Stanford on-line AI course draws 58,000 — but is it real?

August 18, 2011 at 10:34 am 15 comments

58,000 people is a huge demand for an on-line AI course offered at Stanford.  That really argues for the demand for on-line education.  But what kind of a course is it?

From the description in the NYTimes piece, there will be video lectures, and interactive “quizzes and grading.”  I suspect that the traditional, face-to-face Stanford AI course is more demanding. Our AI courses at Georgia Tech have a significant programming requirement, and have a long pre-requisite chain, including courses in software engineering and data structures.  The AI course at Stanford will be joined by a database course and one on machine learning.  Our database course requires using SQL to generate queries, and our machine learning course has a significant math requirement.

A challenge in open-learning is doing significant assessment on authentic tasks (like grading programs) and to respond to the variable background knowledge in the cohort.  What will students do who lack the appropriate math for the Stanford machine learning course?  How will they know that they lack the math background?  Where will the teacher send them?

Is this really a Stanford-quality AI course?  Or is it more like outreach, community education, or adult ed?  The latter is fine, and it’s great to include graded quizzes in an open course.  It’s a step forward. But it’s not really a Stanford AI course, then. It’s an experiment in educational technology and distance education.

It’s an important distinction, between an experiment and a class.  The NYTimes piece quotes Dr. Ng saying, “I personally would like to see the equivalent of a Stanford computer science degree on the Web.”  I’d like to see that, too.  My guess is that Stanford wouldn’t actually count any of these courses toward any degree.  They’re not really undergraduate classes yet.

The Stanford scientists said they were focused on going beyond early Internet education efforts, which frequently involved uploading online videos of lectures given by professors and did little to motivate students to do the coursework required to master subjects.

The three online courses, which will employ both streaming Internet video and interactive technologies for quizzes and grading, have in the past been taught to smaller groups of Stanford students in campus lecture halls. Last year, for example, Introduction to Artificial Intelligence drew 177 students.

The two additional courses will be an introductory course on database software, taught by Jennifer Widom, chairwoman of the computer science department, and an introduction to machine learning, taught by Andrew Ng.

via Virtual and Artificial, but 58,000 Want Course –

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

App Inventor goes to MIT Some Media Computation Teaching Resources

15 Comments Add your own

  • 1. Alfred Thompson  |  August 18, 2011 at 10:41 am

    Additionally I wonder how many will actually complete the course. Online programs traditionally have high rates of non completion. With people not having money invested in this one I can see that rate being pretty high once actual work begins. Even, as you point out, with the work not being as difficult as a traditional course.

  • 2. Alan Kay  |  August 18, 2011 at 10:46 am

    Hi Mark

    I think your observation is correct.

    As far as I can tell (at some point I will email Peter and ask) why they didn’t take the rather easy route of making Javascript into an interactive AI language. Alex Warth did several of these a few years ago using his OMeta translator and a little work to make the equivalent of a Smalltalk workspace in a webpage.

    A few years earlier Viewpoints did such an interactive web page for LOGO just to show how easy it was, and to try to get the Wikipedia people out of their blindness.

    These don’t solve the assessment problems, but they could provide readily accessible programming environments for the examples in Peter’s book.

    (And since I don’t think that S expression Lisp is a very good vehicle for learning these powerful ideas in both programming and heuristic reasoning — especially for the larger audience — this route also provides a way to make a better vehicle for this purpose.)



  • 3. gasstationwithoutpumps  |  August 18, 2011 at 11:57 am

    I believe that the courses are real intro courses in AI and machine learning, based on the syllabi on their web pages. The machine-learning course, for example, does require linear algebra, basic probability, and some programming as prerequisites—exactly what I’d expect for an undergrad machine learning course.

    I blogged about the courses also ( ), but my assumptions were different. I didn’t think that the courses were to be greatly watered down, as you assume, but that fewer than 10% of the people who sign up will complete the courses. Stanford is not giving credit for the on-line courses, but the same lectures attended in person do get credit.

    I do wonder about programming assignments, though, as those are essential for a CS course at the proposed level, and I don’t see how meaningful feedback could be given in a huge class. I have seen CS classes in which programs are just checked automatically for producing the correct output for a given set of inputs, with no one reading the code to see if it is well-written. (This explains, in part, the absolutely abysmal programming I’ve seen from some grad students—no one had ever looked at their code before.) I suppose that they could have set up the machine learning and AI courses with this sort of automatic grading even of programming assignments.

  • […] Guzdial has commented on the Stanford AI course on his Computing Education Blog: Stanford on-line AI course draws 58,000 — but is it real?. He contends that the courses are not going to be real courses—that they will be watered-down […]

  • 5. Lisa Rubini-LaForest (@RubiniLaForest)  |  August 18, 2011 at 6:22 pm

    I am viewing it as a non-major course. That doesn’t mean it’s “not real” but it does mean it’s more of a survey course that will focus on theory and the applications of AI than practice of creating AIs.

    Looking at the course syllabus:

    *The (in person) AI course does not require a CS prereq (“Prerequisites A solid understanding of probability and linear algebra will be required.”)

    * The (in person) ML course doesn’t ask for much more (“Knowledge of basic computer science principles and skills, at a level sufficient to write a reasonably non-trivial computer program.
    Familiarity with the basic probability theory. (CS109 or Stat116 is sufficient but not necessary.)
    Familiarity with the basic linear algebra (any one of Math 51, Math 103, Math 113, or CS 205 would be much more than necessary.)”)

    Even the course numbers (CS221 and CS 229) imply this is probably a second year course at Stanford.

    I actually think this is a good idea. While someone interested in being able to implement systems using AI or ML techniques will be more interested in taking a more traditional AI course that is at the fourth year (or grad school) level, there is nothing wrong with attracting a wider base of students to CS by offering them material they find highly interesting. Reminds me of the two “popular science” courses my alma mater offered to non-science majors at the second year level.

    • 6. Jeff Forbes  |  August 19, 2011 at 11:33 am

      I wouldn’t view this course as a non-major course. 200 level courses at Stanford imply advanced undergraduate/beginning graduate not second year. If they follow the syllabus from previous semesters, then this course will certainly be at the depth one would expect from an AI course for undergraduates.

      The Russell & Norvig text (used at most top schools including Georgia Tech) is a challenge for most undergrads, so a course that can distill most of the important concepts and techniques from the book for a general audience would be a real achievement. The fact that they were able to attract 200 students to CS 221 last year is pretty amazing.

      Andrew Ng’s Machine Learning course is a really challenging undergraduate course. The reason why it doesn’t require much CS background is because it focuses far more on algorithms and assumes that students will be able to pick up the necessary programming required.

      These courses appear to be Stanford courses. Stanford undergrads and grad students who sign up for CS 221 will listen to the same lectures as students online. The concern about whether students will complete this course is certainly valid. As I write this message, 107,418 students have signed up for the course, but they haven’t committed to do anything other than typing their name and email address into a form.

      • 7. Mark Guzdial  |  August 22, 2011 at 1:17 pm

        My argument is about learning outcomes and faculty expectations. Learning arises from the activity of students. If students aren’t programming, they aren’t learning the same things as those students who do. “Covering” material is not the same as learning. Faculty would typically expect students to program in an AI class. Therefore, my prediction is that few faculty would recognize that all those students who complete this course (receive their certificate) will have completed an undergraduate-level AI class. The course number does not predict learning outcomes.

        • 8. gasstationwithoutpumps  |  August 22, 2011 at 3:15 pm

          I agree that without programming, few CS professors would see it as a real AI course. Since I did not see how to scale grading programs to a large class, I was dubious about that part also.
          One commenter on my blog suggested that Stanford might use “their new CodeScore system to analyze code for correctness, completeness, style, extensibility, etc. …”

        • 9. Jeff Forbes  |  August 22, 2011 at 3:35 pm

          Agreed. What students do is far more important than what a course or instructor covers. I was responding to Lisa Rubini-LaForest’s comment about the course syllabus and prerequisites.

          While the Times article highlights the online quizzes, since that’s presumably the novel part of the online version, it doesn’t say that students won’t also be doing all of the traditional problem sets and programming assignments. I signed up for the course, so we’ll see.

  • 10. John "Z-Bo" Zabroski  |  August 20, 2011 at 7:15 pm

    Courses are always a matter of what the student puts into the course in order to get out.

    Teachers tend to overestimate their contribution, sometimes to the point of negative effects.

  • 11. Fred Martin  |  August 21, 2011 at 9:45 am

    In the article published by the CHE, Thrun is described as saying that the “purpose behind grading the online students is to encourage them to work as hard as the Stanford students to meet deadlines and watch lectures.” He also says students should plan to spend 10 hours per week on the course (

    Regarding level, the online course’s web page points directly to the regular Stanford AI course for details ( Their regular course describes itself as “tailored towards advanced undergraduate or early graduate students, new to Artificial Intelligence, who wish to learn about the excitement in the field.”

    I set up two directed study sections for UMass Lowell students (one undergrad, one grad) so that our students who want to partake can do so for real credit. I’ll be meeting with them weekly — not as much face time as if I were teaching the class myself, but enough (I hope) to support their learning, and be able to connect my cohort together.

    Also, I am requiring that my students do an original implementation project with accompanying literature review, public demonstration/poster session, and write-up. Last year I taught AI and my students did some nice projects ( I agree that students should write code in an AI course 🙂

    The online class describes itself as a “a bold experiment.” I hope it works and I’m looking forward to experiencing it directly.

    • 12. Alan Kay  |  August 21, 2011 at 9:53 am

      Hi Fred

      This sounds good. Please keep us all posted on how things go at Lowell. With the kind of support you will be giving, this should be a terrific learning experience for the students.



  • 13. Stanford – More Free Online Courses «  |  August 24, 2011 at 3:36 pm

    […] Stanford on-line AI course draws 58,000 – but is it real? ( […]

  • […] that learning.  That’s the contribution that I see the Stanford AI class making – doing assessment, at least in the form of quizzes. And the education could be far cheaper, because there would be no expensive instructor and […]

  • […] interesting take, from the NYTimes, on how Stanford’s president sees the on-line Stanford AI class experiment.  The virtual campus is for “specialized programs,” for going beyond the undergraduate […]


Leave a Reply

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

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

Twitter picture

You are commenting using your Twitter 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 9,052 other followers


Recent Posts

Blog Stats

  • 2,030,788 hits
August 2011

CS Teaching Tips

%d bloggers like this: