Arnold Kling  

Education and Disruption

August 1: Freedom for Western... Would Arnold Support an RFPB?...

Benjamin Lima writes,

With MOOCs, now anyone in the world with an internet connection can download and watch lectures from eminent experts at top universities, for free, and hundreds of thousands have done so. This is indeed a huge leap forward in the area of knowledge transfer. But the equivalent leap in the area of evaluation and feedback has not yet taken place.

Pointer from Tyler Cowen. Remember that I say that teaching equals feedback. In fact, I agreed with so much of Lima's long essay that it was difficult to excerpt. Read the whole thing.

I went to Udacity and spent about an hour in Sebastian Thrun's intro statistics course. My impression is that it is way too difficult for the typical student in my high school AP stats course. In eleven years of teaching, I think I have encountered only about 7 or 8 students who could have followed Thrun's course.*

That said, one way to think about what Thrun is doing is that he is disintermediating the Stanford admissions process. Instead of having the admissions folks filter potential students, Thrun puts his course out there and students sort themselves into ones who can handle it and ones who cannot.

Lima puts a lot of weight on the campus experience, including interacting with other students in extracurricular activities and enjoying campus amenities. But, as he points out, colleges seem to be competing extra hard in these areas, to the point where marginal costs are probably getting to be high relative to marginal benefits. One scenario he offers is one in which the non-top-tier schools stop trying to compete on the basis of campus experience.

I am not convinced that feedback is necessarily tied to geography. Let's stipulate that a human can provide more useful feedback than a computer on a student essay. I see no reason why the human has to be someone who resides on the same campus as the student.

If we wiped out everyone's memory of how higher education is organized and started from scratch, would the institution of a campus, which has medieval origins, still emerge? It obviously is not needed for content delivery. I do not think it is needed for feedback. I think there are more cost-effective ways to obtain consumption amenities. My guess is that the value, if any, is in the informal, serendipitous interactions, including student-professor conversations. Just as a city can facilitate informal knowledge transfer, a campus might serve the same function.

Related: Terry M. Moe writes (in a "retrospective" from the year 2030)

technology was also breaking down the geographic basis of schooling because students and (many) teachers--now meeting in cyberspace--no longer needed to be in the same physical location.

Somewhat less related: Kevin Carey on "credit for life experience." He refers to something called the Open Badges movement. I wonder what Bryan thinks about it. Are all alternative signals inherently inferior, because they signal nonconformity, or is there a point at which alternative signals reach critical mass and no longer have adverse connotations?

*As an aside, I think that Thrun's opening teaser is a swindle. He begins with a "proof" that, in expectation, someone else is more popular than you are. Go through the following logic.

1. There are two types of people. Type A has many friends and type B has just a few friends. Half of people are A and half of people are B.

2. Therefore, you have a 50-50 chance of being an A or a B.

3. If you pick one of your friends at random, the chances are higher that this friend is an A than a B (the intuition may be clearer if you substitute "sex partner" for "friend" And, as Thrun points out, the intuition is even clearer if you assume that B's have zero friends).

4. Therefore, a random friend is likely to have more friends than you have.

Thrun stops there. I actually think this is a swindle. Suppose that you and I are friends. If we follow the logic from my point of view, you are more likely to be popular than I am. but if we follow the logic from your point of view, I am more likely to be popular than you are!

What is odd here is that Thrun is saying that the fact that you and I are friends tells me something about you but not about me. That is, it now is more likely that you are popular than that you are not. That is fine. But we continue to pretend that we have no information about whether or not I am popular. I think that produces a result that, in my view, makes no sense. In other words, I think that (2) is a swindle in this context. I am what I am--an A or a B. Saying that I have a 50-50 chance of being one or the other is, in this instance, an abuse of the language of probability.

If the observable fact is that you and I are friends, then we can draw just as many inferences about my popularity as we can about yours.

Comments and Sharing

COMMENTS (11 to date)
allen writes:

With regard to feedback, that's always been part of the cost equation of eduction and has always been shaped by those considerations.

Testing - feedback - is also a cost to the education institution so the less of it the better so long as such value as there is in testing to the school isn't impacted.

What that's boiled down to is a relatively few tests taken at relatively long intervals for high stakes. But that's at the higher ed level. In K-12 testing has no direct value to the school so the tests are entirely for the benefit of the student. The SAT, for instance.

Schools don't aggregate and extract information from those tests because there's not much reason to do so.

But testing doesn't have to take six hours once a year or some such and neither does a kid's future have to hinge on that single test. Khanacademy shows how that works using a dictum from the technology sector - test early and test often.

Most mini-lectures are linked to small tests to determine absorption so test results are immediate, fine-grained and clearly linked to a given lecture.

That approach is sometimes approximated in classroom instruction with instructor-generated snap quizzes. The problem with that approach is that the instructor's time is taken up preparing, grading and analyzing the snap quiz so it's always a balancing act between time spent on lecture and time spent determining whether the lecture's effective.

In higher education the snap quizzes have value to the institution in that they may help identify the students most likely to excel sufficiently to add value to the institution but in K-12 it's largely a matter of the instructor's pride that motivates the use of snap quizzes.

But technology is altering the cost equation and schools that embrace those changes will prosper at the expense of those that don't meaning that what's now allowed by the technology will soon be required by the customer. That applies in both higher education and K-12.

Steven Sailer writes:

I probably have 300 hours of Teaching Company courses sitting around the house. I've probably listened to or watched about 120 hours. They are good stuff, but they haven't revolutionized my life and I doubt if putting them on the Internet would revolutionize the world.

In short, the world has had distance learning for well over a century (e.g., "correspondence courses"). During that period, famous universities have only grown in prestige and wealth. Why assume that the Internet will suddenly change that trend, especially because the Internet, so far, has only increased demand for admissions to elite colleges?

Mike C. writes:

Let's try a small scale example of the friend thing:

6 people, 3 unpopular people that each have one friend and 3 popular people that each have 3 friends.

There are 6 total friendships (3*1 + 3*3)/2 out of the set of 15 possible friendships.

The friendships can be either between two unpopular people, two popular people, or one of each. There are 3 possible friendships between two popular people, 3 between two unpopular people, and 9 with one popular and one unpopular.

How many ways can we delete 9 of these while maintaining our original constraints? In this case, all three of the popular people are friends with each other and each of the unpopular people is friends with a different popular person.

If you are an unpopular person, there is a 100% chance that your friend is more popular than you are.

If you are a popular person, there is a 50% chance that a random friend is less popular than you, and a 50% chance that they are equally as popular.

So all in all, if we choose a random person and then a random friend, there is a 50% chance that the friend is more popular, a 25% chance that they are equally as popular, and a 25% chance that they are less popular.

Just for fun lets try a different configuration, where 6 people have 5,4,3,3,2,and 1 friends. In this case if we select a random person, then a random friend, there is a 6% chance that they will be equally popular, 29% chance that the friend will be less popular, and a 65% chance that the friend will be more popular.

Doesn't really sound like a swindle to me. The reason we can draw more inferences about the friend than we can about you is that your friend was selected by a different process than you were. The first person is chosen after one random selection, but the second person is selected from that persons friends at random. But popular people are over-represented in friend groups, so we have more information after the second selection.

Michael Rulle writes:

Agree with your characterization of education.

One interesting question: why has this system continued to expand? The college experience is enjoyed by college age people. I believe it is largely a consumption item wrapped around an opportunity to learn, but not a necessity. One can get a degree and not really learn all that much.

Yet, even now there are "arbitrages". One could always go to community college for two years and transfer to excellent reputation schools for two years. This has a two fold advantage. If you are not college material, you will not make the expensive second two years. If you are college material, you basically get half price education.

It is also clear that integrating online learning, even in a campus setting, would lower the cost of education and improve it for those interested in learning. (lets assume the cost of living is "a wash" versus no college). But we have a significant Collaberative effort between higher education and the Government keeping the current high cost system going with straight cash payments going to the schools and paid back by either students or taxpayers.

The system is transparently corrupt. The last argument of a corrupt system is "signaling".

Ken writes:

Lima's article is the best I've read on the subject.

I took the Stanford online Machine Learning course last fall. I hadn't taken Linear Algebra in 30 years and was amazed that Professor Ng's course brought me up to speed so quickly with what I needed to know as background for the course.

It's interesting that Arnold says Udacity's Stats course is too difficult. I'm currently taking the course and I find it to be easier than the Stanford's AI course that over 30,000 students successfully completed. That said, I think Arnold is has a good point about the course filtering students. Which means the course found tens of thousands of very bright students. Online could provide a better signal than the current paradigm in some cases.

I'm not sure that that the current system for developing credentials is better than the MOOC system coming immediately down the pipeline. Employers may need to do a better job assessing candidates but, that is already the case.

How a new educational paradigm emerges will be highly influenced by employers.

prof.g. writes:

really? you think this statistics course is too difficult? every college freshman i've talked to who has taken this has said it is insultingly easy.

Brock writes:

Anyone who finds the Udacity statistics course too difficult can take a refresher on Khan Academy.

I actually think that Khan Academy is a better model for the future than Udacity or Coursera. I understand why the University system has broken up learning into semester-long coursework, but they did that because of logistical and overhead limitations, not because learning is best done that way. Anything that can be learned in a semester can be broken down into much smaller, bite-size modules. I'm sure Professor Kling could imagine taking any of his courses and making 20+ mini-lessons out of it.

Teaching is feedback, and the more targeted the feedback, the better.


I think the campus will continue to exist. The main reason for this is the students' benefit of being around other people who are also learning. Being around people who are working or goofing off all the time creates social pressure to blow off homework.

Ben Lima writes:

Dr. Kling, thank you very much for the link and kind comments. I have learned a lot from your blog.

My main point about feedback would not be that it is geographically constrained, but that it is time and labor intensive, needs to be done by experts, and hence can't be cheaply scaled up to a 10,000-student course, the way that lectures (i.e. knowledge transfer) can be. (At least, feedback on upper-level work, as opposed to basic feedback on grammar, arithmetic, etc.)

Geography would, however, be a constraint for the kind of learning (again, upper-level student learning) that needs physical resources such as laboratories and libraries (and perhaps cafeterias with brilliant lunchtime conversation partners, i.e. the "informal serendipitous interactions" you mention) to take place.

Extra-curriculars, I argue, both develop and signal valuable non-academic forms of human capital like leadership, teamwork, social skills, etc.

Vs. amenities which are more about consumption. Anecdotally they say many enrollment decisions are driven by amenities such as good food, good dorms, a good gym, etc. Agreed that returns are diminishing.

Mr. Sailer, thanks for your comment as well. The implication is that these Teaching Company courses were mainly about knowledge transfer, and that they did not in themselves provide a valuable boost to your human capital in the form of writing skills, thinking skills, etc.

I would say that the main difference between online learning and old-fashioned correspondence courses is the same as the difference between books, newspapers, LP records, etc., and their online equivalents, i.e. super-efficient, very low cost reproduction and distribution -- which is a pretty big deal, right?

TravisA writes:

Thrun is right:

TravisA writes:

The easiest way to see it is through an example:

Suppose that the average number of friends a person of type A is 50. The average number of friends for a person of type B is 10. Thus the average number of friends of a randomly chosen person is:

.5(50) + .5(10) = 30

Now randomly choose a friend of one of the first random chosen people. That person has a greater than 50% chance of being type A. Suppose that chance is 80%. Then the average number of friends that a friend has is:

[.80(50) + .20(10)]*.5 + [.80(50) + .20(10)]*.5

= 45.

(the .5 is the probability of randomly choosing a Type A person initially).

Thus the average number of friends of a friend will always be greater than the average number of friends of a randomly chosen person if the probability that a friend is of Type A is greater than 50%.

Dan Weber writes:

The footnote about friendship reminds me of the Two Envelope Problem.

Say there are two envelopes, one with twice the money of the other. You open one and find $100. You can keep it, or swap with the other envelope. So you risk $50 to gain $100, so you switch. Of course, the logic is to always switch, which doesn't make sense.

It's a fun thing to work out.

Comments for this entry have been closed
Return to top