Blog

Founders of DeepMind, Duolingo Focus on Tech’s Benefits

August 04, 2015

There are some striking similarities shared by Mustafa Suleyman co-founder of DeepMind, and Luis von Ahn, co-founder of Duolingo: Both are young, idealistic, non-U.S.-born entrepreneurs pushing the envelope of new technologies that also have groundbreaking social implications. DeepMind was founded in 2011 and bought last year by Google for $400 million, while three-year-old Duolingo is one of the most popular free apps on Google Play and on iPhones.

The stories part ways when it comes to approach, however: Duolingo’s claim to fame is the crowdsourcing nature of its iterative methodology for learning a foreign language.

London-based DeepMind takes a rigorous, programming-intensive dive into the world of artificial general intelligence (AGI) and machine learning. Both co-founders described their products and dreams at recent IDE events.

At the IDE annual meeting in May, Suleyman (pictured above) explained how his AGI systems employ neural networks and deep learning methods to solve tasks without prior programming. Using Atari computer games as its test case, DeepMind programs learned automatically–from recognizing raw images–how to reach high game scores with only pre-training instructions. The program figured out how to succeed at nearly 50 Atari games without any foreknowledge of how to play them.

Suleyman has high goals for AGI. Rather than fearing its power to replace humans and perform devious actions, he sees AGI tackling some of the world’s biggest problems including clean water access, financial inequality, fraud detection and reducing stock market risks. “Maybe AGI can shape a better world,” he claimed. Discussions over ethics and safety measures are certainly needed, along with verification and security methodologies, he acknowledged.

Google sees huge commercial potential in the technology and already has used AGI in its Streetview map app, photo apps and to replace 60 hand-crafted systems across the company. AGI has also made huge strides in speech recognition as in Android phones and Google Translate, which Suleyman said reduces transcription errors by 30 percent.

Clearly, “AI has arrived,” he said at the IDE event, though AGI’s full potential is still a long way off. The programs are weak at conceptualization, an area where humans can work with AI to add more abstract thinking.

Crowdsourcing Language Learning

Guatemalan native Von Ahn (pictured above) also aimed high when he tinkered with app dev as a young software designer. He created and sold two companies to Google (including reCAPTCHA) while still in his 20s before creating, Duolingo with co-founder Severin Hacker, and he also teaches at Carnegie-Mellon University.

His initial goal was universal education, he told attendees at the July 10 IDE Platform Strategy Summit. When he stumbled on the fact that two-thirds of the world’s 1.2 billion language learners are studying English to rise up from poverty, he decided that a free or low-cost app was the key to their success. Duolingo was the result.

His plan was to make it simple and game-like (offering bonus points and incentives for to move to the next level, for instance), to attract and keep learners. Soon, the site scaled rapidly and began crowdsourcing methods and tips so that students also become the teachers. Today, more than 100,000 schools are using the program and it has 40 million registered users—mostly outside of schools.

What’s evident from these two examples is the rapid-fire pace and unlimited potential of many digital technologies. Not incidentally, maybe they will improve the world along the way.

For more about MIT’s robotic efforts, read the blog here.

For more on DeepMinds see: