I think this is a fundamental problem with STEM programs in general. Why have grades and tracks at all? The important thing is to learn the material, and not what grade you got on a test at some arbitrary point in time. This is what is so brilliant about MOOCs like Khan Academy. No grades, and you can repeat the material until you know it.
Actually, I think that's one of the strengths of MOOCs (or, at least, a symptom of its strengths.)
> The benefit of having a physical class is that you can turn to the professor or TAs for help if you're having trouble. It's also more difficult to completely fall behind because there are milestones in place (tests, assignments, projects, etc.).
The MOOCs I've been in EdX and Coursera all have had both milestones of the type described, and TAs to go to for help.
What they don't have is a large up-front cost to enter, so there's no reason for people not to try out classes when they aren't completely certain are the right fit -- and little reason, even if the fit is good, not to drop out of one iteration and come back and take the course again if other life events that do have a cost for not dealing with now erupt.
So, yeah, a higher rate of people starting MOOCs and not completing them compared to traditional courses is to be expected -- and that's a good thing.
Khan Academy definitely gets this right where other MOOCs really fail. The quizzes are actually quite engaging and the 'gamified' point giving, despite being incredibly cheesy, really seems to work.
I agree that there is no replacement for TAs and the professor, and I think that's where a lot of confusion about MOOCs stems from. This is where 'flip' teaching comes about. Instead of wasting time doing daily lesson planning and grading papers, the TAs and the professor spend their time helping students out who are having trouble. The students can watch/re-watch pre-recorded lectures and then take quizzes during class time which can be graded by the computer.
The quick program for Harvey Mudd seems to be 1) Make the problems more practical in application (ie controlling a robot, modelling a disease) 2) Giving students choices in what problems they are interested in 3) Segmenting students according to skill set, thus putting people who have been programming in another course that will match their pace better and allowing students who are more new to have their own pace.