Koller and Ng eventually realized that they could achieve both their goals — outreach and on-campus reform — by pooling their efforts. In late 2010, they started work on a software platform that would support discussion forums, video feeds and all the other basic services of an online course, so that an instructor only had to provide the content. But making social interaction work on a large scale turned out to be a research project of its own, says Ng. For example, standard online discussion forums are a fine way to bring communities together — for 100 or so users. “With 100,000 it gets more complicated,” he says. Hundreds of students might end up asking the same question. So the developers implemented a real-time search algorithm that would display related questions and potential answers before a student could finish typing. Ng and Koller also let students vote items up or down, much like on the link-sharing website Reddit, so that the most insightful questions would rise to the top rather than being lost in the chatter.
The two researchers even set the system up so that students could mark one another's homework for essay questions, which computers can't yet handle. Not only is such a system essential to scaling up learning, says Koller, but it also turns out to be a valuable learning experience. And experiments have shown that if the criteria are spelled out clearly, grades given by the students correlate strongly with those given by the teacher (R. Robinson Am. Biol. Teach. 63, 474–480; 2001).
By early 2011, Ng and Koller were planning to demonstrate the platform on campus, and other faculty members were paying attention. Among them was Thrun, a robotics researcher who was splitting his time between Stanford and Google in Mountain View, where he worked on the development of driverless cars.
It was Thrun's idea to go big, using a platform of his own based in part on Ng and Koller's ideas. He says that he was scheduled to teach an artificial-intelligence course that autumn, along with Peter Norvig, Google's director of research, “and I thought it was a social responsibility to take it online, so we could reach more than the 200 students we would get at Stanford”. But even he hadn't imagined how big it would get. This was the course that registered 160,000 people from 195 countries after just one public announcement, a post to an artificial-intelligence mailing list. “It shocked everybody,” he says.
In response, Ng took Koller's machine-learning course public using their platform, while department chair Jennifer Widom did the same with a database course. Each attracted roughly 60,000 students. With those numbers, venture-capital funding quickly followed.
Thrun announced his company Udacity in January 2012. Arguing that most professors don't have a clue about how to exploit the online medium, he and his colleagues elected to develop their courses in-house, working with education experts to make the pedagogy as effective as possible.
Ng and Koller announced Coursera in April 2012, and took the opposite tack. They partnered with big-name universities — Stanford and three others, to start — and let them provide the content while Coursera provided the hosting and software platform.
Anant Agarwal, former head of the computer science and artificial-intelligence laboratory at MIT, had been experimenting with online learning for a decade, developing an electric-circuit simulation package called WebSim that tried to give online students an effective substitute for hands-on laboratory experience. In December 2011, inspired by goings on at Stanford, he launched MITx: an independent, not-for-profit company that would offer massive online courses from MIT on an open-source basis. It became edX in May 2012, when Harvard joined.