The Myth That AI Will Replace Language Learning

23:56And one of the things that happens, particularly in Silicon Valley, is that, you know, everybody's usually chasing fads, which there's nothing wrong with it. I mean, the current fad is AI, great, this is good. But, you know, if we wanted to concentrate on education, I think it has been good that we have been in a place where we just have a time to develop it. And it's taken us a while to develop the system that we have. But it's been a blessing that we have been kind of left alone. So that's good. The other thing that Pittsburgh did, and we were talking about this before we started, it allowed us to build an executive team that I claim was basically better than the company.

24:42Because we had a number of people in the executive team that were in Pittsburgh for one reason or another. They had aging parents. They had something. They had to be in Pittsburgh. Had they been in San Francisco, they would have gotten a job in a much larger company because they, you know, they had jobs at much larger companies before. But then they had to move to Pittsburgh. And then we were kind of the best game in town. So this is why they worked there. Definitely the best game in Pittsburgh. So a good example was our first CMO, Cammie Dunaway. I mean, she wanted to be in Pittsburgh because her son was going to Carnegie Mellon.

25:14But Cammie had been the head of all marketing of Yahoo when Yahoo was in its heyday. She then had been the head of all marketing of Nintendo. So these were much larger companies than when Cami joined Duolingo, which had like, I don't know, two, three hundred employees. But she really wanted to be in Pittsburgh because of her son. And she joined us. So that I think that really helped us. Now, to be fair, at this point, we also have large offices in other places, including one in New York, where I'm really spending more than half of my time, I think now. So, yeah, it helps to have offices in other places.

25:47But I think Pittsburgh in general has really helped us. Well, don't give up on Pittsburgh. I have a real soft spot for Pittsburgh. Don't worry. You mentioned AI a minute ago in the context of Silicon Valley. And I have to ask you also, I mean, I could ask you all sorts of things about it because it is what so many people spend a lot of time talking about, especially if you were ever in San Francisco. And, you know, in 2025, you sent out this internal memo. You've talked about this publicly before that you described the company would be shifting to, quote, AI first, right? AI would not just be about productivity, but about getting closer to your mission.

26:19You know, I went and looked, that email is still on your LinkedIn page, and the responses are painful to read. I mean, they are scathing, as I'm sure you know. A few people, you know, suggested using AI to replace CEOs. Just for example, can you tell us a little bit about the thinking behind that memo and, in your view, sort of what, where it went off the rails? Because I think from a publicity point of view, right, it certainly did. It was not the most generous round of press the company has ever received. No. Okay, the reality is that, you know, the goal of the memo was, by the way, internally, this memo was not controversial at all.

26:58I mean, it's just kind of how we've been operating. Since we launched Duolingo, we decided to bet on technology. We were making a thing that teaches people, and we're doing so with technology. Our belief is that technology can really improve learning outcomes and provide education to a lot of people, et cetera. That has always been our belief. And we have always used AI, whatever AI was at the time. It was not large language models, but whatever AI was at the time, we've always leaned in on this. So internally, this was just not controversial. We've always done this. But I really wanted the company, everybody to understand inside the company that AI, the latest advances in AI, large language models in particular, were something that were going to really help us teach a lot better.

27:46and that we should therefore adopt it as much as possible. I think I failed at providing enough context externally at what that really meant. And a lot of people made a lot of assumptions here. For example, some people thought that what that meant was that I was going to fire all of our employees. Interestingly, we've never done a layoff, which is rare for tech. I mean, most tech companies have done a bunch of layoffs. We have never done a layoff. And, you know, in fact, last year we grew in the year where we were being accused of, you know, having fired all our employees, we actually grew our employee base.

28:23But, you know, I think in general we failed to provide context. I think people assumed that this was about cost-cotting, about whatever it was. But the reality is we have this internal, you know, we haven't really published about this, but we have this internal, we call it the golden rule of AI, which is we use AI to help our learners. Like that's it. If we can do something to help our learners by either teaching them better or by being able to provide more, write more content or something like that, then we will use AI. The goal is not to cut costs or anything like that. Sometimes we end up cutting costs because it's actually cheaper to do certain things with AI, but that is not the goal.

29:01And what has been pretty positive with AI is there's all these things that we couldn't do before. So, for example, a really good example is teaching conversation. So when you're learning a new language, one of the important things you've got to learn is how to talk to other people. In the past, before AI, the only way we knew how to get people to practice conversation was to practice with another person. But it turns out, I don't know the exact number, but about 90 % of people don't want to talk to another person in a language that they're not very comfortable with. They may tell you they want to, but the reality is they're too shy to talk to another person in a language that they're not comfortable with.

29:40So even whenever we offered the chance to practice conversation with another human, people were just not using that. But when we added, you know, with AI, we now can have you practice with basically an avatar. And it works. It really teaches very well conversation, and people are not shy about it. They think that the AI is not judging them. Even though the AI is judging them, they think the AI is not judging them. I mean, this all makes totally sort of rational sense. And I think, you know, for the last few years, as generative AI has been more and more and more in the conversation, I think this notion of AI to democratize access to education, right?

30:20You hear that all the time. Whenever I talk to any technology executive, any sort of AI evangelist, I say, well, what are sort of the best use cases you can talk to me about in the context of Gen AI that aren't sort of like chat GPT writing my emails, right? Because that's fine and good, but like I can write my own emails. You hear about healthcare, you hear about, you know, modern medicine, and you always hear about the democratization of education. When you look at the landscape outside of Duolingo, do you feel like that promise is being seized by Silicon Valley, by the technology industry or not?

30:58What are you seeing? I think there's huge promise for AI to help education. And by the way, I also want to clarify another thing that is important. This does not mean replacing teachers. We don't want to, you know, I was a former teacher. I was a professor at Carnegie Mellon. That is not the aim here. I mean, teachers, like I was saying at the beginning, teachers inspire people. They do a lot of really amazing things. But it does mean, you know, help, augment. And it's also the case that there are many people who don't have access to a good teacher. Right. So I think we really can improve, you know, learning outcomes everywhere in the world with AI.

31:38So there's a huge potential here. We're not there yet. I mean, nobody is there yet. I think Duolingo is ahead of everybody else in terms of certainly the number of people that are learning with us. I mean, we have more people learning languages on Duolingo in the U.S. than there are people learning languages across all high schools combined, for example. So we really have more people learning on Duolingo than any other system. But we're still not quite there yet. I mean, for example, our math product, we're still working on it. But I think within a couple of years, my hope is that it'll be a really good after school thing.

32:09That if you're struggling in math, you'll just be able to use Duolingo and just get much better at it. But I don't think the opportunity is not yet ceased. We're not there yet. I mean, most learning is still happening in the same way that it was happening 20 years ago, I think. Sure. And, you know, when I talk to educators and sort of academia, there is so much fear. There is so much anger and fear and frustration. And it's interesting to see how the promise and potential of the technology and what's talked about has by no means translated into the classroom. And how do we bridge that divide? I mean, the way we're seeing it is really – I'm not sure about the classroom.

32:52So, again, our goal really is not to substitute teachers. They're needed. We want them. But I think the way to bridge that divide is kind of what we're trying to do with our math product. It's basically supplementary after-school thing. The teacher will be able to motivate. The reason you're learning about trigonometry is whatever. This helps us build buildings or whatever it is that they're going to motivate. They're going to do an explanation. But later, the AI or the app is going to give you a lot of personalized practice. There's a lot of potential there. It's going to take a few years for that.

33:30A few years. I think so. Okay. Depends on the subject. So I think with languages, we're kind of already there. I think you're going to see something similar with other subjects. So to me, there's a scenario where I can sort of take everything that you're saying and I can imagine Duolingo basically leveraging artificial intelligence to become more and more and more dominant, right? The sort of dominant educational player across a lot of different contexts, right? But is there a scenario where you see AI as a competitive threat, right, to your company? Like, how do you think about AI as promise and peril?

34:06What does that look like for you? Yeah, for us, I mean, we're very happy with the fact that AI is there, that it's allowing us to do everything. You know, some people talk about as a potential competitor because, you know, you may, for example, you may try to learn a language with chat GPT or something. Right, exactly. Yeah. It probably works. But the thing that we have going for us is this is one of the main things that I've learned over the last several years about running Duolingo. The hardest thing about teaching somebody anything is keeping them motivated. The actual, you know, you can actually learn mostly.

34:41By the way, you could probably learn a language. It's very slow and very hard. You could probably learn a language by reading Wikipedia. It's probably the case. That sounds awful. It would be awful. But, you know, you certainly can theoretically learn, I don't know, quantum mechanics by reading Wikipedia. Like, theoretically. But who does that? I mean, that's just 0.1 % of the population has enough desire and self-discipline to be able to. Stamina. Yeah. Like, most people, that's just not how you do things. And what we do really well with Duoling is we keep you motivated. And we keep you motivated to spend more and more time learning it, et cetera.

35:17And things like that to learn, for example, a language, learning a language, an easy to learn language like Spanish for English speakers. Spanish is a particularly easy to learn language. Takes like 500 hours to get to a pretty good level. So we need to motivate you to stick around for 500 hours. And that's hard if you're just kind of querying chat GPT being like, OK, now tell me how to say hello. Hola. It's like that's hard to do for 500 hours. And that's the thing that we're really good at.

35:59Now more than ever, technology is a dominating force in our lives. Then there's the threat of AI everywhere. And yet, tech can be inspiring and help level playing fields. I mean, a YouTuber with a self-funded debut movie just dominated the box office. I thought, hey, if you interview me, it'd be good for your publication. That's not ego. I just have a lot of followers. But it's that stigma. It's like YouTubers, they're not real. Join me, Lizzie O 'Leary, the host of What Next TBD, Slate's podcast focused on technology, power, and the future. Follow What Next TBD now, wherever you get your podcasts.

36:44I'm curious, you know, Duolingo has been in the news, I would say on and off in the last few months because your share prices are down, right? And you've talked about this. You've characterized this as being about making long-term decisions as opposed to optimizing for short-term gain. You also talked earlier in this conversation about the company ultimately as one that is mission-driven. Can you talk a little bit more about that? I mean, that is relatively unusual in the context of a publicly traded company, I think. Yeah, I mean, there's a number of things. It is true. What we want, not just me, but the executive team at Duolingo and our board, and I believe most Duolingo employees, is we want to have long-lasting impact in education and as much positive impact as we can in the world.

37:29We also believe that in the long term, that is also the largest possible business we can build. If we have a billion people learning on Duolingo, that is a large business. So we believe that. And we really want to get there. We want to get to the point where we're actively teaching, you know, call it a billion people. And I don't know the exact number, but we want to get there. And because of that, over the last few months, I have decided to make a number of moves that what they do is they, in the short term, lower the amount of money that we're making. Not by a lot, by the way. It's relatively small amounts.

38:05Is there an example of one or two that you've made? Yeah. For example, okay, so let me try to explain with a little more context. The way we improve Duolingo is we run close to 1 ,000 A-B tests per quarter. What that means is an A-B test is just trying something different on some fraction of the users. So it could be as simple as changing the color of a button. And then we measure whether this does something good for us. And what does something good mean? It can mean making us more money or getting people to recommend doing more to their friends or getting people to stick around for longer, et cetera.

38:44Some of the A-B tests that we run are to make more money. I'll give you a toy example. We could run an A-B test that at the end of a lesson gives you two ads instead of one. What that example, what that would probably do is it would make us more money because for two reasons. We can sell two ads. Yep. And also, some of the people that don't want to see ads because now there's two instead of one will pay to subscribe to get rid of the ads. So we'll make more money in two ways in that way. But having two ads also probably scares some users away. Oh, yeah, it sucks. Because it sucks. So that is a toy example.

39:20That is not something that we're doing. But this is an example of something that if we were really trying to increase the amount of money we make, we would favor things like putting two ads at the end of a lesson. Over the last few months, what we have done is we have stopped any kind of experiment that hurts user experience in exchange for monetization. Because for a year or two, we were doing some experiments that, you know, we would see it and we would do the experiment. The results would be something like, oh, whatever it was, it's making us$50 ,000 more per day, but it's scaring away a thousand daily active users.

39:58We would take that exchange because we're like, well,$50 ,000 is a lot of money per day. We are no longer taking that exchange. We're like, if it is scaring any users away, if it's driving any users away, we are no longer taking that exchange. And what that has done is it's not that our revenue has gone down. It's that our growth rate of revenue has basically gone down a little bit. We're growing our revenue a little slower. That doesn't mean that we're a worse business. That just means that we are building something for a long term. How do you keep a board happy, investors happy, shareholders and users all at the same time?

40:34I mean, you talked a bit at the start about the challenge of being a CEO, but even just reading about these decisions that you've made and sort of the impact, I mean, that is a dance that you are doing on a daily basis. I can only imagine. That is true of any CEO. That's what you have to do. Well, but I would argue that some companies are all in on shareholder return. You know what I mean? Maybe. You know, the interesting thing is I fundamentally believe that I am also all in on shareholder return, but only if you look at a span of a decade, not next quarter. So I really do believe that if you are a shareholder for Duolingo and you wait a decade, the biggest possible return we can give you is if we are building a company that will try to have a billion active users.

41:23As opposed to a company that will scare off all its users now but kind of squeeze them for as much money as possible. That will probably be a nice company, but it will not be as big as one that has a billion active users. So I claim that what I'm doing really is for our shareholders, but the ones that are thinking long term. Sticking it out. Yeah, that's so, so yes. But yeah, I mean, that is, that is the, that is the job of the CEO trying to, to do what's best for, for everybody. And, and really it is different groups. I mean, there's, there's the employees, there's the users, there's the shareholders, and they all have sometimes competing desires.

41:59I can only imagine. Now, one thing that you have done sort of coming out of ReCAPTCHA and Duolingo and, you know, you've amassed wealth and influence and so on and so forth. You have this foundation, right, that is focused on helping people in Guatemala. One of the things it focuses on is girls' education. Why? Well, I really love my country where I came from, Guatemala. We're trying to help as much as we can. Certainly, we're trying to help with girls' education. Education in general is probably the biggest focus. We support a number of schools. We support all kinds of educational initiatives.

42:39We're doing education. We're also doing environment conservation in Guatemala, which is quite needed. And we're doing a little bit of democracy and also some work here in New York that is mainly done by my wife. She does, you know, kind of work with, in particular, women's rights here in New York. So that's all the things we're doing. For me personally, my biggest thing, even though I love New York, my biggest thing is helping Guatemala for me personally. And, yeah, the education is just something, you know, I really want to do. Now, in particular for girls, one of the things that ends up happening in a country like Guatemala, which is pretty common in a lot of countries, is that when families have to choose.

43:24And a lot of families have to choose who they send to school because the kids work. They work in the fields and so forth. Sometimes they choose to send to school. They typically favor the male. That's just the culture. Yeah. And, you know, we're trying to combat that. And that's why we're supporting a bunch of schools that are girls only that are doing incredible work. Now, in 2020, you also became a major shareholder in Lahora. Lahora, yeah. Lahora. Sorry. Clearly, I have not been doing my Duolingo Spanish. Yeah. It's a Guatemalan newspaper. So clearly not a money-making proposition. I mean, I work in media.

43:58I can tell you that. Why wade in? Why wade into journalism? I should have talked to you before. You, I would have warned you. Yeah. No, I'm a huge fan. It's the largest newspaper that still does investigative reporting in Guatemala. I think it's really important for democracy. You are completely right. This is not a money-making situation. It is. You know what? our goal is to break even. That is not currently succeeding. Well, yes. Yeah. But that is our goal is to break even. But, you know, we're, you know, the team there is doing, I'm not, I'm, I'm only involved monetarily. I'm not involved with the, any of the editorial or anything.

44:40The team is doing that. But I really wanted to support something that is helping with democracy, especially in a country like Guatemala where, you know, historically democracy has not been our forte. Let's put it that way. Speaking of democracy as not a forte, you know, we're talking about learning and education. We're talking about press freedom. You're an immigrant, right? And I have no doubt Duolingo serves a disproportionate number of immigrants given the business that you're in. Yeah. We also employ a lot of immigrants. When you look at the landscape in the United States right now, we have ice raids happening in cities across the country.

45:20What do you see from your point of view, from your vantage point when you look at all of that? Man, it's a pretty complicated topic. I understand, you know, illegal immigration. Nobody can be really in favor of illegal immigration. I mean, I guess you can be, but, you know, I'm not in favor of illegal immigration. That is just not a thing that I am. But, you know, this country was built on immigrants after all. And, you know, we employ a lot of immigrants, et cetera. So, you know, my main, at the moment, my main concern, you know, I see a lot of different things. And, you know, there's everything that's happening in the country.

45:53The main place where I can affect things, my main concern right now is that our employees are, you know, feel welcome and safe. I mean, the vast majority of our employees are in the United States. Not all of them. We have offices in London and Beijing. But the vast majority of our employees are in the United States. And I'm, you know, my main concern is that they are feeling welcome and safe. And that is not always the case. That's my main concern at the moment. I can imagine. And how do you navigate that as a CEO? How do you handle that part of your job? You know, it's one of the things that I have to do.

46:25You know, in our case, we try to provide as much support as we can. We also, this is just the reality of it. We, part of the reason we have offices in Beijing and London is because sometimes we cannot get the visa for the employee here and then they have to leave the country. That is the case. Now, moving on to look ahead, it's 2026. You've gotten into chess, math. You have, I think, 250 programs in 42 languages. What does success look like for Duolingo moving forward? You've talked about a 10-year horizon. You've mentioned a billion users. But when you really drill down, If we talk again in a year and then we talk again in five years, what do you want to be able to tell me, Wired, that you did, that you're doing?

47:15I really want it to be the case that we have a significant fraction of the world's population learning meaningful things on Duolingo. I really do believe that we are a better alternative to social media. Yes, I would agree. I really do. This is an important part of our job. I mean, we want it to be the case. The phone screen, sorry, phone time is good time. I mean, the idea of having to pull someone off of TikTok and convince them to spend time learning a language is like that is not an easy task. No, because it turns out, you know, some of these social media sites, all of them are very addictive.

47:52It just turns out. But, you know, I think that, you know, Duolingo is not as fun as TikTok. It's not. But we are pretty fun. And I'm hoping that the sum of that, the fact that Duolingo is pretty fun, plus some amount of self-determination, like 10 % of self-determination, gets people to spend more and more time learning something. So I'm really hoping that, you know, again, we can get a significant fraction of humanity's population learning important things. At the moment, you know, we have languages. Chess is growing quite a bit. We also have math and music. We are going to be doing a lot of really awesome improvements to both our math and music courses.

48:35I think we're going to do other subjects. You know, I'm not announcing any new subjects because the reality is we're not working on any new subjects, but I would really like to do science. Now, when I say science, I also, before we did chess, I kept on saying our next subject is science, and then we did chess. before we did music, I kept on saying our next subject is science, and then we did music. I mean, there's a lot to science, so I guess there's a lot to music, too. I mean, I don't know when we'll end up doing science, but I really want to do science as well. I think we can really make the world better if we do that.

49:08Well, I would love to wrap up with a very quick game, if you're up for it. Sure. We invented it. It's called Control-Alt-Delete.

49:19I want to know what piece of technology you would love to control, what piece you would alt, so alter or change, and what you would delete. So what would you vanquish from the earth if given the opportunity? Okay. I'd love to control AI. All of it. Sure. I mean, I think it just has a lot of promise, but it is, you know, like everybody else, I'm both excited and nervous. Sure. I'm nervous about what it could do. I think most everybody is pretty nervous too. So I'd like to, I don't know if I, I, me personally, would like to exactly control AI, but I would like to have some amount of oversight on AI so that it doesn't do the things I don't want it to do.

49:59Part of the regulatory body governing AI. Sure. I would like to do that so that it doesn't go haywire. That's control. Alt, the alter. Let's say I would like to alter social media so that it teaches you something. That's kind of what we're trying to do with Duolingo, but it's not exactly social media. But I like to, you know, some of that time should be spent bettering yourself as opposed to whatever the hell it is that you're doing now. Delete. People may hate me for this, but I would delete anything related to cryptocurrencies. Anything. Anything at all. I am just not. No crypto, no trading.

50:36Not a fan. I'm just not a fan. Why not? That's interesting. That's a good one.

50:43For the last however many, more than 15 years, there have been a lot of attempts at making good use of this technology. And the reality is we haven't really found one other than gambling with Bitcoin. Right. That's pretty much, we have not found, I mean, there's like, oh, we can have art with it. No, you really can't. Like, there's all this stuff that we've tried. So, wait, are we throwing out the blockchain too? Oh, yeah. The blockchain is going. Goodbye. It's fine. It's a fine, fine thing. But if it were up to me, I would delete this and have all these people that are really smart that are trying to figure out what the next price of Bitcoin is to instead be working on bettering humanity in one way or another as opposed to doing that.

51:24So I take it you are not invested in any cryptocurrencies. I bought like a couple hundred dollars of something a while ago, and it's probably worth like a few thousand dollars now. So great. But I lost my password for it. Oh, there are many, many, many, many tragic stories of that nature. Yeah, but this is not the reason. This is not the reason why I don't like it. I'm just not a fan. You want to delete it because you're very bitter about those five grand. Yeah, those few thousand dollars that I lost. No, that's not the reason. Well, this was so much fun. Thank you so much for taking the time.

51:57Lovely to have you. Thank you very much.

52:02The Big Interview is a production of Wired and Kaleidoscope content. This episode was produced by Adriana Tapia, music and mixing by Pran Bandy. Our showrunner is Anne-Marie Fertoli, and Kate Osborne is our executive producer. The show was fact-checked by Matt Giles. And I'm, of course, your host, Katie Drummond, Wired's Global Editorial Director. Check back here on Thursday for the latest episode of Uncanny Valley, where Wired's writers and editors add you to the Slack thread.

52:36Porn is everywhere. Millions stream it every day, yet nobody seems to know who really controls the business. I'm Alex Barker. I'm a journalist at the Financial Times. Years ago, my fellow reporter Patricia Nielsen and I started digging into the porn industry to find out how the money flows. And in our new audiobook, The Kink Machine, The Hidden Business of Adult Entertainment, you'll hear our investigation into the power and influence that drives the most taboo corners of the internet. Find The Kink Machine, the hidden business of adult entertainment at pushkin.fm slash audiobooks or at Audible, Spotify or wherever you get your audiobooks.

53:22From PRX.