Peter Norvig [8.1.08]

What motivates people? What is it that people want to figure out, and when do they decide, "Hey, I've got to go find some information," and how can we get them to do that more, to find the information, and then use it in their life?

PETER NORVIG is Director of Research, Google Inc., Fellow of the AAAI and the ACM; Co-Author, Artificial Intelligence: A Modern Approach (the leading textbook in the field).

Peter Norvig's Edge Bio Page

[11:00 minutes]


[PETER NORVIG:] I used to spend a lot of time thinking about what's the right way to get information to people. They have a query, they come to Google, they type something in, and we have to find the right answer. That's a challenging problem. I spent many years thinking about that. Lately, I've been thinking more about a different problem, which is: Maybe we were cheating in a way, that when people come to Google, and they type in some words, and we have to find the right answer, it sounds like a hard problem, but really, the users did the hardest part of all, they came up with an interesting question that they wanted to know about.

That's the question I'm really interested in now: what motivates people? What is it that people want to figure out, and when do they decide, "Hey, I've got to go find some information," and how can we get them to do that more, to find the information, and then use it in their life? I came to this realization when we had Teach for America come visit, and a teacher told us, " they gave me a seventh grade math class, and they were two years behind level, and by the end of the year I had them all up to the proper grade level." That's an amazing achievement, and I was really impressed by that. But then I thought, wait a minute, I'm at this company and our motto, our mission statement is "Access to All the World's Information." And this great achievement that this teacher was able to do had nothing to do with information. They had a textbook her year, they had a textbook the year before. It wasn't that the textbook was any better. It didn't have to do with information, but it had to do with motivation and determination, and it had to do with this relationship, that the teacher said to the students, "Look, I'm going to be there for you. I'm going to teach you this, and you're going to work hard for me, and together we're going to do this." And it worked.

The teacher did have to know enough about the subject matter to get it across, but the main part was bringing that determination. Now I'm trying to understand that a littlie bit better. What motivates people? Why are they interested in something? If they're interested in this, what's the next step? On the one hand, can we personalize that to say, "I want to help you in your quest to do what you want to do with your life, and bring the right information to you, make it more personal." On the other hand, we also want to make it more social, because we know that it's hard to maintain, and work, and improve yourself on your own. It's easier when you have some motivation from other people that are doing it along with you. Trying to get that balance just right, of how do we do something that's just for you, but also do something where it's "I don't want to let my buddies down. We're doing this all together."

One way I came to all this was, in addition to my day job here at Google, I also have a job as a former academic, and I still have a textbook in artificial intelligence. Every few years it's time to revise that textbook and write a new edition. The last edition, we did a couple years ago, it struck me much more than the previous editions of how limiting it is to try to put everything down onto this dead tree form, in a number of ways.

One is, I'd be going along writing a paragraph, and say, "Ah, wait a minute. This isn't coming out very well. It's confusing. I'm doing a bad job. Let me throw this paragraph away and start over." Then I'd do another one. Now I've got two paragraphs, and I have to decide which one is better. I said, "This is crazy. Why should I be the one deciding which one is better? Because I'm just the author, I'm not the reader." The way we should really decide which one is better is show these two paragraphs to different readers and let them say, "this makes sense to me, this doesn't." Let them vote on it. Then we have an idea of what works and what doesn't work, and maybe we have a good model of the reader, we can say, "Hey, for you, you have a lot of mathematical background, you might like this type of explanation. For you, you have a different background, you might like a different explanation." We should be able to customize the book to the reader. You can't do that with paper, but you could do that online.

Then there are other things that I really missed in trying to get it done. For example, it's a technical book about artificial intelligence, where you see plots of data. I was making one three-dimensional plot of some data, and so what did I have to do? I had to write a program that ran a simulation that generated some data, and then took those data points into a 3D plotting routine, and then played around with it on my screen, and rotated it until I said, "Well, here's the one view that gives the best overall summary of what this 3D surface looks like." I learned an awful lot by doing that playing around, and manipulating, and so on.

But the students just see this one view. It's like they've got this little peephole, and they have to look through it, and they see one view of this mountainous surface, whereas I got to do sort of a flyover and see the whole thing. So why do we restrict students to do that? It's because that's what paper can do. But if we had the program that we used online for those students to play with, they'd gain a lot more. So I started thinking, how can we go from books to a more interactive experience?

I've done some teaching at Stanford as a visitor. It was my turn, along with my co-teacher, Sebastian Thrun, to teach the AI class, and we said let's do this: let's teach the Stanford class the way we've done before to the Stanford students, but let's simultaneously put this online and open it up so that anybody in the world can take it. And we did that. We wanted to do things a little bit differently from some of the classes that had been put online in the past. There's a long history, and lots of people had made a lot of good innovations, and we were just following up on that.

One of the things we wanted to do was say, when you take an hour-long lecture and squeeze it into a video, and students view it on their small screen, it's just a lot more boring than seeing things in real life. So we said we don't want to do that. We don't want to just record the classroom. Instead, we want to make materials that specifically design for this media, so that's what we did. We said, we're going to teach the class as always, but in addition to that we're going to make videos of small units, a couple minutes at a time. We learned from Khan Academy that a couple of minutes is better than an hour. We're going to intersperse that with these kinds of interactive environments where the students can interact. We can ask some questions. They can answer. They can play with something, see how it works, run it, get some responses, and have more back and forth rather than one-way lecture. We thought that was successful. We were able to do that. We were able to get a good experience for each individual student. We were able to put it online for 100,000 students to take, and we were able to have the group component. So they got that individual instruction, and then they also got the group component of this all sorts of discussion forms, in which they can participate with other students in the class.

Another thing we wanted to do was shift the emphasis from what the teacher does to what the student does. Herb Simon has this quote saying, "learning is something that's exclusively done by the student, and the only thing the teacher can do is influence the way in which the student does learning on their own." So it's not what do I say. It's what do you do? How do you interpret what I say? We wanted to acknowledge that, and we wanted to give a forum where the students could perform that learning on their own, where they were figuring things out, where they were making up models, where they were correcting the misconceptions that they had, rather than just having them listen to me and memorize it. There's a great experiment where passages from a textbook were given to students. They said, "Here's a new topic. Read this passage. Then we're going to ask you a bunch of multiple choice questions." They chose passages that were graded as being exemplary examples of very good writing. The students read the passage, they took the test, and they got a certain percentage right.

Then in the second group they took the same passage and crossed out one of the key sentences and gave that to the students, and then quizzed them on it. Turns out those students did better. What's happening is the first time around the passage would say, "Well, the first day, and that causes B and that causes C." The students would look at that and say, "Yeah. That makes a lot of sense. I think I've got it." But they didn't really engage with it fully. It came in, it made sense as they were listening to it, but they didn't make it their own.

The second group of students had said, "First, there's A and then there's C." The students would say, "Wait a minute, that doesn't make any sense. C doesn't follow from A. What's going on here? Maybe there's a B in the middle there. That would explain the whole thing." That way it was the students who built the model, rather than just listening to it, and that made it their own, and that enabled them to do better.

That's the type of experience we were trying to do, sort of a Socratic approach, where the students invent the subject matter on their own, due to our hints, and that works great if you've got Socrates in the room. But if you don't, and if you only have this electronic forum, you can try to get there but it's not always going to work because every student is different. So that's where we're relying on the discussion forums, where you go in the teachers and the TAs were part of that, but the most important part was the students themselves helping each other out.