I'm here to tell you about MOOCulus. Which is really a part of course that me and

my team launched on Coursera about a month ago.

The course is Calculus One. We have the requisite lecture videos that

you might come to expect, we have a free open source textbook that you can download, we

have weekly quizzes. The weekly quizzes have 10 questions per quiz,

which raises a question about questions, how many questions should there be on this quiz?

And the problem is that depends on the student. Some students should have more questions on

their quiz than other students. So to make this possible we have developed

another website. mooculus.osu.edu. Which might leave you wondering, what is MOOCulus.

MOOCulus is not Calculus, it is really an online homework system that we have built,

but unlike other online homework systems, MOOCulus comes backed with a hidden Markov

model. It is also Open Source, you can download it

right now on GitHub and install it on your own computers and if you were to do that this

is what you might see. You might see an exercise computing the derivative

of a polynomial, and there is a polynomial there.

And in order to facilitate you know questions about randomly generated polynomials we have

built a computer algebra system in JavaScript. We have also got a hint system, there is a

button there that you can click on to get help and if you click that button you will

start being lead to the problem. It will explain how you do the problem.

We have this progress bar, this is the hidden Markov model.

The problem is that student understanding is invisible.

I'm fairly certain that attaching the brain scanners that I would like to put on my students

is a FERPA violation, what is visible is the hints and answers that they submit, whether

they are correct or incorrect. The hidden Markov model takes that visible

data and produces data and estimate of student understanding.

So the students get a page that looks like this with green bars that filled on tasks

that we think that they understand. Yellow bars that they should do more work

on. And these red bars that they seem quite confused.

What is the real benefit to something like this?

Well part of it is that you might think that this is cheaper.

Now we don't even need TAs, but it is more than that.

It's really supposed to be better. We have already commoditized education right;

this is not the world's educational experience. It is a lot better if we get more people doing

mathematics. That is a better experience for the student,

I would much rather that they would just do more problems.

So if we are truly trying to personalize their education, so that leads to the question.

If I want this to be about 1 on 1 education what is so massive about this experience.

People have this mistaken idea that the M in MOOC stands for huge enrollment numbers,

but I guess it does, but the enrollment isn't really the big deal the point is that is only

a means to an end and the end Data! A lot of data.

In the last ten weeks, 10 person years have been sacrificed to MOOCulus.

And people have submitted two million correct answers on our system and this lets us produce

histograms and covariant matrixes where you can kind of see certain problems, if you interpret

this appropriately, are really more important in some sense or better predictors of overall

performance than other problems. That is really interesting data.

So what are we going to do with all of this data?

The point is that we can now start to try to evolve better classes.

This stuff is providing a fitness function that tells us when education is working.

The problem is that education also requires some sort of heritable variation and that

is not something that I can get with a regular, but a with a MOOC which fixes the form of

the interactive experience, I can now have something that is evolvable.

I fix the form of the classroom, which means that it is the sort of thing that you know

that evolution can act on. If this doesn't make a lot of sense here is

a soft of analogy. Here is some random walk and you can do lots

and lots of random walks, and my claim is that teaching is lot like this random walk.

Every year I'm slated to teach another course which is very similar course that I taught

last year and I do it slightly differently for no go reason.

And contrast this with if I actually an honest feedback mechanism, a very small amount of

feedback makes me get better. You know. And the point here is not emphasize

that MOOCS are massive or that they are online, the point is that MOOCs are for data-driven

feedback, which actually might improve education in the long run.

So we are really are at some fort in the road, one of those is that MOOCs.

MOOCs might not be great right now, but they are rigged to get better, because they have

got this feedback mechanism in place. The question is where or not the rest of the

university is going to get on board with that. And if you think about 50 years from now,

my grandchildren will have some very exciting choice.

They will either be taught by a human being or they will be taught by a Cylons.

And the Cylons will have some sort of hidden Markov model and the best cognitive science

available to really understand how to teach effectively.

And I much rather, that my great-grandchildren be taught by the Cylons.

So, thank you.