Slack off on your daily Spanish lesson and a cellphone alert will auto-generate at the same time you used the app the day before. Duo appears on your screen as an emoji, nagging you to keep up with your drills.
Millions obey Duo’s commands. Created seven years ago by MacArthur “genius” computer geek Luis von Ahn, 40, Duolingo has hooked everyone from Bill Gates, Khloe Kardashian and Jack Dorsey to Syrian refugees in Turkey.
“The moment I felt proudest was when I realized, Wow, the richest man in the world is using the same system as the lowest people on the economic scale,” says Von Ahn. “That is for me really special and pretty big.”
Duolingo has lots of room to grow bigger still. More than 2 billion people worldwide are studying a foreign language, and increasingly they are doing it online. Digital language-learning generates $6 billion in revenue, and that number is projected to rise to $8.7 billion by 2025. But it’s a highly fragmented market with dozens of players scattered around the globe, and it’s begging for a dominant one.
Von Ahn certainly knows how to execute on a grand scale. When he was a 21-year-old grad student he gave the world CAPTCHA—better known to billions of internet users as those annoying squiggly letters that you have to transcribe to prove that you are not a bot. Later he sold two inventions to Google, netting more than $20 million.
He splurged on a Lamborghini and a Tesla Model S but otherwise leads a modest life, remaining in the six-bedroom house he bought with his ex-wife in Pittsburgh’s Point Breeze neighborhood, close to Duolingo’s offices. “I could have gone to Guatemala to live in a villa, but I didn’t want to,” he says. Instead, he has thrown himself into building the world’s most downloaded education app.
Duolingo already offers more languages than its competitors do, 36 at last count. They include little-spoken tongues like Hawaiian, Navajo and Gaelic and the fictional language High Valyrian from the HBO blockbuster Game of Thrones (1.2 million users are studying it). À la Wikipedia, Duolingo enlists volunteers to help create its more obscure courses.
Seven years after its launch, it has nearly 30 million active monthly users, according to its own figures. Venture capitalists have pumped in $108 million, ballooning Duolingo’s valuation to $700 million in 2017, $150 million more than the market capitalization of Rosetta Stone, its 27-year-old publicly traded rival.
Along with teaching more languages (Rosetta Stone offers 25), Duolingo attracts users because its basic ad-supported version is free, compared with the $120 a year Rosetta Stone charges its 500,000 subscribers. Another competitor, Berlin-based Babbel, says its revenue of more than $115 million comes from the $85-a-year subscription fee it charges its million-plus subscribers.
Only 1.75% of Duolingo’s users pay for its ad-free version ($84 a year), but because its base is so huge, revenue hit $36 million last year. Von Ahn says that will climb to $86 million in 2019 and $160 million in 2020 as more users sign on and pay for the premium app, which will attract them with new features. Employee head count will rise from 170 to 200 by year’s end.
Duolingo’s headquarters, in a converted furniture store in East Liberty, a gentrifying neighborhood not far from Google’s Pittsburgh office, is set to expand to a second floor. The company isn’t profitable, but Von Ahn says it will be cash-flow positive this year. He is planning an IPO by 2021.
He likes to say that the money he forfeits giving away the app is equivalent to the cost of his rivals’ bloated marketing budgets. He also boasts that users are less likely to quit Duolingo than they are to bail out of competing language apps. “Our retention is comparable to games,” he says.
The comparison is apt. Duolingo grabs users with gamification tricks like points, treasure chests and “streaks” for continuous use. The app’s three-minute lessons are designed with a simple interface. In a typical exercise, the sentence “I eat bread” appears above seven bubbled words. Drag the words onto a line, hit “check” and “You are correct” appears at the bottom of the screen with a satisfying chime. Fans tweet about funny sentences generated by the software, like “I am selling my mother-in-law for a euro.”
Duolingo has gotten bad press from writers who try the app and don’t learn much. But Von Ahn promises only to get users to a level between advanced beginner and early intermediate. “A significant portion of our users use it because it’s fun and it’s not a complete waste of time,” he says….Read more
In 2013, James Evans, a University of Chicago sociologist and computational scientist, launched a study to see if science forged a bridge across the political divide. Did conservatives and liberals at least agree on biology and physics and economics? Short answer: No. “We found more polarization than we expected,” Evans told me recently.
People were even more polarized over science than sports teams. At the outset, Evans said, “I was hoping to find that science was like a Switzerland. When we have problems, we can appeal to science as a neutral arbiter to produce a solution, or pathway to a solution. That wasn’t the case at all.”
Evans started his study on Amazon. You know the heading that says, “Customers who bought this item also bought”? Evans and his colleagues analyzed the top 100 items in this list for two “seed” books: Barack Obama’s Dreams from My Father and Mitt Romney’s No Apology. They repeated this process for each book in the top-100 list until they ran out of new titles.
“The resulting ‘snowball sample,’ ” Evans and company wrote in their 2017 Nature Human Behaviour paper, “contained virtually all books in the largest strongly connected component in Amazon’s directed co-purchase network,” or 1,303,504 unique titles.
After performing a co-purchase network analysis—the sort used to study co-citation and co-author networks—on this dataset, the scholars concluded that political ideology guided people to science books. With some curious results. Liberal readers preferred basic science (physics, astronomy, zoology), while conservatives went for applied and commercial science (criminology, medicine, geophysics).
“It seems like conservatives are happy to draw on science associated with economic growth—that’s what they want from science,” Evans said. “Science is more like Star Trek for liberals: traveling through worlds, searching for new meanings, searching for yourself.” Science turned out to be “a huge example of confirmation bias,” Evans said. “You expect something to be true, you want it to be true, you read books that affirm and confirm those truths.”
The thing most disturbing to me is the onslaught of claims about fake information and fake news.
Looking at the polarized results, Evans had an idea. What would happen if you put together a group of diverse people to produce information? What would the results look like? Evans knew just the place to conduct the experiment: Wikipedia. Evans and Misha Teplitskiy, a postdoctoral fellow at the Laboratory for Innovation Science at Harvard, and colleagues, studied 205,000 Wikipedia topics and their associated “talk pages,” where anybody can observe the debates and conversations that go on behind the scenes.
The scholars judged the quality of the articles on Wikipedia’s own assessments. “It’s based on internal quality criteria that is essentially: What do we want a good encyclopedia article to be? We want it to be readable, comprehensive, pitched at the right level, well-sourced, linked to other stuff,” Teplitskiy explained.
In their Nature Human Behaviourpaper, “The Wisdom of Polarized Crowds,” Evans and Teplitskiy concluded that polarization doesn’t poison the wells of information. On the contrary, they showed politically diverse editor teams on Wikipedia put out better entries—articles with higher accuracy or completeness—than uniformly liberal or conservative or moderate teams. It’s a surprising result and so I caught up with Evans and Teplitskiy to offer their interpretations.
What does Wikipedia tell us about diversity?
James Evans: People talk about the importance of diversity. It’s not diversity in general; it’s diversity in specific. If you have these different ideologies, it’s associated with different filters on the world, different intakes of information, and so when it comes to constructing reference knowledge on an encyclopedic web page that’s supposed to thoroughly characterize an area, you do a much better job because you have a lot more information that’s attended to by this ideologically diverse group.
Editors working on a social issues page said, “We have to admit that the position that was echoed at the end of the argument was much stronger and balanced.” Did they begrudgingly come to that? They did, and that’s the key. If they too easily updated their opinion, then they wouldn’t have been motivated to find counter-factual and counter-data arguments that fuel that conversation.
We found that more diversity is associated with longer conversations. If they were immediately willing to give up on these things, then it wouldn’t have produced the sustained competition that ended up generating the balance that they, themselves, came to appreciate.
Which pages on Wikipedia benefit most from political diversity?
Evans: Political pages. The second most are social issues pages, which have a substantial political content. Even science pages benefit because sciences resonate with different political ideologies. And there’s no question that the science articles that benefited the most were the science articles that are associated with political polarization.
I would be surprised if we didn’t see that across the science pages associated with the environment, which include climate change, but likely a lot of other things, including biodiversity. Those are kinds of science articles that benefited the most from political polarization because they’re the ones, unsurprisingly, for which diverse political perspectives end up offering really different filtered information.
Misha Teplitskiy: Psychologists and organizational scholars call this “task relevance.” It’s the idea that diversity in ideas should help only for tasks for which diversity is relevant. You expect ideological diversity to be most relevant for politics, less so for social issues, and less so for science. The surprising feature there is that it’s at all relevant to science, but generally we expect it to matter less and less the farther you get away from task relevance.
The evidence for our contribution to climate change is unimpeachable. So might a Wikipedia entry that encouraged diverse opinions not produce a high quality result? Would that run counter to your study, to the importance of diversity?
Evans: You’re saying there are some things where diversity can just generate noise. In general, one could imagine this, and there’s a wonderful book called Merchants of Doubt which explores precisely this issue, that companies in a number of areas increased the amount of apparent diversity in a decreasingly diverse consensus about, for example, smoking’s influence on lung cancer.
That’s certainly taking place in the world. But for some reason, and this is a tribute to some of the standards in the context of Wikipedia, people discipline each other and are effectively disciplined by higher-level editors. There’s also a whole host of different perspectives that people might take with respect to global warming.
Even though there might be general agreement that human activity is increasing greenhouse gases and higher temperatures, it could be one assumption has you thinking there are human solutions to human problems, and another one has you thinking of the importance of human stewardship over the earth. So different perspectives aren’t just generating artificial conflict in these contexts.
At the same time, our experience is for broad topics. There are few places where there’s enormous amounts of certainty in the sciences. My guess is in places where there is strong certainty, we’re not going to see a big effect from political diversity. Political diversity is not a magical substance. If the distribution of political perspectives aren’t correlated with useful information about the topic at hand, then you’re not going to see a benefit. You’re going to see noise. You might even see a detriment.
What do you think about fake news?
Evans: The thing most disturbing to me is the onslaught of claims about fake information and fake news. In some sense, all information is fake. All of it has a purpose, an angle. But the fact that now it’s just so easy to claim that it’s fake without any particular support for that claim, and it’s popular to do so, means it’s easier to discount alternative information than ever before.
Angles are useful. They motivate people to look in a certain place, to search out information that you probably wouldn’t have searched out if you weren’t motivated by the possession of a belief. Angles end up having a lot of value, unless you discount them all. It begins with Trump arguing that everything’s fake news and then people arguing that Trump’s producing fake news all the time.
There’s this cloud of fakery out there, and, of course, it’s exacerbated by the proliferation of bots and other things generating noise. I see that in mass media news in the same way I see it online. It’s a new level. It’s like we’ve just discovered that there’s bias in the system and so everything is biased, categorically, and we can agree or disagree with it at will.
I hope that we can begin to persuade people to really value the importance of bias, that bias is critical to how we view things, that there isn’t an unbiased position.
Why are the highest quality articles overseen or written by an ideologically diverse group of people?
Evans: More collective insight is generated when you draw people who have non-random and minimally overlapping sets of information or knowledge exposures and you put them in a forum that’s well-regulated by a set of norms, which can be appealed to and are, in fact, appealed to. I was really struck by the fact that people often experience this. When they experience balanced debates on these sites, they really described the process as painful and beleaguered but the outcome as satisfying.
Teplitskiy: Ideologically diverse teams end up debating more. These people are carrying different bits of knowledge. When they bring it together, they’re spending more effort to aggregate it into good content. Even aside from increased effort, we’re also finding that the kinds of debates they have are a bit more focused. They zero in on a smaller set of issues and really hash out those issues that are presumably most problematic.
They end up having more conflict and rely on policies more for regulating what we call their “task conflict,” or conflict that’s oriented around creating content, and they also have a lower relational conflict—they gang up on each other less and harass each other less on a personal level compared to more unbalanced teams. Those that are more balanced have a lower harassment prevalence.
What happens when editor teams are politically unbalanced, or overwhelmingly left- or right-wing?
Evans: When you have a single person going in and describing a set of pages as, “They look like Russian propaganda,” those people don’t recognize that they are entering a system of 30 or 40 people who have constructed this page in conversation, and they are coming in and really just trashing that characterization.
Almost invariably, they just got beaten up, labeled as trolls, sent out of the community on a rod with tar and feathers. The homogenous group has a sense of, “We’ve built the social contract and then this person is coming in from the outside.” We found empirically, in our study, that there was a lot more toxic language when you have these imbalances.
Teplitskiy: Our data is suggestive. You would prefer diverse teams that are in a moderate position. Shifting in either direction away from the middle or the moderate position as a team is negatively associated with quality.
Are there some key lessons that groups that produce or evaluate ideas can take from your Wikipedia study?
Teplitskiy: One lesson that our work raises is around branding or creating a culture and letting people know about it, and letting it be the mechanics of how you organize a platform. One interesting thing about Wikipedia is it’s got a very strong culture. If you want to play in the sandbox, you should be ready to back up your claims, cite your sources, cite sources that are reasonable, listen to others.
That clearly discourages some people from joining, people who are not willing to play by reasonable rules. They do more filtering up front on who can play, not in a heavy handed way, but more by signaling their culture strongly, and people who don’t like it don’t stick around.
Compared to the science-book study, the Wikipedia paper sure seems to hold out hope for consensus.
Evans: Yes, and I hope that we can begin to persuade people, with this kind of paper, to really value the importance of bias, that bias is critical to how we view things, that there isn’t an unbiased position. Only when we begin to demonstrate the value of bias can we battle the cloud that bias is bad. Everything’s biased, so we have to reach into our core values and use those to guide our way through this world. There’s a strong scientific value behind bias, so our hope is to begin a conversation about the value of polarized crowds.
What can scientists learn from your results?
Evans: My hope is that not just scientists, but people with opinions and political stakes in general, can seriously consider the fact that people who don’t share their political viewpoints have something valuable to say—and even if they don’t have something valuable to say about a particular political topic, that their different experience and perspective has likely given them access to other kinds of information that will be valuable and new to you.
That’s the key to unlocking the potential of polarization: to allow people to constructively contribute to knowledge projects and other projects together. If you know enough about Wikipedia to open up the talk page, which anybody can do but almost nobody does, you’ll see extensive discussions going on. You’ll see people carefully, painstakingly employing diverse perspectives that are perceived by experts as being systematically better. It just produces more robust knowledge because there’s less ideological filtering going on.
Aronson, Elliot (2010). Social Psychology. Upper Saddle River, NJ: Prentice Hall. pp. 273.
Myers, D.G.; H. Lamm (1975). “The polarizing effect of group discussion”. American Scientist. 63 (3): 297–303. Bibcode:1975AmSci..63..297M. PMID1147368.
Isenberg, D.J. (1986). “Group Polarization: A Critical Review and Meta-Analysis”. Journal of Personality and Social Psychology. 50 (6): 1141–1151. doi:10.1037/0022-3514.50.6.1141.
Bray, R. M.; A. M. Noble (1978). “Authoritarianism and decisions of mock juries: Evidence of jury bias and group polarization”. Journal of Personality and Social Psychology. 36 (12): 1424–1430. doi:10.1037/0022-3514.36.12.1424.
Lord, C. G.; Ross, L.; Lepper, M. R. (1979). “Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence”. Journal of Personality and Social Psychology. 37 (11): 2098–2109. doi:10.1037/0022-3514.37.11.2098.
Taber, Charles S.; Lodge, Milton (July 2006). “Motivated Skepticism in the Evaluation of Political Beliefs”. American Journal of Political Science. 50 (3): 755–769. doi:10.1111/j.1540-5907.2006.00214.x.
Brauer, Mark J.; Judd, Charles Mosley; Gliner, M D (1995). “The effects of repeated expressions on attitude polarization during group discussions”. Journal of Personality and Social Psychology. 68 (6): 1014–1029. doi:10.1037/0022-3514.68.6.1014. PMID7608855.
Moscovici, S.; W. Doise; R. Dulong (1972). “Studies in group decision: II. Differences of positions, differences of opinion, and group polarization”. European Journal of Social Psychology. 2 (4): 385–399. doi:10.1002/ejsp.2420020404.
Stoner, J.A. (1961). “A comparison of individual and group decision involving risk”. Unpublished Master’s Thesis, Massachusetts Institute of Technology.
Forsyth, D.R. (2010) Group Dynamics
Moscovici, S.; M. Zavalloni (1969). “The group as a polarizer of attitudes”. Journal of Personality and Social Psychology. 12 (2): 125–135. doi:10.1037/h0027568.
Van Swol, Lyn M. (2009). “Extreme members and group polarization”. Social Influence. 4 (3): 185–199. doi:10.1080/15534510802584368.
Vinokur, A.; Burnstein, E. (1974). “Effects of partially shared persuasive arguments on group induced shifts: A group problem-solving approach”. Journal of Personality and Social Psychology. 29 (3): 305–315. doi:10.1037/h0036010.
Hinsz, V.B.; Davis, J.H. (1984). “Persuasive Arguments Theory, Group Polarization, and Choice Shifts”. Personality and Social Psychology Bulletin. 10 (2): 260–268. doi:10.1177/0146167284102012. S2CID145085635.
Hogg, M.A.; Turner, J.C.; Davidson, B. (1990). “Polarized norms and social frames of reference: A test of the self-categorization theory of group polarization”. Basic and Applied Social Psychology. 11: 77–100. doi:10.1207/s15324834basp1101_6.
McGarty, Craig; John C. Turner, Michael A., Barbara David; et al. (March 1992). “Group polarization as conformity to the prototypical group member”. British Journal of Social Psychology. 31: 1–19. doi:10.1111/j.2044-8309.1992.tb00952.x.
Feilitzen, C. (2009). Influences of mediated violence: a brief research summary. International clearninghouse on children, youth and media. ISBN978-91-89471-81-8.
Bastug, Mehmet F.; Douai, Aziz; Akca, Davut (2020-07-02). “Exploring the “Demand Side” of Online Radicalization: Evidence from the Canadian Context”. Studies in Conflict & Terrorism. 43 (7): 616–637. doi:10.1080/1057610X.2018.1494409. ISSN1057-610X. S2CID115806907.
Sia, C. L; Tan, B; Wei, K. K. (2002). “Group Polarization and Computer-Mediated Communication: Effects of Communication Cues, Social Presence, and Anonymity”. Information Systems Research. 13: 70–90. doi:10.1287/isre.13.1.70.92.
Taylor, J.; MacDonald, J. (2002). “The effects of asynchronous computer-mediated group interaction of group processes”. Social Science Review. 20 (3): 260–274. doi:10.1177/089443930202000304. S2CID220160579.
Iyengar, Shanto; Westwood, Sean (2014). “Fear and Loathing Across Party Lines: New Evidence on Group Polarization”. American Journal of Political Science. 59 (3): 690–707. doi:10.1111/ajps.12152.
Walker, Thomas G.; Main, Eleanor C. (December 1973). “Choice shifts and extreme behavior: Judicial review in the federal courts”. The Journal of Social Psychology. 2. 91 (2): 215–221. doi:10.1080/00224545.1973.9923044.
Kunovich, Robert; Deitelbaum, Catherine (2004). “Ethnic Conflict, Group Polarization, and Gender Attitudes in Croatia”. Journal of Marriage and Family. 66 (5): 1089–1107. doi:10.1111/j.0022-2445.2004.00080.x. Myers, DG (2007). Exploring Social Psychology: Fourth Edition. McGraw Hill.
