Thrun/Norvig/Stanford introduction to artificial intelligence - just short of 90,000 enrolments in 2 weeks

(Other posts tagged ai-course.)

Note - 20/8/2011. In the small font below I am unsystematically tracking the growth in sign-ups for the three Stanford course.

Two weeks ago I wrote something about (and signed up for) the Stanford University online introduction to artificial intelligence, to be taught by Peter Norvig and Sebastian Thrun. (Stanford's Introduction to Databases [201108200619 - 23,671. 201108242232 - 35,321. 201108261846 - 37,419.] and Introduction to Machine Learning [201108200619 - 25,935. 201108242233 - 35,685. 201108261846 - 37,636.] are also available online this autumn.

Within 2 weeks - after extensive press coverage - just short of 90,000 people worldwide had enrolled on the course. [24 hours later the number enrolled topped 100,000. 201108190712 - 103,290. 201108200619 - 110,775. 201108242227 - 128,477. 201108261846 - 132,469.]

This means that the amount of data that could be collected and analysed about learner demographics, learner-behaviour and learner-performance, and about the behaviour of the supporting infrastructure is going to be large, even if only a fraction of those who enrol actually get involved in the course.

Alongside this, getting videos and tests served to users smoothly will be a technical challenge, depending on the way the course is designed and on how users are distributed geographically, especially if the number of people accepted onto the course increases substantially. (Has delivery of this kind of course at this kind of scale been tried before?)

So, Norvig and Thrun are taking a big, interesting, and quite a brave personal risk; nor is the venture without risk for Stanford University.

I would be interested in other people's questions about the course, and please feel free to add to this list below by commenting.

1. Did you plan for 10^3, 10^4, 10^5 or 10^6 students and if your plans were for much lower numbers than you have got, have you needed to adapt them and if yes, how, as numbers have risen?
2. On whose infrastructure will the course run and how is it intended to scale up as student numbers increase (assuming that it needs to)?
3. What if any demographic data will be collected about students?
4. Any reason for not having some lightweight pre-enrolment testing of potential students to help them decide if they could cope with the course?
5. What plans have you to analyse data about learner performance and to draw meaning from it?
6. What sorts of representations of learner activity and performance will be available to you as the course progresses?
7. What if any VLE or equivalent will be used to host the course materials and activities, or pointers to them?
8. How big is the team behind the course and with what roles?
9. Have all the course materials and activities all been prepared in advance or will some of them be being created as the course progresses, possibly influenced by judgements made about how the course is going?

Comments

[Stuart Sutherland]

Thanks for this post and these valuable and stimulating questions Seb. My questions are all variations on the one theme: which is about how this course might harness and benefit from peer-to-peer activity within that huge body of motivated learners:

What scope might there be for learners to learn from each other or to form special interest or research groups on some of the topics within the course?

Given the volume of learners willing to participate, how might we harness the sheer volume of motivation to produce meaningful activities or research or tangible outcomes that other learners can access and benefit from?

Given the number of learners who have shown an interest in the course, what scope, if any, might there be for learner-led group work amongst manageable groups of learners?

[Bob Harrison]

Thanks for this Seb, I am on the campus at Stanford and have spoken to some faculty. They are excited by this project and although it was slow to pick up interest once it went viral it has surprised everyone. This has enormous implications for learning in general and HE in particular. Readers will hopefully recall my previous visits to Palo Alto to study the Stanford Virtual High School and the iStanford project. Interestingly the inspiration for these short courses came from the Khan MIT videos and Stanford will award a "certificate of accomplishment". I love forward to seeing yours at ALT-C :)

[Pabitra]

Could answers to some of the assignments end up as new heuristics? When such large numbers across the world are involved, such results are within the realms of possibility.

[John Vornle]

I bought the book. So far, the text is boring and repetitive and is only defining the nomenclature associated with current discussions as the egos of the founders of the "new" field are choosing to define it.

For the "create a yardstick for myself" experience, I'm hoping that I will be able to participate fully. I will probably read their 1,000 page text book. That part is easy. Unfortunately, depending on what it says, it will determine whether it is worthwhile to even bother to participate in the course.

I found a disconnect between the initial brouhaha that said that treatment was "exactly the same as a participating Stanford students" and the later paperwork, issued on Stanford stationary, stating the legal interaction. It would seem that Stanford administration has stepped in and forced changes to protect their "unique education" experience.

Interesting points that raise themselves include:
1) Accreditation. If "treatment" is the same, than it should only be a matter of time before accrediting organizations issue "degree certificates" to those who receive "completion certificates" with grades, according to whatever curriculum the accreditation authority requires for a degree. This is a positive development.

2) As I see it, the business model of revenues for text books sold will quickly dominate salaries paid to professors. This suggests that professors will soon start to pay universities to be able to teach there and for permission to interact with their student bodies. I imagine this as a welcome change but one that will initially generate a lot of resistance.

3) Professors will be forced to increase their communication skills and "content relevance" so that they keep themselves interesting, relevant, and meaningful to the exponential increase in shared global information. This is welcome.

4) University students and their bell curve performance will be measured against the performance of a much broader audience which will either be positive or negative for any school, professor, syllabus, or student. This will be a welcome change to the status quo.

We live in interesting times.

[Seb Schmoller]

1. Rather by chance I used WebCite to take snapshots of the course home page and you can see how the course description changed between 3 August and 18 August.

2. I do not agree with John Vornle that the book is boring and repetitive: interesting, thorough, and hard work would be more my description.......

[Dick Moore]

Such an interesting experiment and one that can generate real meaningful data.

To pick up on Seb's point 6:

1. How will the infrastructure and learning materials be "instrumented", and will the data be made public?

2. Is it only the assessment elements that will use transactional interfaces such as SCORM or will learning materials also include data stores and adaptive paths.

Other areas for exploration:

a) Traditional assessment and review techniques with numbers like this are just not possible, being much more akin to national public examinations which measure the output of the educational process.

b) How, if at all, will the course design bring along and support those students that are struggling?

Such an exciting experiment, 150,000+ people who will have a better grip on AI - if they turn into real rather than simply curious learners :)

[C Bell]

Hi,

I have zero background in AI or anything remotely related.

OTOH, I read a lot, have various degrees in other fields, learn quickly, and am interested. I've had basic science courses as part of a liberal arts background.

Do you think I have any chance of keeping up with this class, even if I want to, and even if I work hard, with no background in any related subject - and just my bare curiosity?

Thanks!

===

The short answer is that I am not sure, and that the course might prove very challenging for you, though I am sure you would learn a lot.

I am slowly working through "Artificial Intelligence, a Modern Approach", and though its mathematics is, so far, not too bad, it uses abstract notation right from the start and is written as if the reader has a fair understanding of programming, which I do not.

The most sensible approach might be to enrol on the advanced track being ready quickly to "drop" to the basic track if it turns out that the assignments are too challenging.

Seb