Khan Academy’s founder on AI: “Hope isn’t a strategy” | Sal Khan

Transcript

Introduction

Nate: Good morning, good afternoon, good evening, everybody. Welcome to another episode of the Getting Smart podcast, where we talk about the future of education. Here at Getting Smart, we’re laser-focused on building the future so that all learners will thrive. And like everyone else, AI has been top of mind in terms of disruption.

We know investment in technology is many, many times higher than investment in how that technology will be used in education. Speed of development is much faster than the speed of education changing. And then we have this big leap, just like the internet and social media’s previous big leaps, where AI will be deeply immersed in everyone’s lives, as it already is, and is going to become more so.

So as we think about learning in schools in our podcast today, we see four focal areas for generative technology like AI. There’s something around efficiency, something around safety, security, and policy, and all that work schools are trying to do to adapt. There’s definitely something about learning and how it will enhance learning. And there’s something about what AI and AI literacy are in general. So we have one big fundamental question in addition to those four areas: will AI upend education as we know it, or will it just be absorbed into the existing structures and systems?

Today we’re talking to Sal Khan. He’s been thinking about AI and learning and technology for a number of years. He’s the CEO of Khan Academy, which he founded in 2008, providing a free, world-class education for anyone, anywhere. He has so many other ventures that’ll come up today, ventures like Khan Lab School, Khan World School, and Khanmigo, the AI tool inside Khan Academy.

He’s written a couple of books, and I think he has a new book coming out called Job Shock, really thinking, as an entrepreneur, about how we make teaching and learning accessible for everybody. So Sal, welcome to the podcast. Excited to have you here today.

Sal Khan: Great to be here. Thanks for having me, Nate.

A Mixed-Age Classroom in Louisiana

Nate: All right, first question I ask all of our guests: when you think back on your K-12 experience, what was the best learning experience you had, in terms of feeling most engaged, feeling like you belonged, and learning a lot? What was that experience?

Sal Khan: Yeah, there’s a bunch. The one that jumps out right when you ask is second grade. I give my sister a lot of credit, she was three years older than me and a very strong student, and she was put into what they used to call the gifted and talented enrichment program. This is Louisiana public school. I remember in first grade she used to tell me about it, that she’d get taken out of class every day and go to this enrichment center. In first grade, I thought I was in that program too. It turns out it was speech therapy.

But by second grade, because I was Farrah’s younger brother, I think they figured there must be something here, and they tested me, and I got into that gifted and talented program in second grade.

I remember, this is really the only thing I remember from second grade. I walked into a classroom, and the seats were all over the place. There were kids of all different ages in that room. There were two teachers at the front of the classroom, and they weren’t facing the students, they were facing each other. As soon as they saw me, seven-year-old Sal, walk in, they said, “Hey, okay,” and called me forward. Miss Kraus and Miss Roussel said, “So, I think you’re Sal, right?” And I said, “Yeah, I’m Sal.” And they said, “So what are you into?” And I was like, “You’re asking me?” They said, “Yeah, what are you into?” I said, “Well, I like to draw.” They said, “Okay, well, do some drawing.” “I like games.” “Okay, we’ll play some games.” I’m like, “What is this place?”

But when I really think about it, that experience started in second grade and continued through the rest of elementary school. It’s where I learned the most. A lot of the things I now preach, that class time shouldn’t be lecture-based, that it should be active learning, that you can get a lot from a mixed-age learning environment, all come from there. I learned so many artistic skills, programming skills, chess skills from the older kids in that class who happened to be pulled out at the same time.

We did a very basic version of mastery learning. There were these things called mind benders, deductive reasoning problems, essentially like LSAT questions. I thought for a while I wanted to go to law school, and when I actually took the LSAT, those mind benders I did in third grade were the same kind of problems most pre-law students struggle with. The teacher would make photocopies, you’d do level one, and as soon as you thought you’d figured it out, you’d show it to her, and she’d say, “Okay, great, now go get level two from the filing cabinet.” So it was personalized mastery learning. If I didn’t get it right, I’d keep working on level one.

That experience still stands out to me as an ideal learning experience.

Nate: I love that. A couple things you noted: you walked into the classroom and felt like you belonged. There’s the idea of mastery learning, which we know from research does work when done well, but it’s very hard to implement across the scale of K-12 and higher ed. I also think about agency and interest-based learning, you had a lot of things that were of high interest to you, and when our brains are engaged and interested, we engage more and learn more. So thanks for sharing that.

Let’s jump into the future of work. You’ve been thinking about this, and we’ll get to the learning piece too, but you’re writing a new book called Job Shock, correct?

Sal Khan: That’s right.

Nate: Is that out yet, or when’s it coming out?

Sal Khan: I owe the manuscript to the publisher this week.

Nate: Okay, so coming out soon.

Sal Khan: Yeah, probably spring 2027.

The AI Job Apocalypse, or a Demographic Job Cliff?

Nate: Perfect. So I think the big question we’re wondering about, and I know you’re thinking about deeply, is this: you have people saying there’s going to be an AI job apocalypse. Then you have people like the Burning Glass Institute saying, “Actually, this isn’t the issue, we’ve got a demographic change happening, and there’s going to be a lot of job openings.” There are these counter camps. So what’s your thesis? What’s going to happen with AI, and what are the important skills students need to think about as they graduate and move into the world?

Sal Khan: My current thesis, I’m interviewing a bunch of people with very diverse points of view here, but my best guess is that in a five-year timeframe, you’re going to have pockets of pain. I don’t think it’s going to be a jobpocalypse. We can brainstorm many areas where I think jobs are very safe, even areas people have worried about for years, like radiology.

Even before generative AI, we had AI that could look at films and diagnose pretty well, and people said radiology was going to go away. It turns out radiologists assisted by AI can now go through more films. The cost of radiology has gone down. One person I interviewed is the head of the Mayo Clinic, and he said, “We’re doing more radiology than ever, and there’s a shortage of radiologist techs,” the people who walk you into the MRI machine and make sure you’ve taken your metal off. AI created a job because it lowered the cost of doing that thing.

So I think anything human-facing, teaching, sales, the care economy, hospitality, there’s no way AI takes that away. Even if someone argues robots will be your waiter, I’d point out that going to a restaurant has never been the most efficient or cheapest way to get a good meal. We’re willing to pay a premium for it. We want to go to the restaurants where we have to wait in line, that are crowded, where it’s hard to even hear. We like that energy, we like having a conversation with the server.

But there are going to be real pain points. I ask everyone I interview, “What about the Uber driver?” Their answer is usually, “Well, I don’t know if Waymo can produce fast enough to meet all that demand.” Fair enough, but it is going to take some of that demand away, and being a rideshare driver isn’t most people’s first-choice job. My uncle is a rideshare driver, and it’s not his first choice, he kind of fell into it. I’m not sure what folks like my uncle are going to do next, so I think there’s real pain there.

It’s not a small thing. Driving some form of vehicle is one of the largest employers of men globally between ages eighteen and fifty-five, not just Uber and taxi drivers, but delivery and so on. If you have a large fraction of men feeling useless, that’s a societal problem.

Call center workers, we’re already seeing that getting automated very quickly. Some economies, like the Philippines, have six to eight percent of their GDP in business process outsourcing, mostly call centers, getting automated fast. I don’t know where those people are going to go, and that has follow-on effects on their entire economy.

There are open questions around software engineering too, which AI does very well. The consensus in Silicon Valley, and I just interviewed Reid Hoffman yesterday, actually, and I agree with him on this, is that the top software engineers, the systems thinkers, the creative ones, are going to do great. If anything, some of them are already making NBA-player-type salaries. If you’re the kind of engineer who understands the customer and has strong communication skills, you’re going to be great.

But if you’re someone who just says, “I can migrate your old code base from COBOL to C++, don’t talk to me, just give me the spec,” you’re already getting replaced by generative AI. So there are going to be mixed pockets, but even what I’m painting is actually the optimistic scenario. Even that’s a pretty tough scenario where you could see structural unemployment of five to ten percent in a lot of places where people thought they had secure white-collar employment. The negative case is you go five or ten years into the future and fifty to sixty percent of white-collar jobs have been automated away, which is a very tough scenario.

We don’t know where we’re going to fall, but I tell everyone hope is not a strategy. We should have some type of plan in place to help people navigate this.

Nate: Right. We’re seeing the same thing, there’s safety in the trades, safety in human connection, healthcare, mental health, teaching, probably safety in AI infrastructure and cyber, less so in coding and around UX and background code, we’re doing a lot of internal coding now at Getting Smart using Claude.

Durable Skills and the New Khan-TED Institute

Nate: So we acknowledge there are going to be pockets of pain, whether optimistically at a lower level or pessimistically at a higher level. You’ve talked a lot about durable skills, the whole sector is talking about durable skills. I know you have an emerging partnership with ETS for a new undergraduate-type program. Do you believe in durable skills? Can they be replicated by AI, or will they be uniquely human?

Sal Khan: I’ll define durable skills first, because the name is a little misleading. When you just hear “durable skills,” it sounds like any skill you keep, that doesn’t go away. But when people use the term now, they’re talking about what used to be called soft skills: communication, collaboration, empathy, resilience, creativity.

We’ve always known these skills are very important. A lot of the argument for why in-person schooling matters, which I tend to agree with, is that you get all this socialization. You learn how to resolve conflict. If you just wanted to learn math as fast as possible, you could sit at home with a parent making sure you did your Khan Academy and some extra problems, and you’d learn a lot of math, quickly. But you wouldn’t learn how to deal with a bully, how to resolve conflicts, how to give a presentation in class, how to take feedback when a peer or teacher gives you something you don’t want to hear. Those really are the things that differentiate whether someone succeeds in the job market or not.

The interview process, which most people don’t think of this way, really is an assessment, a very imperfect, inconsistent, non-psychometric one, but they’re usually going after those durable skills. If they just cared about intellectual horsepower, they’d say, “What’s your SAT score? Great, show up.” Instead, they’re trying to see: can you communicate, how do you deal with conflict, are you a job hopper. The reason we’ve historically called them soft skills is that they were hard to measure, it could feel subjective, just one person’s opinion.

But what’s interesting, and we’ve already seen this in some pre-generative-AI projects, is that it’s not purely subjective. It might be subjective what Nate thinks of Sal, but if there are 200 people who’ve worked with me and you start norming their feedback, it becomes more objective about what my communication or collaboration skills are actually like.

Now AI opens up a whole other front we have to be careful with, but it’s also powerful. I’ve seen assessment platforms where you upload a video and get pretty good feedback on how good a speech you gave, whether you maintained eye contact. We’re looking at assessments with ETS, and other parties, that can monitor how you’re using a computer, even how you’re using AI tools, and use AI to say, “This person was a really good problem solver.” My inner cynical seventh-grader always says, “There must be a way to game this, to give the AI what it wants.” But the technology is getting very, very good, so I’m cautiously optimistic about broadening the aperture of what we can assess, both durable skills and traditional academic skills.

Take the SAT: people say the score is good, but it doesn’t measure all your academic skills. You don’t really write on the SAT anymore, and the math questions are mostly multiple choice. What if you had to do a proof? That’s expensive to grade by hand, which is why the SAT doesn’t ask for it, but generative AI could make that possible.

So what we’re creating with Khan Academy, TED, and ETS is what we’re calling the Khan-TED Institute. We’ve publicly announced a bachelor’s program, and we’re likely to also do a certificate and a master’s program. We have an interest list of over 30,000 people now, and most of them actually already have college degrees. What they’re telling us is, “I’m really afraid of losing my job, I want to make sure I’m building these skills, and I want to be able to showcase them to employers.” We’re also talking to major employers, Bain, Accenture, Microsoft, Google, and they want these signals too. Their old way of doing things, looking at a resume with bullet points, or seeing someone with a high GPA from a top university, doesn’t tell them anything about durable skills like communication or collaboration. They try to get at that in an interview, but it’s limited and inconsistent.

We’re planning to launch the certificate program in the next six months and the master’s program within the year. The goal is high scale, low cost. If we can show employers that this is a more valuable signal than someone who graduated from an Ivy League school with a 4.0, then I think we’re onto something.

Practice, Feedback, and Self-Sovereign Data

Nate: One of the things we’ve wrestled with is that assessing these more durable skills requires a lot of practice reps, like an athlete or performer, and a lot of opportunity to collect that data in context, in conversations, in the hallway, with a teacher. But that gets into surveillance territory, where you’re constantly being measured and assessed. Do you think the future of durable-skills assessment looks like practice, practice, practice, simulations, and then a curated portfolio of your best work? Or are we heading toward an AI companion plugged into your ear all the time, ranking you every night?

Sal Khan: It’s hard to know what the human race as a whole will create, but I’ll tell you what I believe we should create and are trying to. Even before AI, we had a platform called Schoolhouse.World, another nonprofit we set up during the pandemic. A lot of people were using Khan Academy, but we wondered, wouldn’t it be great if people could learn from each other in a very human way? Zoom donated licenses to us, and we ran an experiment: what if volunteers tutored other people?

First question: how do you certify someone knows calculus or physics well enough to tutor it? We figured out a way. On Khan Academy you can take an assessment, but there’s no authentication that it’s really you. So what if you record your face and screen and explain your reasoning out loud while you take it? If you hit a 90% threshold, which has to be high because we want you to be a tutor, you submit the video, and it gets peer-reviewed by several people you don’t know, who check you followed the protocol, didn’t cheat, and can explain your reasoning. If it checks out, you’re certified in, say, calculus, and you can start tutoring.

Once you begin, you can tutor an AI simulation first and get feedback. Then you do your first few real tutoring sessions with a more senior tutor observing. Once you’re on your own, the students you tutor rate you, and there’s an AI giving you private feedback, not your public rating, on things like, “You spent a lot of time talking to Nate but not to Susan,” or, “You gave a lot of explanations, maybe ask more questions instead.”

As soon as we launched this, about fifty top universities, University of Chicago, Yale, MIT, Caltech, asked if they could use it for admissions. They were interested in who could get certified, but even more interested in who could tutor. If you can tutor calculus and have a high rating doing it, that’s more powerful than just saying you got a five on the AP calculus exam. Now they know you can communicate, you have patience, empathy, strong listening skills. Anecdotally, they’re admitting students who can prove this at a much higher rate than the general applicant pool.

They also told us, “Is there a way to give students practice and feedback on having constructive dialogue about things they disagree on?” So we tried something: what if high school kids filled out surveys on the toughest topics, gun control, abortion, Israel-Palestine, affirmative action, and we paired kids with opposite viewpoints over Zoom for a structured conversation? Most people said that would turn into a disaster, screaming matches and threats. We ran the pilot, told everyone, “This isn’t a debate, you don’t need to convince the other person, you need to listen, say your point of view respectfully, and then give each other feedback afterward.” It would bring tears to your eyes in a good way, people with very different viewpoints actually listening to each other. People are craving that right now.

And to your question about whether everything gets exposed on your profile, the answer is no. If you tell me, “Sal, you spoke all over me, you were obnoxious, you made me feel bad about myself,” it’s good that I hear that feedback, most people who do that never get told. But I can choose whether to share that with a university or employer. I just know I need to improve on those dimensions. Hopefully, over the next ten conversations, I get better, and eventually I start getting feedback like, “Sal, you’re a good listener,” and I can selectively display that.

For the Khan-TED Institute, across all the skills, whether it’s a portfolio of AI projects or feedback from these structured conversations, we put you into groups from around the world to solve authentic work tasks over video conference, and you get peer review. Then you elect what you want on your profile. A graduate’s transcript might look like: here are the projects Nate did, he understands vibe coding, he knows how to validate information from an AI, he’s strong on communication, and here are the five projects he wants to showcase, plus notes like, “He has a spike in curiosity and working under pressure, colleagues said he’s the guy you want when the rules change at the last second.” We hand that to employers, and they’re telling us, “Yeah, this is what we want.”

Nate: Right, it’s real-time, and there’s a clear set of skills you’aiming for, lots of practice, feedback-rich environments, and then a curated showcase, what we sometimes call self-sovereignty, you have control over what’s out in the world and the signals you’re sending. It reminds me a bit of what ETS was doing at one point with the Mastery Transcript Consortium, though that relationship has since changed.

So, on the Khan-TED Institute, is there an opportunity for people to bring in outside experience, “I want to add this because it already demonstrates this,” or is everything contained within the projects you create?

Sal Khan: Yeah, absolutely, it’s possible, especially once we’re in the bachelor’s program. We do want people to show they know their statistics or have some base knowledge in general-ed courses, history, writing, whether through a college class, an AP exam, or something in their work experience. But for the main meat of the programs, certificate, bachelor’s, and master’s, we want to put people into authentic work tasks, the kind people actually need to do right now. “The four of you have eight hours, start a business,” and that doesn’t just mean a business plan, it means a prototype, a pitch deck, a video presentation, things that used to take a team of ten people six months.

Unless someone can showcase something they’re doing at work, and the problem with work is, if you work somewhere like Apple, everything’s confidential, you can’t bring a camera into meetings. A lot of people also aren’t in jobs where they even get the opportunity to do these kinds of tasks. A family member of mine was recently asked, as part of a job application, to design something that would automate a hospital process, using AI, in a week. She spent a lot of time on it, showed it to them, and never heard back, no idea if they lost interest or just stopped hiring. If she’d had that project well documented, she might have been able to submit it as one of these Institute projects.

We’ll have individual and group projects, leaning more heavily on group projects, because one stereotype of online high-skill education is that it’s impersonal and asynchronous. But if ninety percent of the time you’re connected with other people, working with them, talking to them, rating them, getting feedback, that’s actually richer than most in-person programs, where the real connection happens in the dining hall or the dorm, and the classroom itself is three hundred people taking notes. I remember in college I knew maybe five people in a three-hundred-person class, just because they lived in my dorm hall, the rest were just faces I saw every day and never interacted with.

So I’m hoping that when you graduate, you not only have a portfolio, but you’ve interacted with hundreds of interesting people around the world, a network that’s unheard of in online education, and honestly unheard of outside a handful of elite universities that make a real effort to build that. I went to Harvard Business School, where every interaction is case-based, and the most valuable thing there is that you’re constantly interacting with other people. If you ask me about any of the eighty or ninety people in my section, I have a strong opinion about them, and I’ve hired some of them, or wanted to work with them, based on that, ten or twenty years later. I hope we can build that inside the Institute.

Nate: I appreciate that, and the idea that human and social capital make a difference, there are real equity issues there, and there’s an opportunity to build social capital through specifically designed programs like this. Kudos to you all for thinking about it. Most of higher ed is still lecture-based, and while I wouldn’t say most of K-12 is, it’s still the fallback, talk to students rather than have them talk and do discourse with each other.

What AI Learning Actually Looks Like: Khanmigo’s Evolution

Nate: So let’s pivot to learning in general, learning and AI. You started Khan Academy a long time ago, it’s had massive success, and it’s very much about content. There are changes coming, gamification, more transparency in results, tutoring built directly into practice with Khanmigo. What’s your thinking about AI and learning, given what you’ve learned from Khan Academy? Is it going to be the ultimate tutor, a partnership with humans, in support of them? What’s it going to look like?

Sal Khan: It’s going to be some of all of the above. A lot of the conversation in AI circles, even in that job discussion we were having, is about diffusion rate, even if an AI is capable of something, how fast do people actually adopt it. I think that’s true in education too. Someone who’s hyper-motivated and curious today can already use AI as an unbelievable tutor.

I’ve been geeking out with various AIs about the Iran conflict, the Strait of Hormuz, the scenarios that might play out, how the Strategic Petroleum Reserve works, pressures in caverns, all of it. I think I now know more about this than if I’d gone and done a master’s or even a PhD, because I had the interest, and it just fed me information that led to more and more interesting questions. I understand so much more now about military strategy and supply chains, like the fact that a land invasion needs hospital vessels, I’d never even thought about that. If you’re curious about something today, you’re unstoppable.

I’d argue this was true even five or ten years ago for someone hyper-curious, Khan Academy has been around almost two decades, MOOCs for a decade and a half, YouTube has a lot of trash on it, but also a lot of amazing things, and someone who can avoid most of the trash can learn a ton. Now you add AI, and you can ask unlimited questions and get deep answers.

The problem, and this is what we learned with the first version of Khanmigo, which we launched in 2023, the same day as GPT-4, OpenAI gave us about a six-month preview before ChatGPT even came out, so we’ve been working on this longer than almost anyone. The first version was an AI aware of the work you were doing on Khan Academy, there to support you if you had questions. For some students that was useful, if they got a little stuck and the teacher was busy, they could ask about the video or the exercise. But it was like me walking into a classroom and saying, “Hey kids, I’m Sal, I think I’m a pretty good tutor, I’ll sit in the back, come ask me if you have questions.” Some will, but a lot will be embarrassed, or won’t know how to articulate their questions.

What we learned is that younger kids, four or five years old, do ask questions freely. Something happens along the way, some of it is socialization, and socialization isn’t always positive. There’s a socialization in school where you’re supposed to be quiet, and it’s not always coming from adults, sometimes it’s from peers. I remember in school, outside that gifted and talented program, if I got excited about something, I learned not to bring it up in front of my peers, “Look at that nerd, he’s geeking out.” You learn not to ask questions, not to stick your head out. Kids actually forget how to do that over time, and they didn’t know how to structure their questions, didn’t have the metacognition to know what they didn’t know.

So in the new version of Khanmigo, which we just launched in classrooms about a month ago, we’re making it more proactive and central to the learning process, not just inserting AI for the sake of it. Our goal has always been personalized mastery learning, and how you help a teacher do that. In the past, you’d get your first question wrong on Khan Academy, maybe look at the solution, and move on. Now, at least in the classroom version, as soon as you get a question wrong, the AI shows up and says, “Hey, you got that wrong, why don’t you review it with me?” In a couple of months, it’ll be, “If we can work through this together, I’ll give you credit for that question,” or, “If you got it right and can really explain how you did it, maybe I won’t make you do four more, it’ll save you time.” That creates a strong incentive to engage more deeply, something a teacher with thirty-five kids can’t do for every single student after every question.

I’m also hopeful about giving teachers more than just, “This is what you assigned, here’s who did it, here’s the percentage correct, here’s how long it took.” Now we can say, “Nate got a lot of questions wrong, but when I discussed it with him, he seemed to really understand it, he just kept making careless errors,” or, “Sal got a lot of questions right, but I think he just memorized a formula, or he did them so fast he might be cheating.” That extra context is what a good human tutor would give a teacher after sitting next to a kid, and I think it’s really valuable.

We’re also launching interim assessments on Khan Academy, work we’ve been developing for three or four years, going to market this coming fall. This was funded by Reed Hastings, founder of Netflix, few people know he used to be a public school teacher, and he really believes assessments are broken. He told us Khan Academy would be in a great position to build assessments that are transparent, accessible, and actionable. These are psychometrically valid assessments, but on some questions, after you answer, it asks, “Can you explain your reasoning? What method did you use? Why didn’t you choose another approach?” So when a teacher, student, or parent gets the report back, they don’t just see “Nate’s in the 74th percentile, growth was nine points,” they also see, “Nate really understood the concept but kept making careless errors,” or “He leaned heavily on this approach but felt uncomfortable with that one.” That’s much more actionable, narrative-based information.

Nate: I’d assume that’s exactly what you’re going for, when an educator gets a set of narratives back alongside growth and proficiency scores, which most are already well-versed in reading, they can rethink what their classroom rotation looks like, when they’re working one-on-one versus in groups. It becomes a much more data-rich environment than just a set of numbers.

It feels like the tutoring piece is already true in a lot of ways, my water heater broke, and I could figure out how to fix it.

Sal Khan: Yes, or my big hack, take a picture of your water heater, describe it, and hand it to Gemini or Claude, it’s amazing.

Nate: Right, and they’ll identify the part. But that all requires motivation. That’s sort of the holy grail in education, and it sounds like that’s what you’re trying to hit with the Khan Academy updates, how do you increase motivation, increase nudges, increase transparency.

Cognitive Offloading and the Case for Effort

Nate: As we wrap up, I have two big questions left. The first: what you just described doesn’t really allow for cognitive offloading, because there’s accountability built in, on the system, in your tutor program. My pessimistic view of a world filled with AI is similar to my pessimistic view of social media, the algorithms are designed to tap into evolutionary shortcuts, if something can be made easier, we’ll take the easier path and use it. It feels like we might be headed that way with AI too, things become easier and easier, and we just cognitively offload more and more. Does that change us as a species, and what do we have to do in education to push back against that? Am I too pessimistic, or is there an optimistic view here?

Sal Khan: No, you’re not too pessimistic, and there are patterns of this in the past. The Industrial Revolution amplified human physicality, you don’t have to walk, you can take a car or a bus, tools amplify your strength. Until about the seventies, people didn’t realize our bodies were atrophying. Our grandparents looked like they were in pretty good shape, they were out in the fields, farming. Now we sit all day, drive everywhere, barely walk, eat processed food, and we’re dying because of it. That’s when gym culture showed up, and it’s only accelerated over the last four or five decades.

You’re already starting to see the mental equivalent, and people are aware of it. Ironically, my biggest pet peeve at the gym right now, this happened yesterday, is people sitting on a machine, just doom-scrolling on their phone, hogging it, not realizing they could easily spend half an hour just sitting between sets because they’re hypnotized by the phone. I actually want to make a spoof of the Odyssey scene where Odysseus ties himself to the mast to resist the sirens, except in my version, all his oarsmen have earplugs in, except one guy who’s just got his phone, completely unbothered, doom-scrolling TikTok or Instagram while everyone else struggles against the sirens’ call.

That’s a real problem. I have a friend who uses GPS every single day to get to work, the same place every day, and their sense of direction has completely atrophied. That’s cognitive offloading. So yes, it’s a real problem. Kids are cheating, people are asking AI to make decisions for them in their actual lives. We need to protect ourselves from that.

I’m somewhat addicted myself, most of my screen time is the crossword puzzle, which I can almost convince myself is healthy, but I don’t feel great about it even then. So I force myself to read for half an hour to an hour every day, real books, I’m reading War and Peace right now, my kids would say I’m “aura farming,” and yeah, I am. I go to the gym, I turn fifty in October, I’m doing deadlifts and drinking protein shakes to make sure I don’t atrophy that way either.

I think we have to be thoughtful about this, the same way we don’t let our kids skip PE and just sit around physically, hopefully we can fend off the mental equivalent, but as adults we have to model that ourselves. This technology can absolutely lead to a lot of cognitive offloading, and we have to be scared of that, but there’s also a lot of cognitive onloading it makes possible. Like your water heater, in the past you’d just call someone and there goes $400. Now you think, “Maybe I can actually solve this myself with a little help.”

I’ve fixed a toilet a couple times in the last year that I didn’t think I was capable of fixing, the AI reassured me, helped me find the part, I ordered it on Amazon, it showed up that afternoon, and I felt like Superman. Even my wife looked at me a little differently. So it can go both ways, and I think we have to take responsibility for ourselves to make sure we go in the good direction.

Nate: I agree, it feels like we have a superpower we can harness, and that harness can send us one way or the other. In education specifically, we need to make sure every learner understands how to use that superpower well, and understands the challenges, the same way we’re trying to do with social media. I think every middle schooler and high schooler should have a neuroscience class, specifically about the neuroscience of addictive technologies, because it’s so influential in the world. The same should be true for AI, every student should have a real-world, relevant AI literacy class in middle and high school as they’re entering this world.

Advice for High Schoolers

Nate: Sal, we’ve been going for a while, let’s finish with one last question. I’ve got one kid who just graduated college and another halfway through. What advice would you give to a high schooler, someone in 11th or 12th grade right now? I think from other podcasts I’ve heard, you might be in a similar spot with your own kids. What should they be thinking about, in terms of learning, in terms of AI?

Sal Khan: My oldest is a senior in high school, so we’re right in that zone. This might sound a little trite, but: be excellent at what you’re doing. If you’re going into software engineering, and my oldest is kind of headed that way, he’s been writing video games since he was six or seven, been really into math from a young age, also really into piano, be excellent at it. Going back to our earlier conversation, if you’re going to be a software engineer, be one of the top ten percent, don’t just be someone churning out code, that’s what Claude is for.

Regardless, you’re going to have to build these durable skills. I give my son credit, when I was his age, I’d like to think it’s because I have better delivery than my mother did, but when I tell him things like, the other day he had a bit of an interview coming up, and I said, “Let me see your handshake.” I told him, “No, a little harder, okay, no that’s too hard, okay, perfect.” And, “This is how you make eye contact, have a good smile at the appropriate times,” those small things about how you engage with people really, really matter. As a parent, I’ve been paying closer attention and trying to gently give feedback when I see my kids at a party, how they interact with adults.

That’s going to matter a lot. A lot of it is out of our control as parents, so you have to learn to accept that, but you can gently give nudges, usually positive feedback, “I liked how you showed up and talked with the adults and held your own.”

My daughter is fifteen, she’s a waitress at a local Indian restaurant, she’d been wanting a job forever. The more she does it, the more I notice, I actually spied on her once when I picked her up from the restaurant, I went ten minutes early, and I was like, “Wow, that’s Dia?” She was like an adult, joking with people at a table, and I thought, those skills are really valuable.

At Khan-TED Institute, one of our employment partners, a management consulting firm, told us the biggest correlate between new college grads and who eventually makes partner is people who worked in hospitality or retail in high school. My daughter is charming, at home she’s got the standard teenager “whatever mom, whatever dad,” but when she’s waitressing, she’s really good at what she does. As long as she keeps up with her academics, and we have a mastery-based school at Khan Lab School, so if she gets the equivalent of a B every now and then, which she can always improve since it’s mastery-based, I’m honestly less worried about that than about the fact that she’s developing these very important durable skills.

Not just the interpersonal skills, she learned to hustle. She applied for a ton of jobs, got rejected, and kept going. That’s the kind of thing you want to see in your child. If you do that, whatever degree you end up with, I think people will respect that you haven’t cognitively offloaded, that you’re capable of doing hard things, willing to hustle, have good communication skills, and are entrepreneurial. I think you’ll do great.

Nate: I appreciate that advice, and I’d add one more, maybe you already said it in a way, have a superpower of curiosity.

Sal Khan: Oh, one hundred percent.

Closing Takeaways

Nate: Sal, awesome conversation. A few things I’m taking away. One, on the future of work, there will be pockets of pain, and making sure young people have durable skills and curiosity to move through and adapt around that is going to be really important. There’s the question of how people work in groups, going back to the social skills you learn in the service industry or on group projects, which is part of what you’re building into the Khan-TED Institute. I really like this idea of skills, practice, and feedback, and how AI can insert itself into that loop, whether that’s through Schoolhouse.World’s AI tutor training or Khan Academy in general, giving people a lot of practice and a lot of feedback before they go share that with others.

The importance of social capital is going to keep growing as more devices and technology consume our time. And then this idea that motivation is everything, motivation comes from being curious, from feeling safe and like you belong somewhere, from seeing purpose. And your pieces of advice: be excellent, engage with other humans, think entrepreneurially, solve problems when you see a challenge, and be curious.

Sal, really appreciate all the great work you’ve done for the world of education and the world at large, and I really appreciated this conversation. Thanks for being here today.

Sal Khan: Great, thanks for having me, Nate.

Guest Links

Sal’s LinkedIn: https://www.linkedin.com/in/khanacademy/

Sal’s X (Twitter): https://x.com/salkhanacademy

Khan Academy: https://www.khanacademy.org/

Schoolhouse.World: https://schoolhouse.world/

Bio

Sal Khan is the founder and CEO of Khan Academy, a nonprofit with the mission of providing a free, world-class education for anyone, anywhere. He is also the founder of Schoolhouse.world, Khan Lab School and Khan World School, all nonprofits focused on making world class, personalized mastery education accessible.

Sal’s interest in education began while he was an undergraduate at MIT. He developed math software for children with ADHD and tutored fourth- and seventh-grade public school students in Boston. He holds three degrees from MIT and an MBA from Harvard Business School