot of that wealth is going to accrue to a very small, concentrated people. And if it used to take a hundred thousand people to do it and now you're able to do it with 10 people. Well, what happens to the other? What? what happens to the other ninety, nine thousand, nine hundred ninety people in terms of, in terms of work?

Now, there are some good utopian scenarios where it frees up resources for for more human centered work, for nursing, health care, education, teaching. But then there's dystopian scenarios where these people get marginalized. It destabilizes countries, maybe you have to do mass redistribution, and even mass redistribution, whatever people's views are on that. I don't think it's good for the people who get it because they lose their sense of purpose. They're like, well, I don't know what I'm, what I'm contributing.

So I think this is something we have to tackle. Honestly, the best way to tackle it is is to let as many of those people as possible learn how to use these technologies and learn how to how to leverage them.

Now, how do you, how do you run the organization? What are your leadership principles?

you know, what I implicitly like to happen is, you know, try to be as close to the users as possible. This is something I still do. I still make videos. I still, you know, one reason why I started a school, the one school my kids go to, is so that I can, I can be close to learners and understand what they're and teachers and what they're doing. So I try to be as close to the users and understand what their lives are like.

A flat organization as much as possible. Right now, you know, there's, you know, one, we have direct lines of communication between me and any member of the organization. But, you know, there's. there's only two levels of management at most. But even then, it's fairly porous.

So we try to do that. We try to keep pushing ourselves to avoid bureaucracy. You do need some processes as you grow and you scale, to have drive alignment and make sure things are happening as they should.

You have first principles thinking around, like, do we need that meeting? Does it need to be run in that way?

Are there ways that we could do this different, so that we we can get more skill and leverage? And, you know, one of the things I found, a lot of times people think about the size of an organization. And it is true. The main cost in organizations these days is salaries. But one of the reasons to keep an organization as small as possible isn't necessarily like cost savings.

It's actually communication and being able to be nimble. If Khan Academy last year was a 20,000 person organization, I don't think there was any way that I would have been able to make this move into AI. that we did, even with 250 people. It wasn't easy. It's still, you know, we're not a.

we're not a speedboat. We're kind of a, I don't know, like a catamaran, like, you know, but we're not a cruise ship either. And that cruise ship was very hard to change or turn around when you realize that there's an iceberg, if you don't, if you, if you don't turn around. So so I think in going forward, that's going to be even more important than than, you know, it's not about saving money and obviously we are limited by resources because we raise primarily through philanthropy, but it's more about being able to stay nimble.

Part of me, you know, when I went into college, I thought I wanted to be a theoretical physicist, because I just wanted to understand the nature of reality.

Now I've kind of reached the conclusion. I should just meditate my way there. But theoretical physics is still, is still an area that I you know, my oldest son is kind of almost caught up with me in terms of his math and science. And I said, OK, when you catch up with me, we're going to start studying quantum physics together, so that I can, I can, you know, I want to understand the frontiers of reality, of how we, how we understand reality.

D. or two master's degrees?

But I knew I wanted to do something in industry. At least I thought I did. It's funny. The only class I HB at Harvard doesn't fail you and anyone in the business school, but they tell you internally that you would have failed. The only class that I was very cynical about that I kind of got the equivalent of a failing grade was called social entrepreneurship.

And.

and it was because I was such a cynic in that class. I you know, a lot of the cases we studied were, you know, someone's organizing a bicycle ride to cure some disease. And I remember speaking up in class and saying, well, that's not going to cure the disease. It's just a bicycle ride. Like, how is that going to cure the disease?

And most of the money is going to fund all of the, the pageantry around the ride. So you could and I later realized the professor organized that that charity, so it was not good. But the and I now come around, I now see the importance of those types of things because they bring awareness to to an effort. But it is. there is an irony here.

I go to business school, fail social entrepreneurship. And that's what I do now.

They should not make seventh graders read that, especially seventh grade boys read that unless they really want to. They shouldn't force anyone to do it, because you don't appreciate it. I read it just now. I'm like, this is the most amazing book. There's like so much wisdom and wit.

It's hilarious and it's modern in so many ways. So I, you know, I flip flop between revisiting the great classics that I was forced to read in school. And now I have, I can read it properly. So I love doing that. And then I, you know, and I read a lot of hard science fiction.

You know, I've always been fans of, like Asimov and Arthur C. Clark. And and now I'm just making sure I'm reading all of their books. You know, I've read their their most notable books, but I'm reading all of them.

And they're like, I wish someone exposed me, exposed this to me this way when I was 12.

. So I would just say just start learning. And it doesn't have to be through Khan Academy. Obviously, Khan Academy is there. But just reading, exploring, you know, a lot of these A.I.

tools are fascinating. to have conversations with. a conmigo. You can have a Socratic dialogue. You can get into debates.

I personally really enjoy it when I, when I have time to to dig into a debate that I'm, you know, I'm dying to have, but I can't find someone to have with me.

So big congratulations to real life class.