Even before their first day of class, many students enrolled in Stanford's first set of comprehensive, free online computer science courses were hard at work.
The three classes - Introduction to Databases, Machine Learning and Introduction to Artificial Intelligence - officially started on October 10. By then, however, the 66,000 students signed up for Computer Science Department Chair Jennifer Widom's database class had collectively viewed 290,000 videos, taken 10,000 tests, asked 224 questions and offered over 2,000 replies.
The 72,000 registered students in associate professor Andrew Ng's Machine Learning class had watched more than 850,000 video clips and submitted some 50,000 quiz answers. And the artificial intelligence course, taught jointly by Stanford Research Professor of Computer Science and Google Fellow Sebastian Thrun and Google Director of Research Peter Norvig, had attracted an unprecedented 160,000 students from over 190 countries who were collectively querying the course database at more than 7,500 times a second.
"We pre-launched about a week and a half ahead, putting out all the materials for the first week and also the videos for the second week," explains Widom, who had emailed those who'd expressed initial interest in her class to tell them it was time to formally register.
Like her colleagues, Widom expected only a subset of that group to make the transition. After all, although the School of Engineering has previously offered recordings of entire Stanford courses online, students in these new classes will also take quizzes, complete assignments and even sit for exams.
"I had thought we'd see quite a bit of attrition," she says. "But we're getting them all back and more."
Classes attract worldwide enrollment
Registrants have signed on for the classes from as close as Stanford's home in Palo Alto and as far away as Ghana, Peru, Russia and New Zealand. Overall, roughly 40 percent are from the U.S., with India accounting for the second largest block of students.
One major country is missing. "The Great Firewall of China is blocking our video," says Ng. "So we aren't able to serve China very well right now."
Much of the interest in the courses has spread virally, Ng adds. "If you watch the web analytics, sometimes a whole country seems to come online at the same time as word spreads in that country's local social media."
New communities - both virtual and real - have sprung up around the classes.
Online volunteers have begun translating the Introduction to Artificial Intelligence class into over 40 languages, and in many places local study groups are meeting in person to talk about the courses.
Social communities a key element
Each course is explicitly designed to capture the community-mindedness of its students. All feature an online discussion forum, for example, where students can ask questions and suggest answers to those questions, both of which can be voted up or down by their peers.
That's essential, says Widom, as she couldn't possibly answer every question posed by tens of thousands of students. And already, she says, "I can see that students are answering each other's questions and that answers from the top students are going to bubble up."
"It all reflects the enormous desire of the world community to have something like this," suggests AI professor Thrun. "It is taking Stanford into the world in a way that I think has never happened before."
Although the online students do not get Stanford credit for their work, they gain access to faculty and three of the Stanford Engineering's most popular computer science courses. They're also participating in an experiment that could transform the way online education is delivered.
None of the classes could exist unless they were also highly automated. Each has at its core a set of prerecorded mini-lectures stored on multiple servers that students access at their own pace as each week's materials are posted. Most of the videos contain in-frame quizzes that are graded automatically. Exercises that test coding skills and mid-terms and finals are also all automatically scored.
Some of the testing technology goes beyond simple multiple choice questions, employing a technique based on the work of Stanford computer science professor emeritus Jeffrey Ullman. Ullman's testing system stores more possible answers than it offers for each question, which allows a student to take the same test multiple times.
"I've been using it for the last eight years in my Stanford classes," Widom reports, "and students really like it. They can take the tests over and over until they get 100 percent, and most of the students do."
Widom's Introduction to Databases and Ng's Machine Learning classes also build on innovations conceived by Stanford professor of computer science Daphne Koller and implemented by her colleague John Mitchell. Their CourseWare platform lets faculty upload videos and handouts, offer interactive online quizzes and track discussions among students and teachers.
Koller has used CourseWare to shift her lectures, quizzes and exams to the web - a move that's freed class time for discussions, guest lectures and interactive problem solving. "At the outset of this effort, we had the idea that we could take the same concept and make it available to the general public," she recalls. "But I think it moved faster than any of us expected."
Because CourseWare was designed to manage only the enrollment of a typical Stanford undergraduate class, Ng combined it with Open Classroom, a web platform he'd created to share Stanford lectures freely with the world. The resulting new platform, which has yet to be named, was re-engineered in a few short weeks this summer by Stanford computer science PhD student Jiquan Ngiam and undergrads Frank Chen, Chuan Yu Foo, and Yifan Mai.
The sheer scale at which people are engaging with the new Stanford offerings provides opportunities for improving education, both offline and on.
Koller, who plans to offer one of her own courses to the public soon, will use that class to explore the effectiveness of different kinds of tests and then refine her on-campus classes. With tens of thousands of students, she says, "you can assign 50,000 students to track A and 50,000 to track B, and then you have enough numbers to tell if a design change is making a difference or not - which is very difficult when you have a class of 50 or 60 students that you can't really split into two."
For Thrun, a major incentive is the chance to improve the quality of online education so it can reach more people more effectively. To that end, he's excited to see how changes in his materials and quizzes will affect learning outcomes. "It's a wonderful platform to study the effectiveness of teaching," he says.
By allowing students to do more than just watch a series of lectures for free, Ng hopes Stanford's three new online classes will help show that the technology exists today to "give a high quality education to a large number of people either for free or at very low cost."
Being able to take quizzes, wrestle with software problems and receive the feedback that comes with accurately graded work all makes for a much more valuable experience for those students, adds Koller.
More classes are being planned for the winter and spring quarters, she reports, including her own popular Probabilistic Graphical Models class.
"High-quality education is sorely lacking for people who have either financial or geographical limitations that prevent them from attending places like Stanford, MIT or Berkeley," Koller says. "Making these technologies widely available is a huge opportunity for making the world a considerably better place."