People sign up and never finish because the courses are downright awful. And there's no mind nor incentive for them to get better. Instructors think that just recording a lecture and putting it online is good education, but it isn't.
Watch Daphne Koller droning on about graphical models as the video shows her standing at a lectern talking, or showing a powerpoint-style frame while she reads the text on the frame to us.
Watch Anant Agarwal go through a *hugely* dense and boring derivation, only to stop before the end and say "but this derivation is too hard, there's an easier way". Twice. For the same result.
Try to figure out how many degrees of freedom a soccer ball has, then argue with Sebastian Thrun because the answer he thought you should have entered is not the mathematically correct one. (Also, see if you can figure out what this has to do with AI.)
For a breath of fresh air, watch Donald Sadoway take you through a delightful and satisfying explanation of chemistry. (Ignore the 1st lecture which is about class scheduling.) It's wonderful.
I could cite two dozen *major* problems with selected online courses - things that go counter to the fundamental goal of learning that would be obvious to someone familiar with human learning mechanisms or a testing group or even a member of Toastmasters. When I point these out to the chief scientist at edX, he responds with "we can't change the way we do things because of X".
Let me repeat that: the *chief scientist* at edX has no control over teaching techniques or video methods or course quality.
Some people (ie - Dr. Sadoway in the link above) have figured out how to do it right, but the vast majority aren't interested in quality. It's unfortunate that edX got all those millions in seed money, because we'll have to wait until they burn through it before they get hungry enough to worry about quality and effectiveness.