Note: This is a guest post from Yaser S. Abu-Mostafa. He is a professor at Caltech. His MOOC on Machine Learning starts at edX on Sep 30. He is the co-author of Amazon’s bestseller on machine learning “Learning From Data”

MOOCs have 3 distinct but interrelated roles. One is outreach; making a top university course available to everyone. Another is the flipped classroom model; allowing a regular course to be taught on campus in a different way, with class time dedicated to discussion while the lectures become a “video textbook” watched at home. The third role is collecting data on education, which takes advantage of the online delivery medium to determine what works and what doesn’t work in teaching a course.

My own experience with MOOCs, which started a year and a half ago when I taught “Learning From Data” on my own Caltech platform, tells me that all three roles are real and significant. I have personally observed in my travels the impact of my course on students all over the world. It makes the huge effort that went into producing a quality MOOC very much worthwhile. I have also experimented with the flipped classroom model at Caltech, and the feedback I got from my students was quite positive. As far as collecting data on education, I have made pointed improvements in my course based on the data I collected in my MOOC platform, and I have no doubt that major platforms like edX are accumulating treasures of data that will profoundly improve the delivery of courses and the learning experience.

Is there any downside to MOOCs then? There can be, if we are not careful. I will address two potential pitfalls here; the dilution of courses and the diminishing learning experience.

The source of the first pitfall is the incentive system of MOOCs. The success metric in many people’s mind has been the number of students who take a given MOOC. While numbers matter, this is not a popularity contest. It’s real education. The numbers game creates an incentive to dilute the contents of the course for popular consumption. The overemphasis on the “fun” part can also come at the expense of rigorous learning. It pushes MOOCs in the direction of becoming video games. For instance, there is a trend to cater to shorter attention spans in MOOCs. IMHO, this is a harmful trend. Serious science requires a significant attention span, and part of the educational process is to train the students to sustain longer attention spans.

Let me mention one of the ramifications of the numbers game, which is the for-profit versus non-profit business model for MOOCs. There is no escaping the fact that, by law, a for-profit company has a fiduciary duty to maximize profit, even if this comes at the expense of academic standards. The good intentions of the company officers cannot overrule this legal obligation. This is why I feel that the for-profit model is not the right model for MOOCs. There is a reason why all worthwhile universities are non-profit.

The second pitfall is the over-reliance on MOOCs. Universities may decide to fully outsource their educational mission to MOOCs, depriving their students from the personal learning experience that comes from interacting with instructors and being in a classroom with fellow students. I believe that the flipped classroom is the right model for incorporating MOOCs into university curricula. Instructors can use the class time for discussion, which is a wonderful way of making the course material settle in. This also enables instructors to venture into courses that are not at the core of their expertise, thus expanding the university curriculum.

All in all, I believe that MOOCs are here to stay and that they will evolve to create a better learning experience for a much wider student body. I also believe that having classmates from the four corners of the world will be a wonderful contribution to the human experience at large.