Introduction to Artificial Intelligence (Stanford online course)

What it was

Introduction to Artificial Intelligence was one of three free online computer science courses that Stanford Engineering started on October 10, 2011^[1]. It was taught jointly by Sebastian Thrun, a Stanford Research Professor of Computer Science and a Google Fellow, and Peter Norvig, Google's Director of Research^[1]. The course was free and delivered over the web, and it was not limited to students enrolled at the university: anyone who could reach the site could take part^[1].

How the course was built

At the core of each of the three courses was a set of prerecorded mini-lectures, stored on multiple servers^[1]. Teaching therefore did not happen in a lecture hall at a fixed hour. What a student worked through was a body of short recordings, and the reason the recordings sat on several machines was that the same material had to reach a very large number of people at once.

Most of the videos contained quizzes inside the frame, and those quizzes were graded automatically^[1]. A student watching a lecture would be asked a question partway through, answer it, and find out immediately whether the answer was right. The exercises, the mid-terms and the finals were also all scored automatically^[1]. Automatic scoring is what makes a class of this size workable: no group of human markers could return work fast enough for a student body in the hundreds of thousands, so the questions had to be written in a form a program can check.

The third component was discussion. All three classes had an online forum where students could ask and answer questions, and where peers could vote answers up or down^[1]. With no classroom and no teaching assistant answering every question in person, the voting gave the forum a way of sorting answers, pushing the ones other students found useful toward the top and the rest out of sight.

The numbers

Stanford says the course attracted 160,000 students from over 190 countries^[1]. The scale also showed up in the machinery behind it: the course database was being queried more than 7,500 times a second^[1]. Some of the work around the course came from outside the university as well. Online volunteers had begun translating the AI class into more than 40 languages^[1].

Where it sits

For a university, this was a test of pulling teaching away from the limits of a campus. A recorded lecture, an automatically graded problem set and a student forum can, in principle, serve as many learners as the servers can carry, and the Stanford course put that idea in front of a very large audience all at once.

The combination it used — short videos with questions embedded in them, machine-scored homework and exams, and a forum where answers are voted on — is a compact description of how large open online courses are usually organised. Courses of this kind, free to join and enrolling far more people than any single institution could admit, later came to be called massive open online courses (MOOCs). The figures Stanford recorded for this class give a concrete picture of what happened when a university course was opened to the web at that scale^[1].