A course about and for robots

Fri 19 Aug 2011 10:24 PM

Two computer scientists at Stanford are going to be teaching a free on-line course in AI. As reported in the NY Times, there are now more than 58,000 students worldwide signed up for it. The course designers have set it up to be indefinitely extensible and have getten a lot of bandwidth, so they are undaunted. All interaction with students will be mediated. For example, student questions will be submitted to the collective and voted on by other students. The profs will give a public answer to whichever questions float to the top.

The Times mentions that one of the two instructors, Peter Norvig, is a coauthor of a standard AI textbook. The story doesn't mention whether students are encouraged to buy the book, but students are likely to buy it without encouragement. It should sell enough copies to earn Norvig some walking-around money.

Of course, these 58,000 are sign-ups on the internet. The Times does not ask the obvious question: How many of these are actual people taking the course?

There must be some number of duplicate enrollments, from people who had problems with the system. There must be some number of people who clicked through and enrolled just to see what this thing was. There must be some number of people who clicked through fully intending to take the course, but won't follow through on actually taking it.

Even counting the ones who start out in the course, the drop rate will likely be very high. And if following through on the course really does require linear algebra, many will quit in despair.

Moreover, we can't discount the possibility that some number of these are bogus registrations. Spammers don't actually stand to make money by enrolling zombie machines in an AI course, but it might amuse them.

Setting aside quibbles about the course and its 58K students, it did make me reflect for a moment on the economics of running a university. The course is not for credit, so there's no direct comparison here. But Andrew Ng, who is teaching another Stanford CS course on-line, suggests a bigger picture. He is quoted as saying, "I personally would like to see the equivalent of a Stanford computer science degree on the Web."

For courses which are taught in a giant lecture hall, there might not be much value added for the student to be in the same space as the instructor. Getting the content on-line could be roughly as good.

In smaller classes, where is more opportunity to interact with professors, there's more difference. Yet there are plenty of students who are content to sit in the back and listen, even in a 30 person class. It's not clear that they gain much by physical proximity.

What I got out of a college education came mostly from being able to ask questions and suggest ideas, to engage in the game of giving and asking for reasons. Video lecture and social network Q&A would not have had the same value for me.

Here's the economic point, though: For every student like me who jumped head first into college and interrogated everybody, there were a bunch of disengaged students in the back who were mostly just listening. Their tuition was required in order for me to have the opportunity to engage. A more efficient system for them would have been a failure for me. There is an argument to be made that I was exploiting them, and anonymized on-line courses would have be a better use of their resources.

As an aside, it is possible to have small classes on-line. So the opposition isn't really between meatspace and on-line courses. Rather, it's between indefinitely large, effectively anonymous courses and courses in which students and the teacher interact directly.