Duolingo has announced a major change, saying it is “going to be AI-first,” including replacing contractors with AI.

Duolingo is a popular app for learning new languages. Like many companies, Duolingo is looking for ways to use AI, but it may be taking it further than many others.

CEO’s Email to Employees

In a post on LinkedIn, the company shared an all-hands email from CEO Luis von Ahn regarding its AI-first plans.

“I’ve said this in Q&As and many meetings, but I want to make it official: Duolingo is going to be AI-first.

“AI is already changing how work gets done. It’s not a question of if or when. It’s happening now. When there’s a shift this big, the worst thing you can do is wait. In 2012, we bet on mobile. While others were focused on mobile companion apps for websites, we decided to build mobile-first because we saw it was the future. That decision helped us win the 2013 iPhone App of the Year and unlocked the organic word-of-mouth growth that followed.

“Betting on mobile made all the difference. We’re making a similar call now, and this time the platform shift is AI.

“AI isn’t just a productivity boost. It helps us gets closer to our mission. To teach well, we need to create a massive amount of content, and doing that manually doesn’t scale. One of the best decisions we made recently was replacing a slow, manual content creation process with one powered by AI. Without AI, it would take us decades to scale our content to more learners. We owe it to our learners to get them this content ASAP.

“AI also helps us build features like Video Call that were impossible to build before. For the first time ever, teaching as well as the best human tutors is within our reach.

“Being AI-first means we will need to rethink much of how we work. Making minor tweaks to systems designed for humans won’t get us there. In many cases we’ll need to start from scratch. We’re not going to rebuild everything overnight, and some things—like getting AI to understand our codebase—will take time. However, we can’t wait until the technology is 100% perfect. We’d rather move with urgency and take occasional small hits on quality than move slowly and miss the moment.

“We’ll be rolling out a few constructive constraints to help guide this shift:

“- We’ll gradually stop using contractors to do work that AI can handle
“- AI use will be part of what we look for in hiring
“- AI use will be part of what we evaluate in performance reviews
“- Headcount will only be given if a team cannot automate more of their work
“- Most functions will have specific initiatives to fundamentally change how they work

“All of this said, Duolingo will remain a company that cares deeply about its employees. This isn’t about replacing Duos with AI. It’s about removing bottlenecks so we can do more with the outstanding Duos we already have. We want you to focus on creative work and real problems, not repetitive tasks. We’re going to support you with more training, mentorship, and tooling for AI in your function.

“Change can be scary, but I’m confident this will be a great step for Duolingo. It will help us better deliver on our mission—and for Duos, it means staying ahead of the curve in using this technology to get things done.

“—Luis”

Duolingo Is Betting Big On An Uncertain Future

Duolingo’s CEO clearly believes AI is the future, but there is growing concern about exactly what that future looks like. The challenges encompass both technical and legal issues.

From a technical perspective, AI companies are still struggling to address basic issues, like hallucinations. For example, OpenAI’s latest ChatGPT models actually hallucinate more than previous models, not less, as is often the case with improved versions.

To make matters worse, researchers have discovered that AI models are at risk of developing ‘Model Autophagy Disorder’ (MAD), the AI version of mad cow disease. AI models’ abilities quickly devolve when they are trained on existing AI content—a growing issue considering the sheer amount of AI content being generated.

“The problems arise when this synthetic data training is, inevitably, repeated, forming a kind of a feedback loop ⎯ what we call an autophagous or ‘self-consuming’ loop,” said Richard Baraniuk, Rice’s C. Sidney Burrus Professor of Electrical and Computer Engineering. “Our group has worked extensively on such feedback loops, and the bad news is that even after a few generations of such training, the new models can become irreparably corrupted. This has been termed ‘model collapse’ by some ⎯ most recently by colleagues in the field in the context of large language models (LLMs). We, however, find the term ‘Model Autophagy Disorder’ (MAD) more apt, by analogy to mad cow disease.”

In addition to the technical challenges, AI firms are facing legal challenges in multiple legal jurisdictions, with Thomson Reuters recently winning a major victory vs AI firm Ross Intelligence over copyright violations.

Meta is similarly locked in a legal battle over pirating millions of books to train AI, even going so far as to say they didn’t need to pay the authors since, “for there to be a market, there must be something of value to exchange, but none of Plaintiffs works has economic value, individually, as training data.”

Duolingo clearly believes the potential benefits outweigh the risks, but only time will tell if this gamble pays off as well as its mobile-first bet did.