Khan Academy Founder Sal Khan on the Future of Learning

MOLLY WOOD: Today I’m talking to Sal Khan, founder of the Khan Academy and author of several books, including the newly released Brave New Words: How AI Will Revolutionize Education (and Why That’s a Good Thing). Sal has been blazing trails in education technology for decades, and he has a unique vantage point on how smart companies can leverage AI-powered tools to transform their approach to teaching and skilling at work. He says AI can make on-the-job learning more accessible, more responsive, more personalized, and more effective. We’ll also delve into the latest scientific insights into how we learn, and explain why the new AI era is all about enhancing human interaction and not replacing it. Here’s my conversation with Sal. 

[Music]  

MOLLY WOOD: Sal, as a mom, I am thrilled to talk to you because my son has been using Khan Academy for years, and he says you’re the reason he passed chemistry. But for those who aren’t as familiar with Khan Academy, can you just give us a brief overview of it?   

SAL KHAN: Sure. Khan Academy, we’re a not-for-profit. Our mission is free world-class education for anyone, anywhere. And the way that we’ve tried to do that over the last 15-plus years is by offering software, personalized exercises, teacher tools. A lot of folks know about our videos, everything from pre-K through college-level math, science, humanities. It really started with me tutoring family and it’s kind of grown in a way that we’re trying to use technology to scale the type of personalization that you can get with tutoring. You know, if you look at most of human history, we’ve always known what the gold standard in learning is. You could go back 2,300 years to Alexander the Great. His tutor was Aristotle. And that was an exceptionally good tutor. But through most of human history, if you were privileged enough to get an education, you usually got something like that, a tutor, or sometimes even an army of tutors that would work with you. If you needed to speed up, they’d speed up. If you needed to slow down, they’d slow down. You fast-forward to about two, three hundred years ago, we had a very utopian idea: mass public education. It’s had a lot of positive outcomes because it’s reaching everyone, but it’s also left a lot of folks behind. So the opportunity with technology has been, we have computers, we have the internet. We have on-demand video, and now we have artificial intelligence. There are now ways to increasingly approximate the type of one-on-one personalized experience that the best educations have always been throughout history, but to do it at scale.  

MOLLY WOOD: I really want to hear your insights about how AI will transform skilling and learning on the job. But first, can you tell me how you’re incorporating AI into your organization?   

SAL KHAN: Yeah, AI has always been a little bit of our conversation. We’ve always played with it for things like recommendations or making inferences based on student behavior of what might be the next best thing for them to do. But OpenAI reached out to us back in summer of 2022. And when we saw, months before anyone else did, what GPT-4 was capable of, that really opened our aperture to, you know, things that I didn’t think were going to happen in our lifetimes. We had read about these things in science-fiction books. We very quickly were able to get, with enough prompting and some extra work on just the infrastructure side, to get these models to really not just pretend to be, but in very substantive ways, do what a great tutor would do. Have a conversation, be Socratic. And even do things that a tutor wouldn’t do. Emulate a historical figure, run a simulation. And when we saw that, we said, look, this is the future of Khan Academy. And so there’s obviously a lot of things to address. How do you deal with safety and privacy, especially if you’re using it, you know, for under-18 students; how do you deal with cheating, et cetera. But this was definitely something that we said, okay, those are risks, but let’s invest here. And ever since then, we’ve been investing a ton on that front.   

MOLLY WOOD: So much so that you actually created your own generative AI tool, right?   

SAL KHAN: Yeah. When we saw that, you know, just a model like GPT-4 with a little bit of prompting, could start to really act very similar to the way I used to tutor my cousins when I used to chat with them remotely. But then, it did have issues with math, it did have hallucinations. You know, we need some type of safety oversight, you need to somehow keep students from cheating. So we started building a tool that would do all of these things. Essentially it could be leveraged in an education setting to be a tutor for any student and a teaching assistant for any teacher—help them with lesson plans, grading, et cetera, et cetera. And so we were initially under a non-disclosure agreement with OpenAI for much of 2022. And then when GPT-4 released on March 15th of 2023, that’s when we released Khanmigo as well, which does everything I just said, everything that students are doing, it’s accessible by the teacher, it actively notifies parents and teachers if something seems amiss. It’s really optimized for things like tutoring. It’s Socratic. It won’t cheat, but it will nudge you forward in the right way. It has the context of everything you’re doing on Khan Academy, but it does things that, you know, a typical tutor wouldn’t do. It can emulate historical figures, simulations that you can enter into, talk to literary characters. You can have debates with the AI. But we’re doing everything in a setting so that it really works for schools. Schools can get dashboards to understand how the AIs are being used. Teachers can get reports on it, et cetera. So that’s, you know, it’s evolving daily on what it is.   

MOLLY WOOD: In fact, I understand that at the recent Build event, Microsoft announced a new partnership with Khan Academy to scale Khanmigo—so clearly it’s evolving into big things. Sal, you’re obviously somebody who thinks deeply about learning. We naturally think about kids when we think of education, but learning, skilling, upskilling, reskilling—all of this is so important to the modern workplace and business leaders. What do you think in that context about the changes that are coming to the way we learn?   

SAL KHAN: Well, shameless plug, I have a book out now on exactly this topic. In fact, there’s a chapter explicitly on this. It’s called Brave New Words: How AI Will Revolutionize Education (and Why That’s a Good Thing). And I think it affects that as much as K-12 education. Even before AI, things like online learning become very relevant for the workplace because you’re not bound by time or space. I remember when I first made some videos for my cousins back in 2006, they said they liked me better on YouTube than in person. What they were saying, I believe, was they appreciated me in their life but they also appreciated when it was 11 p.m. and they had a question that they could access the videos—that there was no stupid question to a video. They didn’t feel embarrassed if they had to review something that maybe someone of their age should have known. I think that becomes even more relevant in the workplace when you’re dealing with adults. Obviously adults have even more going on in their life between work and family. I think many of us adults are even more embarrassed to ask what we might think is not a smart question. I think any time any of us have started a new job, you know, you have about a week to say, what does that acronym mean? Or what does this acronym mean? Or wait, what? I don’t get that part of it. And then after that, you kind of say, well, maybe I should know it and you stop asking, which is really bad. You should ask because it turns out most people are probably wondering the same question. But one of the critiques of, let’s call it pre-generative AI, online learning is you couldn’t really engage in terms of asking questions or edge cases. But now you can, you can watch a Khan Academy video or any type of workplace video now and ask follow-up questions. We’ve all seen these workplace training videos—they’re getting better, thank god. [Laughs]—you know, about various policies and security and all of that. But now you can ask questions like, hey, what about this scenario? I think it just allows for a lot more richness: that same one-on-one tutoring type of experience, on-demand anywhere. You’re going to have that in workplace-type knowledge too.   

MOLLY WOOD: Now I want to ask you about the second part of the subtitle of your book, and Why That’s a Good Thing. Expand a little more, if you would, on why you think this really is valuable at work.   

SAL KHAN: I am, in particular, not only optimistic about the education lane, but I think we have to be optimistic about it. And we have to lean in there, because to a large degree that’s the only way that we can mitigate some of our fears that we have in these other parts of life. I talk a lot about using artificial intelligence to improve HI, to improve human intelligence, to improve human purpose. Because what people are going to need to do in the future—obviously they’re going to be augmented by these artificial intelligences—but at the end of the day, they’re just a magnification of human intent. And we’re all going to go quicker, you know, a recent college grad who stays a software engineer, they’re going to go very quickly from entry-level software engineer to software architect, where they’re really managing AIs. But in order to manage it, you have to be able to understand the code and the architecture, at least as well as the AI can. If you’re going to be using AI to help write—if you’re an editor of a magazine, you need to be able to write and have a sense of what great looks like at least as well as the junior writers, and the junior writers of the future are going to be AIs. And so it’s even more important than ever for people to really upskill, have the proficiency, the mastery of the traditional three R’s, reading, writing, arithmetic—it’s always frustrated me that they called it ’rithmetic with an R—   

MOLLY WOOD: Me too. I see you on that pedantic-ness. [Laughter]  

SAL KHAN: But beyond that, you know, really becoming excellent at whatever your craft is, and probably more than one craft. We are going to have pseudo-godlike powers in a lot of these things, but the people who know how to put these pieces together, know what excellent looks like, they are the ones who are going to be really empowered. And to only get there is education. And the good news is, these AIs can be used to act as a personal tutor, they can really help answer any questions you have, you can learn through simulations as opposed to through multiple-choice questions. Multiple-choice questions might be part of it, but now you can do much deeper things in a learning setting using generative AI. So I think the same technology that’s going to introduce some shifts in society can also help us navigate those shifts.  

MOLLY WOOD: I love this idea of junior employees and AI being sort of a rapid up-leveler for them, because you’ve also talked about how AI can help us sort of move education away from drills and rote memorization, into more creative areas. What will it take, do you think, for employees to kind of grasp that potential?   

SAL KHAN: Well, I think you’re already seeing in the workplace, there are employees that are embracing it. I think you’re seeing it most clearly in coding. You can see the difference between the engineers who are using the AI copilots. Now, once again, it’s not writing all the code for them, but it is, you know, the autocomplete is awfully powerful now with coding. It can write the whole loop, write the whole function. You know, we’re seeing this at Khan Academy. I talk to a lot of friends who run engineering teams, who run start-ups. And I’ve heard folks say anywhere from 50 percent increased productivity already to two, three X productivity. I’m encouraging anyone at Khan Academy who’s writing. It helps you go from zero to first draft very fast. As a nonprofit we’re always writing grants, writing proposals. Now we can do our brain dumps from a conversation, sometimes even a transcript of a conversation or an email thread, and have a first draft of a proposal within five minutes. And then very quickly, the whole team, we can stick it in some type of a shared document, and the whole team can hammer that, and within a day it might look very different than that shared first draft, but what used to take a week or two is now taking a few hours. Every position in any company is some combination of writing, presenting, spreadsheet making, coding—and even design, image creation, video creation, video editing. All of that can be accelerated with generative AI already.  

MOLLY WOOD: Can we, in your view, also upskill creativity?   

SAL KHAN: Simple answer is yes. And there’s another chapter about that in Brave New Words. [Laughter] And, you know, at first, people’s gut reaction when they see something that could write a novel story or write a poem or create an image in the style of Van Gogh, it’s like, oh, this is going to undermine creativity. But what I remind folks in the book, first of all, people had these types of reactions when they first saw the camera. The camera actually helped release painting from the mundane. It allowed the portrait artists to become cubists and impressionists as opposed to, let me just capture things as realistically as possible. Let me capture my emotion, my impression. And it created a whole new art form. The other thing I point out is, if I’m alone, I am less creative than when I’m in a room with other creative people. And then I will extend that to other creative entities now. You know, the most creative people in the most creative times in history: you can imagine people in a coffee shop in Paris, you could imagine painters hanging out in Renaissance Italy. They riff off of each other versus, you know, somehow saying, oh, you can do that? Then I don’t need to do that anymore. And so, we’ve already seen this with Khanmigo. When students are able to riff with the AI, they become more creative. And then we’re going to have more outlets for that creative expression. The same way that the internet and things like YouTube—we didn’t have to go through gatekeepers anymore to publish our creative expression. We’re entering a world where, if you have a great idea as a writer, you’re not going to have to go through the studio to get your movie produced. You are going to be able to generate the imagery. You’re going to be able to edit yourself. A movie that used to cost 100 million dollars is going to cost 10,000 dollars to produce. So you’re going to see a lot more content. The creatives are going to be the ones in charge as opposed to the gatekeepers.   

MOLLY WOOD: So in your book, you write that deep and broad skills are going to become even more important in the AI era, as well as soft skills like communication, collaboration, and empathy. So one, we have to teach those skills, especially to the leaders of the future, but we also have to filter for them as business leaders today. And I wonder if you have advice about how you even look for that.   

SAL KHAN: As technology empowers us to do more and more and more, you know, we’ve already said you have to get at least as or comparable to the AI in order to be able to manage the AI and be able to use it well. What I also write about is, the people who are most empowered are going to be the people who can go deep in several of these areas but then also put those pieces from the different areas together. I give the example of even before generative AI. Let’s call it 50 years ago. If you wanted to start an education company, you would have had to hire a lot of people and get a bunch of capital to put it together to even get started. The reason why I was able to do Khan Academy back in 2006, 2007, 2008, one, the tools got better. Things like computers, screen-capture software, the internet, on-demand video, all of these things, the programming languages became easier and easier to use, the cloud services. But I had to be deep and broad enough to put all those pieces together. We talk about when I incorporated as a not-for-profit, I had to know enough about incorporation and law and finance, so I didn’t have to go hire lawyers to do or accountants to do that part. I had to know enough about coding to be able to start putting that first software together. I had to know enough about video creation and explanation. And as the tools become more and more powerful, we’re going to see that dynamic even more and more. So as we expand to AI, someone who isn’t just an expert at coding, isn’t just an expert at writing, but someone who can put those pieces together, they’re going to have a huge advantage. And then the soft skills, whether it’s empathy or just being able to communicate in multiple forms, oral communication, written communication. Because no matter what, you’re still going to have to sell what you’re doing, whether it’s your own skills or whether it’s some kind of product or service that you’re actually producing. Now one of the things I’m really excited about, and I write about this as well, it’s another chapter on the future of hiring and assessment, is traditional assessment, it’s very hard to do something fairly broad. You can give someone a multiple-choice test, et cetera, et cetera. But if you want to do something richer, we’ve had these very intensive interview-type processes that most of us are familiar with. The problem with those are they’re very inconsistent. In the future with AI, and obviously we have to be very careful to make sure that the AI isn’t introducing new biases or new biases that we don’t want. Some biases are good. You want to hire someone who’s going to be hardworking. You want to hire someone who knows the skills, but you don’t want to filter out someone based on their gender or their age or their ethnicity or their religion or something like that. But you can imagine more scalable ways to do a first pass of assessment of a much broader group of people. We see this at Khan Academy, every time we put a job posting out, there’s three, four hundred resumes come in. Our hiring managers, our recruiters, they obviously try to be as equitable as possible. They try to look—but if you’re going through four hundred resumes, realistically you’re spending a few seconds on each one. It’s human nature. So they did kind of, I see that word that’s in the job description, maybe. And so you whittle it down to 20 or 30 people to get these first-round interviews, but you probably missed a hundred people or maybe 50 people who are also good, but you know—our brains, our neural nets didn’t pattern match on those resumes when we were looking at it for 15 or 20 seconds. We’re entering a world now where we can give a richer set of evaluations to the whole applicant pool, which I think is very exciting. You could also imagine a world, you know, a lot of jobs, they give you these case-based interview questions. You know, how many pennies can you fit inside of a 747? And they really want you to figure out, are you asking the right questions? Are you good at estimation? Things like that. You can, in fact, we’ve been working on this. You can imagine an AI running those types of case-based questions or interviews, and then giving an evaluation and then filter the folks who are doing really well on it. And for the folks who are worried about the bias, and I am too, the beauty of these AIs is you can stress test them and make sure that they’re not doing an incorrect or inappropriate bias in ways that you can’t do with human beings.   

MOLLY WOOD: Right. And then I want to ask you about the kind of learning paradigm, because so much, you know, when we talk about upskilling, especially at work, it is this sort of paradigm of lifelong learning. Now there is broad access to tools that can enable that. Do you see people thinking about learning differently or wanting to continue to learn differently as they get access to these tools and realize what might be available to them?   

SAL KHAN: I think it’s just starting. I’m still amazed how much of corporate learning, say, or on-the-job learning is still fairly traditional, where people still go and take a bunch of time off and go sit in a classroom, et cetera, et cetera. A lot of that has obviously started to move online. So even though online learning has been a thing for 20-plus years, it’s only now really catching on, and especially in a corporate environment. I think with AI, this is a world where a corporation could slurp in their entire career rubric, their entire competency frameworks, and the AI, knowing those things, you could do a check-in with employees and their managers and the people they work with on a regular basis. I’ve been telling our team, you know, we still do the classic employee survey. That’s nice. There’s certain metrics you can get from that. When people write free responses, we kind of look at it and try to make sense of it. But I was like, well, couldn’t we use generative AI now to do a richer survey? To essentially ask people and ask follow-up questions and then have the AI synthesize the key insights for us so it becomes a little bit more actionable than someone from HR just trying to read 500 comments and trying to make sense of it. And you could imagine in an organization like Microsoft, instead of 500, you might have 500,000 comments. How do you distill that? Well, that’s what AIs can do.   

MOLLY WOOD: And then that also becomes a leadership question. As you see people start to adopt AI-related skills, add them to their LinkedIn profiles, how can, and how should, do you think, leaders help people who may be a little bit less AI literate, let’s say?   

SAL KHAN: I think you just have to encourage folks to use it and give them space for failure. I’ve been trying to do this with our team and I haven’t had perfect results yet, but I’m telling them, look, if you’re not using it at least a little bit, or at least thinking about maybe it’ll be useful, you’re going to be in trouble in a couple of years. And I’m not saying this just for your productivity at Khan Academy. I’m saying this for your value as an employee, generally. And the more that you lean into these tools, your market value is going to go up dramatically. So one, encourage folks to use it. Maybe it’ll make them a little bit less productive initially, but it’s going to very quickly catch up. We’ve also played with things like giving our team special days, or like a week where the whole point of that week is, this is the week that you figure out how to use better tools, a lot of which could be generative AI, or even, we start to develop tools that might help different parts of the organization.  

MOLLY WOOD: That’s amazing. Both the space and the concept of the space for failure, because it also feels like these learning opportunities and the access to the skilling is a great way to retain talent. What do you think about, you know, it sounds like you’re being very intentional. Some of the best companies are being very intentional about offering access to these tools, about adopting them. In our research, we’re also seeing, though, that people who don’t have access to these kind of officially sanctioned tools are finding a way. They are finding a way around it and that that’s also happening in learning environments. What do you make of that phenomenon and what advice would you give to organizations about that?  

SAL KHAN: Yeah, to your point, it’s happening anyway. And it’s better when it’s where people know that it’s happening. And there are some risks that people should be aware of. I think we’ve all read that article about that lawyer who wrote a case brief and, you know, using an AI and it hallucinated and they got in trouble. And so you don’t want people in your organization misunderstanding how to use these tools. They go to ChatGPT or they use Copilot or something, and they just take that first draft and they just copy and paste and they send it to a client. That can become very embarrassing for the company. It could even be a legal issue, it could expose a liability. So, as optimistic as I am, and I want people to lean into it, I think it’s best done when the organization is saying, hey, but here are things to watch out for. Here’s what a good use of it looks like, and here’s what a shady use looks like, so don’t try to go there. Then you’re going to have the best, you know, you’re going to get the benefits and mitigate the risks.  

MOLLY WOOD: And then one last question for you, because we came across this term in your book. You said that we need to approach this new AI era with educated bravery, which I think is such a wonderful phrase. And I hope you could just expand a little more on what you mean by that.   

SAL KHAN: Yeah, you know, being purely brave sounds good, but sometimes it isn’t. Sometimes you’re doing things that are foolhardy. You know, brave could mean, hey, even if there’s reasons to be afraid, you kind of ignore them and then you can run forward. And on the other side of it, you can sometimes be educated, but I’ve seen that some of the most educated people are really great at talking themselves out of something, or really indexing on all the risks and then not wanting to take action because they’re so paralyzed by those risks. And so educated bravery is really recognizing that tension, but using information, getting educated about what the real risks are, what the pitfalls are, but not using that as a reason to not take action. Write them all down, turn them into features and move forward and be brave about how you’re doing it. Because, and I think this is true with generative AI more than almost any other technology, the risks of inaction are much higher than any of the risks of action, especially as long as you take that action in a reasonably educated way.   

MOLLY WOOD: Sal Khan. Thank you so much for the time today. We are really grateful. This is fascinating. 

SAL KHAN: Thanks for having me, Molly.  

[Music] 

MOLLY WOOD: Thank you again to Sal Khan, founder of the Khan Academy, and what a wonderful way to end this season of WorkLab, the podcast from Microsoft. And thanks to all of you for joining me in these incredible conversations. Don’t worry, we’re coming back with another season. Please subscribe and check back for that. We’ll continue to explore AI’s potential to transform every aspect of how we work. If you’ve got a question or a comment, please drop us an email at worklab@microsoft.com, and check out Microsoft’s Work Trend Indexes and the WorkLab digital publication, where you’ll find all of our episodes along with thoughtful stories that explore how business leaders are thriving in today’s new world of work. You can find all of it at microsoft.com/WorkLab. As for this podcast, please rate us, review us, and follow us wherever you listen. It helps us out a ton. The WorkLab podcast is a place for experts to share their insights and opinions. As students of the future of work, Microsoft values inputs from a diverse set of voices. That said, the opinions and findings of our guests are their own, and they may not necessarily reflect Microsoft’s own research or positions. WorkLab is produced by Microsoft with Godfrey Dadich Partners and Reasonable Volume. I’m your host, Molly Wood. Sharon Kallander and Matthew Duncan produced this podcast. Jessica Voelker is the WorkLab editor.