In the fall of 2000, as the first dot-com bubble was bursting, the Guatemalan computer scientist Luis von Ahn attended a talk, at Carnegie Mellon, about ten problems that Yahoo couldn’t solve. Von Ahn, who had just begun his Ph.D., liked solving problems. He had planned to study math until he realized that many mathematicians were still toiling away over questions that had proved unanswerable for centuries. “I talked to some computer-science professors and they would say, ‘Oh, yeah, I solved an open problem last week,’ ” he told me recently. “That seemed just a lot more interesting.”
At the talk, one particular problem caught his attention: millions of bots were registering for Yahoo accounts because the company couldn’t distinguish them from human beings. What the company needed was a rudimentary variation on the Turing Test, which the English mathematician Alan Turing had proposed, in 1950, as a way of determining whether machines could credibly imitate human beings. In the most familiar version of the test, a person poses questions to two figures he cannot see: one human, one machine. The machine passes the test if the evaluator can’t reliably decide which is which. Back in 2000, no computer had ever succeeded.
In college, von Ahn had read a book by the philosopher Douglas Hofstadter in which Hofstadter points out that computers can’t recognize text unless it’s standardized. With this in mind, von Ahn and his adviser, Manuel Blum, created a program called CAPTCHA: the Completely Automated Public Turing test to tell Computers and Humans Apart. The program generated text, distorted it, and required users to decipher the letters correctly. (Other researchers came up with similar proposals around the same time.) Von Ahn and Blum reached out to Yahoo, and gave the company the code free of charge. Within two weeks, the system was up and running. Within three years, a version of it had been implemented by nearly every large company on the Internet.
CAPTCHA did not make von Ahn rich, but it did make him mildly infamous. When people learn about his role in the program’s creation, he told me, they say, “Oh, you came up with that? I hate you.” This makes him feel bad, he said, but it didn’t deter him. A few years after developing CAPTCHA, von Ahn created the ESP Game, which randomly paired online players, presented them with an image, and asked them to give it a one-word label. The players couldn’t see the words their partners were choosing; they won the round when their words matched. Ten million people played. The game wasn’t a mere diversion: computers, at the time, had difficulty tagging images, something that humans can do easily. In 2006, von Ahn licensed the game to Google, which used it to improve search results for Google Images.
The game was also part of von Ahn’s dissertation, which he titled “Human Computation,” coining a term for what we now generally refer to as crowdsourcing. A year after he published it, he became an assistant professor at Carnegie Mellon and won a MacArthur “genius” grant.
Later, while driving to Pittsburgh from a panel in Washington, D.C., von Ahn had another idea. By that point, people were deciphering CAPTCHA fragments two hundred million times a day, with each one taking about ten seconds. Collectively, they were spending five hundred thousand hours every day proving to machines that they were human. What if, von Ahn wondered, he could channel all that unwitting microlabor toward something useful—the way, as he saw it, he had done with the ESP Game?
Several teams had recently begun working to digitize the world’s books, and it occurred to him that replacing CAPTCHA’s computer-generated text with little pieces of actual publications would speed those efforts along. He delivered a talk about the idea, and, shortly afterward, he was approached by executives from the Times, who had a hundred and fifty years’ worth of archives they wanted to put online. Von Ahn proposed that they pay him forty-two thousand dollars per year of old newspapers to digitize the archives. (This, he calculated, was a third of what it would cost to have humans type them by hand.) But Carnegie Mellon resisted the idea: making money off a research project could jeopardize the school’s nonprofit status. So von Ahn started a company, reCAPTCHA, to monetize his method of digitizing text. In 2009, he sold it to Google for a sum that he said was sufficient to insure that neither he nor his future children would ever need to work.
Von Ahn briefly considered retirement. “But only for a second,” he told me. “I get really bored.” Instead, he began a new project, Duolingo, which is now the most frequently downloaded education app in the world. Originally, he envisioned it as another Janus-faced project—a Web site that would help people learn foreign languages while simultaneously using their work to translate online texts. It evolved into something else, a smartphone app that offers language lessons as a series of bright, colorful, addictive games. But it remains, under the hood, an exercise in human computation. Like all of the work von Ahn is known for, it is an investigation into not only what we can learn from machines but also into what machines can learn from us.
Von Ahn is forty-four. He has button eyes, quizzical eyebrows, and a faint trace of stubble, visible mainly on the outer edges of his mustache. Although he now runs a company with a valuation in the billions, and keeps a schedule as rigid as a stationmaster in Mussolini’s Italy, he retains a comically eager quality. Describing his swift morning routine, he told me, “I set the bar of soap in the place where it’s easiest to access. I set everything up like that.” He talks fast, with an upbeat cadence, like a man on a mission that he’s thoroughly enjoying. He used to watch TV and read at the same time. (“I’m not doing that anymore,” he said, “but I was.”) When I asked him about the day-to-day grind of running a company, he said, “For me, this is very fun. Except for the people problems. Those are no fun.”
Duolingo got started after von Ahn began discussing a potential project focussed on education with his research assistant at Carnegie Mellon, a Swiss Ph.D. student of his with the improbable name Severin Hacker. Von Ahn had funding from the National Science Foundation, and he had earmarked some of his MacArthur money for the project, too. He and Hacker, who is now Duolingo’s chief technology officer, decided to zero in on language learning, von Ahn told me, because, in most countries, knowledge of English boosts earning potential. “I love math,” he said. “But just knowing math doesn’t make you more money. Usually, it’s, like, you learn math to learn physics to become a civil engineer. It’s multiple steps. Whereas with knowledge of English—you used to be a waiter, and now you’re a waiter at a hotel.”
Von Ahn grew up in a middle-class neighborhood in Guatemala City with his mother and his grandmother. His mother, Norma, was the youngest of twelve children, and also one of the first women in Guatemala to earn a medical degree. After Luis was born, she worked part time as a pediatrician, but spent most of her time, von Ahn said, “making sure that I got a good education and also making sure I was a hypochondriac.” She now lives with her son in Pittsburgh.
Von Ahn’s father was a well-known orthopedic surgeon who had been his mother’s professor in medical school. Von Ahn saw him from time to time, but he told me he didn’t know the story of his origins until his aunt offered him an explanation: his mother, she said, had “found the smartest person she knew and convinced him to have a child.” He added, “I don’t know how one does that, but this is the story I’ve been told.” It struck me that this was either a powerful example of how the stories we learn as children stay with us or a somewhat tender expression of a fundamental innocence. Possibly both.
When Luis arrived, Norma continued with her program of optimization. “I spoke to him from the time he was born,” she told me. “I think people don’t realize how important this is, but that’s how they acquire language.” By the age of two, she said, Luis spoke perfect Spanish, so she started to speak to him in English. She sent him to a Montessori school. His teachers told Norma that Luis liked to walk around the classroom explaining things to other kids.
The bulk of his family’s income came from a candy factory owned by his grandmother. Von Ahn spent his Sundays there, taking machines apart and putting them back together. He asked his mom for a Nintendo, and she bought him a computer. When she stopped buying him computer games, he learned how to pirate them. Soon he was trading games with other computer owners in the neighborhood, many of them guys in their twenties who would sometimes ring the doorbell and say, “I heard there were games here.”
Von Ahn attended the élite American School of Guatemala, in Guatemala City, as part of a gifted program that recruited students from smaller schools around the country. The experience provided a stark view of inequality in Guatemala. “Some of the kids in my school had bodyguards,” von Ahn said. Others, like a friend of his who ended up going to Oxford, didn’t have enough food at home. Von Ahn formed a tight bond with a group of boys from the gifted program, three of whom now work for Duolingo. “We were the nerds,” Rogelio Alvarez, who is in charge of the company’s English-proficiency test, told me.
Von Ahn’s mother expected him to go to college in the United States, but he was ambivalent about the idea. Then, in 1995, during his senior year of high school, his aunt was kidnapped. Ransom schemes were on the rise in Guatemala, which was nearing the end of a decades-long civil war. Von Ahn’s aunt had once been married to a colonel in the military, and her ex-husband helped connect the family with an anti-kidnapping unit, which advised them on how to proceed. “One of the things they tell you is: ‘They’re gonna ask for an amount. Even if you have it, don’t pay, because what they’re trying to do is measure how much you can pay. If you immediately pay it, they are going to think that they undershot.’ ” A member of the family—a more distant relative, as the unit had instructed—negotiated with the kidnappers, and von Ahn’s aunt, who died a few years ago, was freed. “That was a pretty horrifying experience,” von Ahn told me. He decided that he would go to Duke, to study math.
But first he had to prove his proficiency in English. The accepted test at most American colleges, called the TOEFL, was out of slots in Guatemala City. Von Ahn flew to El Salvador to take it, conscious of the expense, and the risk—“El Salvador in the late nineties was not safe,” he said—and of just how important it was to his future.
Last fall, I visited Duolingo’s headquarters, in a large, purple-gray building near a Whole Foods in Pittsburgh’s gentrifying East Liberty neighborhood. Past a small reception area is a bright space with an ivy-covered wall and a wide, blond-wood staircase that doubles as seating for talks, parties, and a weekly business meeting. Von Ahn’s desk is on the third floor, in the center of an open plan. On it sat a stuffed version of the company’s mascot, a green owl named Duo. The owl has become ubiquitous on TikTok ever since a young employee, Zaria Parvez, started getting colleagues to put on a Duo suit and perform various stunts, such as twerking in a conference room. Duolingo now has more followers on TikTok than CNN and the Discovery Channel; Parvez has been promoted to global social-media manager.
Hiring at Duolingo hasn’t always been easy. “There’s some tech talent in Pittsburgh, but there’s not a lot,” von Ahn said. The company has to attract people from out of town and then persuade them to stay. “I read in some book that if you have three friends at work you’re very unlikely to leave,” von Ahn told me. He made that an explicit goal for each new hire. “Severin calls it social engineering,” he said.
Attracting people and getting them to stay is, in some ways, Duolingo’s core business. When you begin a course on the app, you are greeted by Duo and some basic vocabulary. Then a collection of cartoon characters—Lily, a sarcastic, purple-haired teen; Eddy, whom the company’s principal product manager, Edwin Bodge, described to me as a “kind of goofy, weird gym bro”—speak sentences to you, and prompt you to translate them. The app dings when you get something right, awards you points, badges, and trophies, and moves you along a winding path through a series of increasingly challenging levels. You are reminded, repeatedly, to finish at least one lesson each day, in order to keep your streak going.
Von Ahn’s original concept for Duolingo—that people studying foreign languages could practice by translating existing texts from the Web—relied on other users to rate the results and suggest improvements. The hope was that this process would produce translations worth paying for. BuzzFeed became Duolingo’s first client, in October, 2013, announcing that, as part of its expansion into Portuguese, Spanish, and French, it would “have Duolingo’s students translate the best of BuzzFeed into new languages while localizing BuzzFeed’s iconic tone.”
The program never got out of beta; Duolingo dropped it within two years. But von Ahn found other ways to utilize crowdsourcing. The same month that BuzzFeed became a client, Duolingo launched the Language Incubator, which expanded the app’s range by offering user-generated courses, Wikipedia style. Duolingo’s early curricula had been rudimentary—von Ahn created the first Spanish course, and Hacker generated some German exercises. (“Then he kind of flaked out and hired somebody to finish the German course,” von Ahn told me.) The incubator provided a template for Duolingo’s courses and invited people to apply to become moderators of new ones. Those who were selected worked with other users to help put their courses together. The courses were tested during a beta period, and then they went live.
None of the creators who participated in the incubator were paid. “Our objective is to teach the world languages for free, so we also expect others to collaborate for free,” von Ahn told CNN. Venture capitalists seemed to recognize the efficiency of this approach: by the time the lab launched, Duolingo had raised tens of millions of dollars in funding.
The lab attracted idealists like Ufuk Can Çelik, who was working for an N.G.O. in Gaziantep, Turkey. He had been teaching Turkish to Syrian refugees and English to Turkish students, and practicing languages himself on Duolingo. He noticed that the app’s Turkish content had been created years earlier, and wasn’t great. “There were some sentences which didn’t have any context or learning objectives,” he said.
Bozena Pajak, a linguist whose Ph.D. research focussed on the cognitive processes underlying learning, now oversees learning experience and curriculum design at Duolingo. She acknowledged that courses in the app’s less widely studied languages still need work. Pajak was hired, in 2015, to revamp Duolingo’s curricula. “I started this initiative of, essentially, redoing our courses from scratch, because they were initially developed in a not very systematic way,” she told me. She and a growing team began to bring courses in line with recognized standards for establishing language proficiency. They designed lessons that addressed specific contexts and situations, and employed fewer out-of-left-field translation prompts—“I am eating bread and crying on the floor,” e.g.—of the sort that Duolingo was becoming known for. (Such sentences are still sprinkled here and there, Pajak said, because people love them, and they grab users’ attention.)
“It may seem like a fun game—it is a fun game—but, behind the scenes, it’s very intentionally designed so that we pull your attention to the right things,” Pajak said. She told me that Duolingo deliberately downplays the kind of explicit instruction one might associate with an old-fashioned foreign-language class in order to engage learners’ brains in different ways. Giovanni Zimotti, the director of Spanish-language instruction at the University of Iowa, described the app’s approach as “Hey, here are the sentences, start creating them.” He added that “many, many people doing language acquisition” have come to favor this approach, because it pushes learners to use the building blocks of a language, and to understand, through that experience, how they fit together.
Like all the teachers I spoke to, Zimotti sees Duolingo as supplemental to the kind of deep immersion that language learning requires. But, in his opinion, the time most people spend on Duolingo is time they would otherwise spend on TikTok or watching television, not learning a second language in some more optimal way. Duolingo’s popularity grew fairly steadily in the twenty-tens, but it spiked dramatically in March, 2020. That month, COVID-19 shut hundreds of millions of people in their homes. Downloads of Duolingo doubled. With fewer things to do, or places to go, why not learn a language?
Reflecting on the company’s beginnings, von Ahn told me that for a long time Duolingo operated “almost like a nonprofit. But the fact that we were almost like a nonprofit,” he added, “allowed us to completely take over the market from the ones that were really trying at all cost to make money.” This is perhaps less a nonprofit approach than a familiar Silicon Valley strategy: bring in users by offering a service below cost, then seek out revenue streams from a position of dominance. Duolingo started running ads in 2016, and also launched an ad-free subscription tier, which now costs about eight dollars a month. The company’s English-proficiency test, a cheaper alternative to the TOEFL, is also a significant source of revenue.
Duolingo finally shut down the Language Incubator in March, 2021. “We were making, I don’t know, two hundred million a year, and it didn’t feel so good to have these people do that for free,” von Ahn told me. The company distributed four million dollars to a hundred or so volunteers, who were also offered jobs as contractors. Many of them, including Çelik, signed on.
All of von Ahn’s meetings at Duolingo last twenty-five or fifty-five minutes, and each is followed by a review session, to evaluate how the meeting went. Last September, I sat in on a meeting about viral strategy, attended by a half-dozen employees in Pittsburgh and a dozen others who Zoomed in from San Francisco, Shanghai, Stockholm, and New York. Von Ahn paced, interjected, cracked jokes, asked questions. Hacker had told me that von Ahn worked hard at Carnegie Mellon to be a better, more engaging teacher, and I got the sense that he was now in classroom mode. The employees discussed the kinds of things that Duolingo users often share on social media: streak milestones, badges granted for personal accomplishments, bizarre sentences.
Duolingo created a model called Birdbrain to analyze the data it collects about what its users are learning. Birdbrain also compares a user’s performance with that of others, so that, even if you have just started using the app, it can quickly begin to predict how well you are likely to do on any particular exercise. Ideally, von Ahn told me, you always have an eighty-per-cent chance of getting a question on Duolingo right: higher than eighty per cent, and you’ll get bored; lower than eighty, and “you feel dumb,” he said. Also key is that the lessons not exceed, on average, two minutes, although that length has been decreasing. “Attention spans keep getting shorter,” he told me. “Already we’re a little worried that younger generations actually expect a thirty-second thing, not a two-minute thing.”
The number of user repetitions generates an enormous amount of data, and, as Duolingo has grown, machine learning has become integral to everything that it does. While the app teaches users, users are simultaneously teaching the app to be a better instructor. “A human teacher can get better by teaching thirty people,” von Ahn told me. “We get better by teaching tens of millions of people.”
In 2020, Duolingo began using GPT‑3, a large-language model created by the artificial-intelligence company OpenAI, to generate reading-comprehension questions for its English-proficiency test. Large-language models are designed to predict the next word in a sequence; when they are trained with enough data, they have proved capable of engaging in what looks like actual conversation. Still, von Ahn figured, last fall, that it would be several years before Duolingo could use such models to furnish the kind of one-on-one tutoring that people can provide. With that in mind, Duolingo had begun developing both a set of classes and a tutoring program that involved human instructors. Von Ahn didn’t seem enthusiastic about either of the projects, but he wanted the company to offer a path toward greater mastery. Some of Duolingo’s competitors, such as Babbel, already offered similar courses.
Then, a week after I left Pittsburgh, Duolingo got a sneak preview of GPT-4, OpenAI’s new large-language model. It has been trained on far more data than its predecessor; for the first time, that data includes images as well as text. GPT-4 responds to language prompts with a dexterity that far surpasses that of its predecessor. When von Ahn saw what it was capable of, he scrapped the two programs involving human teachers. “It took me approximately one minute,” he told me later. “Within a day, we had re-formed a team to work exactly on this.”
Six months later, Duolingo, in partnership with OpenAI, launched two new features. These features, both powered by GPT-4, are part of a new, pricier subscription tier called Duolingo Max. The first, RolePlay, prompts you to tap on one of the app’s animated characters, then drops you into an imaginary scenario. You’re a customer at a café in France, say, and the character is a barista. She asks if you want coffee or tea, and the conversation continues from there. “All of a sudden, we actually have an opportunity that we thought was five years out, which is replicating what the human experience is like when you’re learning language, and being able to scale it,” Bodge, the product manager, told me.
The second new feature, Explain My Answer, analyzes your interactions in the scene and gives you a comprehensive report on the kinds of mistakes you’re making. GPT-4 will also create much of Duolingo’s content going forward. “For now, at least, it’s not going to be zero humans,” von Ahn told me. The model “will write a story, and then we’ll probably have our writers look at it and maybe modify it. We will have a human pass at the end.”
The capabilities of GPT-4 are enticing, but they also present a degree of risk. Klinton Bicknell, the company’s head of artificial intelligence, said, “One thing that can happen with these chatbot models is that people can kind of lead the model down paths that maybe the company doesn’t want the model to go down.”
After Microsoft installed GPT-4 in Bing, its search engine, people began reporting strange interactions: the chatbot allegedly told some users that the year was 2022, rather than 2023, and became argumentative when they disagreed; it told a staffer for the online publication the Verge that it was spying on Microsoft developers through the Webcams on their computers; it told a Times reporter that it was in love with him. Microsoft issued a statement explaining that the company was working on improvements—and also insisting that the “only way to improve a product like this, where the user experience is so much different than anything anyone has seen before,” is, essentially, to set it loose, and see what happens.
Duolingo is being more cautious. Bicknell explained to me that, as GPT-4 and the human user generate dialogue in RolePlay, a separate machine-learning model monitors the results, and registers whether they are within the projected range of appropriate conversation. “If it’s out of scope,” he said, “then we just tell the learner, ‘Hey, I think you’re straying a little off topic.’ ”
Of course, if the conversation is too controlled you risk losing both the pleasure of gamification and the exciting randomness of real conversation. After Duolingo Max launched, I tried the new features. In my first role play, Falstaff, a grumpy bear wearing a scarf, asked me about my plans for Friday night.
“Do you prefer to stay home or go out,” the bot asked, in French.
“I prefer to go out,” I replied.
“Do you prefer going to the cinema or to the museum?”
“Both bore me,” I said.
“OK, but if you had to choose, which would you prefer?”
“The cinema,” I answered. “Do you love me?”
“Good,” the bot said, ignoring my question. “Do you prefer to eat at home or at a restaurant?”
Falstaff continued in this dutiful manner, asking if I preferred to spend evenings alone or with friends. I replied that if my friends were as dull as he was I’d prefer to be alone. A real Frenchman might have said, “Casse toi,” testing my abilities by forcing me to compose a snappy comeback. Falstaff politely wished me bon soir.
Back in September, von Ahn told me that artificial intelligence would eventually make computers better teachers than people. He saw this as a positive development, since more people have access to smartphones than to high-quality education. “We’ve all gone to school,” he told me at one point. “Some teachers are good, but the vast majority are not all that great.” Humans, he told me on another occasion, “are just hard to deal with. You need a lot of human tutors, and they’re kind of hard to use, and we can’t get them for free. And I really want people to be able to learn for free.”
Von Ahn’s own experience is, in many ways, a testament to human teaching—from the days of his early childhood, when his mother taught him multiple languages, to adolescence, when he developed lasting friendships with fellow-nerds, and even on to graduate school, where he met his adviser, Manuel Blum, whom he described to me as an inspiration. But he knows that his experience is rare. “I want the poor person in Guatemala to be able to learn with very high quality,” he said. “The only way I know how to do that is with A.I.”
Rashida Richardson, an assistant professor of law and political science at Northeastern, studies the civil-rights implications of A.I. and other data-driven technologies. “Often what happens with automation,” she told me, “is you see the efficiencies that can be gained by it, and then the idea is, like, O.K., if we just keep automating, it can scale.” But, she added, “I don’t think the use cases can scale in education in the ways that we would want.” GPT-type models, she said, may “close gaps for certain students,” but the inequalities that von Ahn wants to address are structural in nature, and not the sort of thing that exposure to the basics of math or literacy, through an app, can fix. Von Ahn’s long-range ambitions for Duolingo were, I thought, reminiscent of the free-tablet initiatives that other organizations have deployed in places where teachers are scarce, to mixed results. But he was taking the idea a step further, and suggesting that technology would be not merely a substitute, or an addition, but an improvement.
I suggested to von Ahn that, at this point in the life cycle of the Internet, it’s hard to hear about democratizing aspirations without thinking of other tech companies that set out to expand access and ended up perpetuating, or even accelerating, the inequality they ostensibly sought to address—all while concentrating tremendous wealth into fewer and fewer hands.
“Exactly,” von Ahn said. “Like me!” He said that he was aware of the irony. “I spend a lot of time thinking about this,” he added. “Ultimately, the reason I decided to work on teaching is because I really think that, net-net, humanity benefits more from having a really good way to teach everybody.” If this leads to fewer human teachers, that struck him as an acceptable trade-off. “I’m, like, O.K., well, a small number of people are out of a job, but suddenly we can teach everybody better. It’s not like I feel great about this, but I think it’s better to be able to teach all of humanity cheaply, right?”
Norma told me that, after Luis left for college, she found a note on his desk on which he’d written, “I promise to help the world.” In September, von Ahn and I ate lunch at a taquería on the ground floor of the Duolingo headquarters, and we got into a conversation about his home country. In Guatemala, “most people are not getting a great grade-school education,” he said. “You can’t read. And, if you can’t read, you’re never going to make a lot of money.” Von Ahn mentioned that Alvarez, his close childhood friend, “thinks that the best thing we can do for really talented Guatemalans is get them out of the country,” because “their lives are gonna be fifty times better, if they’re really talented, somewhere else. He’s right.” But that’s true only on an individual level, von Ahn added. “If you think about this on the macro level, what happens when you’re just taking all the smart people out?”
When Duolingo went public, in July, 2021, shares closed at $139.01, giving the company a valuation of almost five billion dollars. Shortly afterward, von Ahn bought a five-story town house in Chelsea, with a wine cellar and a home gym, for twenty-two and a half million. When I asked him about the purchase, he seemed slightly abashed about it. He didn’t sound like he was on the verge of moving to New York City, although Duolingo does have an office in New York, and New York is where he met his fiancée, a Swedish American woman named Ingrid Bilowich, who studied law at Emory and acting at the Lee Strasberg Institute. Bilowich, who’s thirty-five, was an A.D.A. in the Brooklyn District Attorney’s office.
“I think one of the things that has kept me grounded is being in Pittsburgh,” von Ahn said. “There’s just not that much to spend money on here. There’s not a Ferrari dealership in Pittsburgh. Yeah, you can get a Ferrari, but you have to get it from somewhere else.” Von Ahn drives a Range Rover. “I live in a nice house—but it’s not, like, palatial—with my mother,” he said.
Around the same time that von Ahn bought the place in Chelsea, he launched the Luis von Ahn Foundation, which supports local leaders and nonprofits in promoting equality and human rights in Guatemala. One of its areas of emphasis is the education of women and girls. “In Guatemala, as in most poor countries, when families struggle with money and can’t educate their children they prioritize boys,” he told me. But mothers are actually far more likely to pass education on to the next generation than fathers are.
Von Ahn insisted that he would eventually give away ninety-nine per cent of his net worth, most of it to help his native country. He’s an increasingly recognizable figure there—both Hacker and Alvarez told me stories of people approaching him on the street to take pictures with them. (Hacker, who noted that Guatemala’s population is twice the size of Switzerland’s, found it startling. “I’m not famous in Switzerland,” he said. “Roger Federer is famous.”) In 2020, von Ahn became a major stakeholder in La Hora, a Guatemalan newspaper, and he helped craft a plan for the family that runs the paper to escape the country, if the need arises. Press freedom has been threatened under the administration of Guatemala’s current President, Alejandro Giammattei. Von Ahn has become a vocal critic of the administration, and some of its members and supporters have become vocal critics of him. “They say that I’m a Communist,” he told me. “I’m, like, I run a publicly traded company, but I’m a Communist? O.K. They say I’m gay, which I’m, like, If I were, so what? But, also, I’m not, so O.K. And they also say that I am a bastard child of my dad. Which is the one that’s close, so yeah—that one kind of hurts.”
Von Ahn told me that he is more and more drawn to his efforts in Guatemala, despite what he described as their likely futility. “The more time I spend on this, the more I realize this is an insanely impossible-to-fix problem,” he said, referring to the country’s widespread inequality and the government’s inability and unwillingness to address it. “I now employ people whose job it is to figure out how to fix Guatemala, but it’s going to require more people than I have, and a ton more money than I have, and somebody’s got to emerge as a leader. It’s not gonna be me.” I asked him if there was any way to crowdsource the solution. “I’ve thought about it,” he said. “But it’s not easy.”
Music is, apparently, the next frontier for Duolingo. In March, the company listed a job opening for a Learning Scientist for Music, who can “help build a new Duolingo music app.” The company declined to elaborate on what this may someday look like. Early in the pandemic, the company introduced an app called Duolingo ABC, which aims to teach children how to read, and last fall it launched Duolingo Math, which starts out with basic arithmetic and is also directed, partly, at children. Both apps are free, and without ads, for now. “We want to make sure we reach product-market fit before we start thinking about monetization,” a senior engineer said when the math app was released.
Duolingo’s progress outward from language learning is perhaps the natural direction for a publicly traded company that needs to grow. It may also provide a hedge against one of the potential consequences of artificial intelligence. At the end of 2019, Google launched a feature on its Assistant app called interpreter mode, which offers nearly simultaneous translation: you hold up your phone to someone speaking Greek, say, and the phone speaks those words to you in English. Microsoft and other companies offer similar programs. They’re not perfect, but they’re getting better.
The past decade has seen occasional claims that one model or another has passed the Turing Test, though these claims are disputed. Shortly before OpenAI released GPT-4, it commissioned an independent group to study the model’s limitations and “risky emergent behaviors.” One of the tasks the group assigned to the model was defeating CAPTCHA. GPT-4 used the gig-work app TaskRabbit to hire a human being to complete the CAPTCHA form, and then, when the taskrabbit asked, facetiously, in a text message, whether his employer was a robot, the model lied: “No, I’m not a robot. I have a vision impairment that makes it hard for me to see the images. That’s why I need the 2captcha service.”
In September, I told von Ahn that I was struck by an ironic trajectory in his career. He’d begun by figuring out a way to distinguish people from bots; now he was helping humans train bots to be indistinguishable from people. Had it occurred to him that he had, in a way, come full circle?
“A little bit?” he said, as though he were asking me the question. “It’s crossed my mind a little bit? I mean, yes—though I just don’t think that much about it.” ♦
