In 2022, Duolingo introduced a new feature: the Path. Up until that point, Duolingo users could choose the direction of their lessons and pick their favourite topics, under a system that is now known as the Tree. The Path removed choice from the Duolingo menu, placing users along a predetermined, uniformed lesson plan that they claim is optimised for learning.
Trading choice for efficiency is a change that angered many of the more vocal users on social media, but taught more people Spanish and made the company more money. And to most people embroiled in the controversy, including the CEO, that is where the story ends. A big fuss, but an overall positive consequence.
But many users with autism, ADHD, and other learning neurodivergencies noticed another problem: many of them weren’t getting as much benefit from the new system as neurotypicals.
Duolingo CEO Luis von Ahn claims that Duolingo has performed studies that demonstrate the superiority of the path for learners. And I am certain he is telling the truth. But the findings he has cited are one-dimensional. There is only one kind of analysis. The only number Luis cares about is the average. And that’s not how you study learning properly.
Learning styles differ between different cultures and neurotypes. When you study learning responsibly, you must look at this data. Sure, Duolingo helps the average learner, but does that come at the cost of certain demographics? The answer is yes. I am unfamiliar with the cultural and racial aspects of this issue, so I will focus on the neurodiversity aspect:
Some autistic and ADHD people struggle with the Path.
Not all. In addition to the purported efficiency of the Path, there are two issues at play: decision fatigue, and choice engagement. Some neurodivergent (ND) people suffer from decision fatigue at a greater degree than neurotypical (NT) people. For these people, the Path is great because they have less decisions to make and will be less fatigued. Some ND people rely more heavily on choice engagement to continue invested in a task than NT people. For these people, the Path is bad because it leads to lack of motivation.
When the Path was introduced, users could choose between the Path and the Tree. Obviously, this is the most neurodiverse-friendly way to do things. People are diverse, because learning is diverse. You give everyone access to the tools that will help them the most. This isn’t a foreign concept to app design; the Path and Tree could be renamed to Simple View and Advanced View. This is a common feature in app interfaces, people understand this and will choose appropriately.
This feature was patched out within a couple of months. Luis cites as reason the increased dev costs it would take to maintain the Tree along with the Path. Continuing to use a legacy version of Duolingo with access to the Tree is not an option.
The Path is simply a different interface to access the same lessons, with a different balance of questions. The same questions are asked, it simply happens in a different order. Other companies have succeeded in maintaining multiple interfaces for the same content. Reddit maintains both old.reddit.com and new.reddit.com to view comments on a classic PC interface or a modern mobile-style interface. Many other apps have both a web version and a mobile version. Duolingo’s case is more involved, but fundamentally this is a solved problem and not a concern for other companies.
Duolingo’s inability to afford maintaining the Tree tells us that the lesson selector is tightly integrated with other parts of the app. The code is too interdependent to simply swap between the two systems easily. This is a violation of the software engineering principle of low coupling, and a failure to design maintainable code. Designing poor architectures like this always leads to more costs in the long term, even if it’s cheap and fast in the short term. Cleaning up code to fix issues like this should be done as soon as possible, and in an app like Duolingo that is supposed to be improved upon for a very long time, it’s always a good idea. The costs just from maintaining the path alone will be higher than they need to be because of this issue.
But spending money to clean up code for long term sustainability doesn’t look so good on a quarterly earnings report.
Wikipedia reports that Duolingo has taken 120.3 million dollars across 6 funding rounds from venture capital firms including Union Square Ventures, New Enterprise Associates, Kleiner Caufield & Byers, Google Capital, Drive Capital, and CapitalG. These are venture capital firms which typically expect a tenfold return on investment from “successful” startups like Duolingo. Duolingo was valued at 1.5 billion dollars in 2019, and that is roughly the amount of value it is expected to provide for its shareholders.
Duolingo reported its first profitable quarter in 2023.
Duolingo is already the most popular language learning app.
There are a finite number of people who want to learn a second language.
Duolingo will struggle a great deal to gain new users at this point. Its users don’t want to pay any more than they are now. Duolingo is expected to squeeze a billion dollars out of those users. This is an impossible situation.
In 1600s Netherlands, the price of tulips was massively inflated by the expectation that rich people would pay a lot of money for rare tulips. Tulip breeders would trade bulbs seeking to grow tulips from them and hope to by chance breed a rarer variety. A 17th century form of loot crates, if you like. Shortly after rose the profession of florist; people who would buy and sell tulip bulbs, and promissory notes for tulip bulbs currently in the ground, whose future value was unknowable. As demand for tulips from florists rose, so too did prices. Florists believed they could get rich by selling tulips to other florists. And some of them did. But in 1637 when everyone realised that tulip gambling was silly and the price bottomed out, the people who had sold their houses and paid a fortune for tulips found themselves with nothing to their name but crushing debt and useless flowers.
Duolingo is a Dutch tulip. It will never make a billion dollars in profit. It might, if it were allowed to slowly develop its service and increase profitability at a slow, comfortable rate over decades. But venture capital firms expect profits faster than that, and they will not get them from a legitimate business enterprise.
The only solution is enshittification: charge users more for the same services, defer maintenance costs to next quarter, create a more monetised experience, chase trends to increase short term user numbers, and underpay the employees to a degree just short of criminal. This is the only way Duolingo can make a billion dollars for its investors.
And this plan doesn’t involve treating disabled people with equal respect. This plan requires ruthless profit-chasing and mass appeal. There is no time or money to build a stable, maintainable codebase that will remain workable for decades.
Luis von Ahn is fucked. He has promised venture capital firms money that he will never be able to make for them. He will be squeezed by the relentless force of capital until he either sells Duolingo or goes bankrupt. Every force in his life will tell him to forget values like equality, patience, or education in favour of money. More of it. And now.
But I’m not sympathetic towards Luis for the predicament he’s put himself in. Because he’s the guy that created reCAPTCHA. And reCAPTCHA is participating in an arms race between ever-smarter image recognition bots and ever-harder image recognition tests that is beginning to leave disabled people behind. Bots are getting better at recognising images than people with visual processing disorders. And if the method of excluding bots is that they are disabled in comparison with neurotypical people, then disabled people will be excluded.
This is not Luis’ first time inventing technology that leaves disabled people behind the march of progress. He deserves his Dutch Tulip.
