Daphne Koller’s TED Talk: What’s new about MOOCs

August 14, 2012 at 7:59 am 17 comments

Daphne Koller of Coursera has a TED talk that was just posted this month (linked here and below) that I highly recommend.  Daphne does a good job of explaining the rationale for Coursera, and offers several example snippets from existing courses (including a couple from Nick Parlante’s course doing Media Computation in JavaScript.)

Daphne talks about the educational research that she’s drawing on. I wondered: What’s new here? Why are people excited about MOOCs?

Here is what I see as new:

  • Video on the Internet. There is an effect of medium and distribution here. Video is compelling. We now have the ability to get lots of video created and shipped anywhere cheaply. When Roger Schank was building his learning systems at Northwestern, they spent a huge amount of effort getting lots of video burned to DVD’s that could be easily accessed. That’s simply not a problem anymore.
  • They’re doing it for free. There have been lots of smallish research efforts in the past. There have been companies started that provide these technologies at scale for a cost. Free changes things, particularly with students and families today bearing a greater portion of the cost of higher education.
  • There is the potential to do more, to make students feel like individuals, rather than part of a 100K herd. When I raised the question of “what’s new about MOOCs” with faculty at Georgia Tech, my colleagues pointed out the potential value of using modern, real-time machine learning and data analytics techniques to get greater insight into learning difficulties, and to better personalize the learning experience.  Daphne says in her TED talk that the Coursera system could recognize the need for more remedial material and provide it.  I recognize that potential, though the technology isn’t in place yet. Current MOOCs have little or no machine learning, and no attempt at personalization.But I see a problem with Coursera recognizing a need and recommending remedial material.  The current MOOCs won’t be able to offer personalization for the audiences that I most care about (e.g., adult learners without previous CS background, non-majors studying CS), audiences that probably would need more background material than the top students, because those students simply aren’t there.  My audiences are most likely in the 80-90% who are dropping out of MOOCs after registering. Even the most sophisticated machine learning and data analytics can’t help you to understand students who are no longer there.  Until you get students who need the remediation through the system, the ML can’t learn about them, but how do you get them through the system without the ML-recommended remediation?

While I agree with the importance of reaching underserved populations, I am not convinced that MOOCs are currently having much of an effect in the developing world or to broaden participation to students who don’t have much prepratory work (say, in CS) in their schools.  I wonder if it’s even possible to make a large impact on the developing world starting at higher education.  Not all K-12 programs in the United States prepare students adequately for MIT, Stanford, and Harvard level classes.  Can we expect that most K-12 programs in the developing world are adequate preparation?  The Open University UK has always been “open,” no pre-requisites, and they provide content at that level.  Coursera prides itself on offering top-notch classes.  That’s valuable, but I find it unlikely that such courses also meet the needs of underserved populations.

Coursera offers demanding courses via video which only a small percentage complete — for free.  That is valuable and interesting.  I don’t currently see the model replacing existing courses, or working well for students who don’t have the background knowledge.

Daphne Koller is enticing top universities to put their most intriguing courses online for free — not just as a service, but as a way to research how people learn. Each keystroke, comprehension quiz, peer-to-peer forum discussion and self-graded assignment builds an unprecedented pool of data on how knowledge is processed and, most importantly, absorbed.

via Daphne Koller: What we’re learning from online education | Video on TED.com.

Entry filed under: Uncategorized. Tags: , , , , .

Cost of college remarkably stable « Gas station without pumps Khan Academy offers kind-of-scaffolded computer science learning: Doing away with the teacher

17 Comments Add your own

  • 1. Deepak Kumar  |  August 14, 2012 at 8:15 am

    Right on, Mark. My guess is that MOOCs will do for interest in education and learning the same way iTunes did for interest in music. Folks will engage in a MOOC and that will help motivate them to learn more, through traditional models of formal education. That is about it. All other claims being made are baseless, in my opinion.

    So, here in also lies a small monetization aspect. Charge a very nominal amount, like $10-25 per course. No credentials, no certificates, no degrees. Just for someone’s own personal interest and choice. With supposedly 100k learners…they will make $1 million+ per course….ok, if anyone uses this idea, buy me a beer for royalty 🙂


  • 2. lenandlar  |  August 14, 2012 at 8:39 am

    I work at a Uni and i can hardly afford more than 256k bandwidth. The videos available on Coursera and other MOOC providers are almost impossible to access with this bandwidth. The access problem is real for developing countries.

    While i believe the course providers are genuinely interested in reach, i believe they also know what subset will be able to make any sense and use of the courses.

  • 3. Jeff Rick  |  August 14, 2012 at 9:07 am

    Fascinating. One concept that is starting to interest me is the separation of learning from accreditation. Universities marry the two concepts together. Students need to learn a lot and then get an accreditation, usually in the form of a degree. Why not separate the two? A highly motivated, intelligent student might be able to learn more quicker from Coursera than Georgia Tech. The only thing they’d be missing is the accreditation (I’m discounting the value of Coursera certificates).

    One idea that occurred to me is opening a pure accreditation university. The student pays $10,000. For that, they spend one month on campus, demonstrating their skills. The university would then provide a precise and fine-grained assessment of what that person is capable of. This could then be officially forwarded on to companies. In the CS domain, I think this would be an immediate hit. A good programmer is worth 10 mediocre programmers. A great programmer is worth 10 good programmer. Let’s assume that all people that can get a BS in CS from Georgia Tech are at least mediocre programmers. My experience is that grades are fairly imprecise to get beyond that evaluation. As such, companies spend 1 to 2 hours interviewing people to assess where they are. That definitely helps, but is also problematic (e.g., I get so nervous during interviews that my skill set drops dramatically). The month long assessment would do a much better job and would be highly valued by top employers and top students.

    • 4. richde  |  August 14, 2012 at 9:37 am


      Aside from the obvious difficulties in the “what is new here” argument (Examples abound: what was new in Google?, Kerry Mullis won a Nobel prize for PCR which hooked together old ideas,…) the problem with taking a snapshot like:

      “Current MOOCs have little or no machine learning, and no attempt at personalization.But I see a problem with Coursera recognizing a need and recommending remedial material. The current MOOCs won’t be able to offer personalization for the audiences that I most care about”

      is that it registers a point in time not where the field is going. What current MOOCs do or have is much less interesting than what they will do in five years.

      • 5. Mark Guzdial  |  August 14, 2012 at 10:11 am

        Hi Rich,

        I agree — I’m describing a point in time (n=0), not what things will be like in five years. Let’s imagine that we’re doing an inductive prediction of where MOOCs will be in five years. What are the incentives or forces for change that will drive MOOCs into n+1 and beyond? Currently, Coursera has no revenue. Are they more likely to get a revenue stream from the top students or serving the students needing remediation? I do believe that they will implement machine learning algorithms and personalization. I’m not convinced that it will be in their interest to reach the students who aren’t succeeding with the current model.

        • 6. beki70  |  August 14, 2012 at 11:17 am

          Perhaps a difference here is something that resembles a question about where the complexity lies. Google’s complexity, while perhaps not novel, is under Google’s control. They build the software. Education seems to me to distribute the complexity of the system more towards the users than Googles. It relies on people in some interesting ways: sustained engagement, their ability to learn, whatever pre-existing learning has past makes a difference, not to mention their connectivity, access to resources, and so forth. Just a thought?

  • 7. mrstevesscience  |  August 14, 2012 at 11:22 am

    While right now I do not “see the model replacing existing courses, or working well for students who don’t have the background knowledge.” I do see it augmenting these course and for a small percentage of students replacing them (those highly motivated and self disciplined and all to small population).

    I used the CS101 with a group of 8 12 to 15 year olds. Where they took the online Coursera course and if they had problems they could email me or discuss in class which met once a week. When one student was stuck on one problem and shared their code, I would make suggestions and ask them to look at certain things and then share those answers with the whole class via email or in class. I would do the same if I though a student came up with an interesting solution or one that I thought could be used to make or re-enforce a bigger idea to the entire class. During the class time I would cover the topics or similar topics in different ways. I even used some of your material from your Python course for HS (re-rendered in Etoys) to demonstrate the idea that its all numbers and allows them to directly play with the numbers and have the images/colors in a “lively” way. Next time I will probably try your image to spreadsheet tool (as this make it more concrete for the kids) as well. I also did a Smalltalk version of the “green screen” along with a “worked example” to try and show the kids the same concepts in a different language and provide them a simple tool to green screen themselves into different scenes.

    If I had tried to develop this on my own or used a textbook, I never would have gotten the quality of material or even had the time to do the course. Thus I believe the kids got a much better experience by having someone who had a very good understanding of the concepts and the ability to explain them, combined with a resource to help answer questions, provide supplementary/complementary materials and guide them along the way.

    • 8. Mark Guzdial  |  August 14, 2012 at 9:57 pm

      Steve, I agree that mixed models, like what you describe and what Fred Martin has been doing, are among the most interesting uses of MOOC’s that I’ve heard about. So cool that you’ve been able to use some Media Computation in what you’re doing — I appreciate you letting me know!

  • 9. jazz  |  August 14, 2012 at 8:12 pm

    Reblogged this on run( ) { and commented:
    One of the most balanced analysis on MOOC’s of recent frenzy.

  • 10. Mark Guzdial  |  August 15, 2012 at 8:50 am

    Someone named “Erin” posted the below comment, which is a (mostly) fair comment, but gave an invalid email and URL, so I spammed it. I’m in favor of an open discussion (and I actually don’t want MOOC’s to fail at all! I just don’t think they solve the problem that I want them to solve and that Daphne seems to be saying they can solve), but I don’t want to encourage Slashdot-style anonymous critiques:

    What was new about Apple and Google? Or media computation? It is certainly easy to dismiss MOOCs and Daphne Koller’s contributions on these grounds, especially since they are still in their pre-infancy. But, whether you like it or not, this is where groundbreaking innovations in education are now most likely to occur. That is why people are excited about them—for their potential to transform education. You may be convinced that they’re going to fail (and hope that they do so), but alot of smart people think otherwise. Only time will tell who is right.

  • 11. Alfred Thompson  |  August 15, 2012 at 9:59 pm

    Getting credit for courses is going to be the thing that pushes these courses past the tipping point. There just are not that many people interested in learning for the sake of learning. Now in theory knowing more “stuff” should make you a better more productive person. I believe that to be true in practice as well but a lot of people are like the scarecrow in the Wizard of Oz and need that paper to validate what they know.

  • […] Daphne Koller’s TED Talk: What’s new about MOOCs « Computing Education Blog – August 15, 2012 […]

  • 13. jasont  |  October 5, 2012 at 10:02 am

    TL;DR. ML is step 12 for this problem, start at 1 with pronounced prerequisites enforced with a quick quiz, and utilize simple time and attempt metrics to morph the DOM to present learning supplements.

    I’m a CS major who left university to help found a small business in financial software in Chicago and I’m currently enrolled at Coursera in some ML and math courses.

    I have often wondered about the ‘steep’ learning curve – actually flat learning curve 😉 – that most detached from CS would run in to with many courses on offer. I see two things that could help push these people further and increase the % of completion:

    1. Utilizing ML may be overkill for the task at hand. Surely there are statistics Coursera holds for students that are simple like the time taken to complete a quiz, number of replays for a lecture, etc., similar to khan academy’s dashboard for teachers. JS can manipulate the DOM depending on where the student is having issues and display helper links or relevant forum discussions.

    2. Make prerequisites more pronounced to prevent unprepared students from tackling too much. One way would be via learning gamification i.e. badges. Short 5 minute exams allowing you to level up to garner a badge that allows for entrance to a particular course. Failing exams could present simpler intro courses or point to resources for picking up prerequisite knowledge.

  • […] I met is teaching MOOCs in Coursera or Class2Go or Udacity, and I got the chance to meet with Daphne Koller, too), where I was asked several times about what questions I would like answered about MOOCs.  I […]

  • […] unchanging between content domains, that does not change for individual students (I know that they hope that it will one day, but it doesn’t now), that filters and certifies those who can learn on their own. The role of education in society is […]

  • […] at this level of abstraction in terms of identifying good teaching.  I’m also concerned that data mining can’t help if you lose 80% of your subject pool — you can’t learn about people who aren’t […]

  • 17. Time to start MOOCing around? | Stephen Engel-Hart  |  October 12, 2013 at 11:29 am

    […] Y. Vardi’s ACM magazine article: Will MOOCs destroy academia? The Computing Education Blog: Daphne Koller’s TED Talk: What’s new about MOOCs Ghanashyam Sharma’s article: A MOOC Delusion: Why Visions to Educate the World Are […]


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Trackback this post  |  Subscribe to the comments via RSS Feed

Enter your email address to follow this blog and receive notifications of new posts by email.

Join 10,185 other subscribers


Recent Posts

Blog Stats

  • 2,060,428 hits
August 2012

CS Teaching Tips

%d bloggers like this: