7 great things Duolingo teaches us about good online learning

Luis von Ahn is the brains, the driver and innovator behind Duolingo. From Guatemala, he’s a mathematician and computer scientist on a mission to keep language learning free. The story is fascinating. It was valued at $1.5 billion at the end of 2019 and their revenues have risen through the pandemic from $400,000 per day to $600,000 per day, with only 20% of revenues in the US. They have x50 more users than their nearest competitor and x5 the revenues.

The primary problem in language learning is motivation. This is where design and AI come in. The bite-size learning chunks and highly visual sense of progress and completion of levels is exactly what is important. This is the good side of gaming. I’ve been involved in adaptive learning for years and really do believe that it offers huge promise in efficacy in learning.

Doulingo is all about making learning habitual. This is the magic dust. I’ve written about h-learning before and it has a long theoretical history. The daily tasks and streaks are achievable and you get visual rewards as you progress. The behavioural science behind the formation of habit is good. It just feels good. This is driven by good design but mostly by clever AI.

Duolingo employ high end AI experts and pay top dollar for this expertise. The personalisation and adaptivity is sophisticated, as it knows what you’ve learnt and, importantly, if you’ve been absent, what you’ve forgotten. This is important. Algorithmic personalisation may have more to do with rectifying forgetting than learning.

But the real application of AI is even more interesting in notifications. They are extremely sophisticated when, algorithmically, they decide what to say and when to say it. This is the clever use of data to automate the learning process, to keep people going. They notify you regularly, but not too much, to keep you going. But the most effective notification is the ‘final warning’. If they feel you have dropped off a timely message, making you feel slightly guilty, works wonders.

This matters. Simple, clean, plenty of white space, consistent palette, no teacher face or teacher avatar, simple progress bar at top of screen. Then there’s scoring, levels, daily goals, completions, green for success, red for failure. Duolingo also works superbly well on mobile and has keyboard as an option. 

No clumsy drag and drop, open input for full phrases and sentences, allows people to type what they hear, remediation when you fail, sentence as audio when you get it right, not scared of repetition, single day streaks, spaced practice. They work hard at this. I’ve seen it improve year after year.

Throughout this whole journey, from late 2012, the primary aim was to keep the service fundamentally free. This was and still is their mission. They are all zealots for free education and hire top-end people at good salaries, who believe in this mission. Keep it free, well largely as only 3% of users pay the subscription) – learning wants to be free.

Duolingo went through some serious innovations and pivots. Luis invented CAPTCHA which was successful in digitising books and newspapers and was keen on similar free solutions to problems that, as a by-product, solved other problems. With Duolingo he tried selling translations but that market was commoditised. He then tried ads at the end of each module, so as not to interrupt the learning process. But his final pivot was subscriptions to avoid ads. This solved the problem. 

His critics, who say that Duolingo doesn’t teach you languages, miss the point he thinks. Most of these apps are about picking up the basics. Learning a second language takes a long time and Duolingo aims at the foothills. Their goal is to get that process kick-started and aim for intermediate level B2, and they’re getting there. B2 by the way is the level of English required to work at Google. He also claims, rightly, that the enormous sums spent in schools, trying to teach languages is a disaster zone, with a tiny fraction ever getting any functional proficiency. Remember, Duolingo is free.

There are other apps. 

Memrise is perhaps better at dialogue with more video clips but clumsier interface and design. It also uses AI and has a Fremium model to add features such as a grammarbot, pro chats, difficult words, speed review, listening skills, learning stats and more.

Babbel is a German alternative but a subscription service. The voice recognition software (uses AI) can be a bit annoying but it also now adopting AI as the driver. It a more traditional structured lesson approach.

Bussuu is London based and another AI-driven app that operates a Fremium model. It is flashcard based but a more social app, allowing you to speak into the app.

There’s debate about what is best but my point is that all use AI to drive pedagogy, go for the bite-size thing, focus on habit and motivation. I still prefer Duolingo.