acm - an acm publication

Articles

An Interview with F-H Hsu
Chess, China, and Education

Ubiquity, Volume 2005 Issue July | BY Ubiquity staff 

|

Full citation in the ACM Digital Library

Feng-Hsiung Hsu, whose book "Behind Deep Blue" told the story of world chess champion Garry Kasparov was defeated by the IBM computer known as Deep Blue, is now a senior manager and researcher at Microsoft Research Asia.


UBIQUITY: Your friends call you CB?

HSU: Crazy Bird. Just some nickname from high school. The first character of my given name, Feng, has the same sound as 'crazy' in Chinese. When I was in high school the kids somehow got the idea that I was in fact a little crazy…

UBIQUITY: What were you like as a student?

HSU: Let's see. I was an OK student early on. And then at some point I realized if I applied myself, I could really do things, and that's when I got good. This happened when I was in primary school.

UBIQUITY: How did you recognize that you were good?

HSU: We were in a painting class, and for some unknown reason I decided to work harder on that particular day, and the teacher gave me a good grade. So I figured out that if I applied myself I could really do things, and that's when I started changing my behavior.

UBIQUITY: Were you good in all subjects?

HSU: Well, in the ones I was interested in.

UBIQUITY: What about in mathematics?

HSU: I actually was OK in most stuff but the athletic things. And even there I think I got better as things went on, but that's a different story.

UBIQUITY: And so what kind of an education did you have after primary school?

HSU: I was born in Taiwan, and back in those days, they stipulated by law that you had to have nine years of education. So I went on to junior high, and in Taiwan you have an entrance exam for senior high school, which I took. And then after senior high school I went to college, and decided to study in the States. I went to Carnegie Mellon and got my Ph.D. and then joined IBM. When I left IBM I joined Compaq, which then got merged into HP, and I decided to move again, to Microsoft.

UBIQUITY: How long were you at IBM?

HSU: I was there for ten years, but the work on Deep Blue actually started when I was in graduate school — that's where I started the project.

UBIQUITY: When did you get interested in chess, and then computer chess?

HSU: I think I started playing chess when I was in primary school. I thought of it as just another game, and liked it the way kids always like to play games. But then when I was in college one day I bumped into a book in the library that was a classic for computer chess, called "Computer Skills in Men and Machines." A chapter in it describes the structure of the program called Chess 4.5, which was a very influential design and was pretty much the basic blueprint for all modern computer chess programs. But the main reason why I came to the States was to study how to design chips, and doing computer chess was actually the last thing on my mind. So it was just purely by accident that I decided to do chess. One day I just figured out, "Wait a second, I know how to do this" — so I decided to go for it.

UBIQUITY: Your Deep Blue chess strategy was a brute force strategy, is that right?

HSU: That was my initial starting point, after reading a paper by Ken Thompson that experimentally verified how you can increase program playing strength by improving computation speed. So we decided to push speed, which we knew how to do and was interesting by itself from a computer science point of view. Of course, when you compete against the world champion you realize you need more than just brute force, obviously.

UBIQUITY: Talk a bit about Garry Kasparov. You've written a whole book about the matches with him, but say something about him now and his recent retirement.

HSU: Yes, his retirement came as quite a surprise, because he's still so young, but I guess he wanted to retire when he was at the top. I think there's good feeling and bad feeling towards him. First, we are grateful that he was willing to play, and not all chess players would be willing to put their reputations on the line — that's human nature. His post-match behavior made Deep Blue's win far less satisfying, obviously. But that's life—you just deal with what you get.

UBIQUITY: How would you describe his behavior?

HSU: It was the behavior of somebody who can't accept losing. Psychologically he has problems balancing himself, and needs some kind of escape valve. Otherwise, it would be hard for him to accept himself. You have to remember that his loss to Deep Blue was his first ever chess match loss in his entire life, whether it was against humans or machines. But I guess he's happy where he is now.

UBIQUITY: He's returning to Russia, and going into politics?

HSU: He's going into politics. And there was some story about someone, who did not like his politics, beating him up with a chessboard. That's a messy business there.

UBIQUITY: Was there a suggestion that he was going to challenge Vladimir Putin?

HSU: I don't know whether he can do that, but he thinks he has the obligation to do something in that area.

UBIQUITY: So where is the state of art of computer chess now?

HSU: Well in terms of real computation algorithms and things like that, I don't think there's any major new development. There have been minor changes and evolution, but it's pretty much the same as when we left the field, except that the computers are faster. The average computer would be stronger than it was when we played Kasparov. Whether what you can buy today is good enough or not is unclear. But the results of recent matches are not conclusive, in the sense that if you look at the match results, yes, they tie with the world champion, but if you look at the results through the match, they were really doing poorly until the end when the human just collapsed. So were they really sufficient to beat the world champion? I don't know.

UBIQUITY: Well, now that your opponent Kasparov has retired from chess, have you also retired?

HSU: Well, we stop doing serious work on Deep Blue a few months after the 1997 match. We were preparing for a new match as any good sportsman would do. But Kasparov was making all sorts of accusations, and IBM decided not to go on with it. When I left IBM, I got sufficiently mad at Kasparov's accusations and licensed the Deep Blue chip technology just in case. I actually tried to set up a match with him at one point but it didn't work. Kasparov wasn't really interested in playing. [He was right to decline. I was planning to create something a hundred to a thousand faster and capable of beating him with a 100 to 1 time odd. Of course, without telling him first.] Anyway, Deep Blue is now in museums. Half of it was donated to the Smithsonian, and the other half is on loan to the Computer History Museum. I haven't really touched that field since the year 2000. I look at things occasionally, but I don't have the energy or the desire to do anything in that area. The way I feel about it is: I can do it but what's the point? If I do anything in that area it's not going to be in chess, it's going to be in some other games. And the big one would be the game of Go, but Go is too hard. The other one is Shogi, which is Japanese chess, but that's less interesting.

UBIQUITY: Go is harder than chess?

HSU: The strategy you use in chess can't be applied directly to Go — and nobody really know what Go strategy should be. But that doesn't mean it's harder. I don't know for sure because, if you think about it, the energy that's spent in computer chess began way back in the 1940s, whereas people didn't really get serious about Go until 1990 or so, when a Taiwanese entrepreneur donated something like $1 million in prize money to get people interested in doing it.

UBIQUITY: Do you have any interest in other kinds of games, like simulation games?

HSU: Well, games are not really where my interest lies. Today I'm mostly working on things that might be relevant to the general public, such as how can we improve the performance or the usability of computers and mobile devices.

UBIQUITY: So you're moving away from games completely?

HSU: I really moved away from games when I left IBM, and even when I was at IBM my interest in games was already reduced at that time. Doing games was for me just sort of an accident. It wasn't really my first choice. We just happened to see a way to do it, and it was within our reach, so we went for it. But that doesn't mean that we wanted to do that forever.

UBIQUITY: Tell us what you're doing at Microsoft.

HSU: I'm in charge of the Platform and Device Center at Microsoft Research Asia, trying to expand the reach of existing platforms that makes PCs more useful everywhere, or create new platforms possibly on new kinds of devices. At least, that is our charter. We do mostly hardware related projects, although more than half of the projects are really software projects, which include mobile system software and camera related software. In the camera related software area, we are mostly interested in the camera as a generic high-bandwidth input device of the physical world. We are also doing some pure hardware work. One of the stranger projects that we are going to start in the second half of the year involves creating a new type of 3D display that potentially can allow a theater-sized audience to experience something similar to a hologram, without wearing glasses. Some of the other things I'm working on, I can't really talk about yet.

UBIQUITY: You're not going to tell us any trade secrets today?

HSU: No trade secrets.

UBIQUITY: What has surprised you in the last 10 years or so?

HSU: The developments in semiconductors in the last 20 years have been staggering. When I was younger I had no idea that things were going to move that fast, so when it happened it was just staggering. I think that was the biggest surprise to me. Now every day we're looking at how Moore's Law can be extended.

UBIQUITY: And you don't see Moore's Law stopping soon?

HSU: I think most people will tell you it's going to stop within the next 10 to 20 years. There are going to be very interesting questions to ask at that point, and even now, while Moore's Law is still valid, there's also a lot of interesting questions to ask. And when processors get another 10 or another 100 times faster, how will we do things? We don't know yet. I think one big thing on people's minds is power consumption. With the way things have been going, the CPU is going to be hotter than the sun.

UBIQUITY: Was there some particular reason you moved from IBM to Compaq to Microsoft?

HSU: Part of the reason for the move was personal. I was interested in seeing what's happening in China, given the fact that that's where most of the growth is happening. And since I'm ethnically Chinese, it makes sense to know more about China as well. So that was part of the interest. I think there are other reasons why I decided I should leave. As I told you, before Microsoft I was merged into HP, and I believed HP was in potentially all sorts of trouble, even before Carly Fiorina was forced to leave. There were serious problems with the business model. The bulk of HP's profits come from ink and from the company's printer-related business, such as its inkjet cartridges. Six or seven years ago the Taiwanese were really looking to go into that field, because the patents had started to expire; once you realize that, you begin to see that things can go pretty bad really fast. Beside these personal reasons, one big attraction was Microsoft Research Asia itself. It is, as MIT Technology Review said in a cover article, the hottest computer science lab in the world. The youthful energy there was just fantastic. Family wise, the decision also worked out quite well. My wife was a venture capitalist on US assignment from Taiwan, and she always thought that she would go back to Asia to continue her career. The move gave her the chance.

UBIQUITY: How long have you been with Microsoft now?

HSU: About two years and two months.

UBIQUITY: In Beijing?

HSU: In Beijing, yes. And I expect to be here at least a few more years. At some point I probably will want to move back to the States. But not before I have better understanding of China, and it could be a long while.

UBIQUITY: How has it been a cultural shock to go back to Beijing?

HSU: Well, it wasn't a complete shock, because I had been in Beijing a couple of times before, back in 1997 and 1995, so I essentially knew what I was getting into. The air pollution is really bad. People smoke everywhere, smog lingers when there is no wind, et cetera. But there were actually some pleasant surprises, and it's much improved since let's say 1997. So it's better than I originally expected. Kai-Fu Lee was the first director here, and he tried to recruit me when they started the lab. At that time, I said there's no way I'm going there — I can visit, but there's no way I'm going to live there. And now I actually live there. There's a significant difference between then and now. But still some of the things, even now, take some time to get used to — like the smog in Beijing, which can get quite bad. Beijing has good days, but during the bad days you can't see more than 100 meters. There was one time it got so bad that when my wife and I went out to dinner it looked so smoggy we decided to wear breathing masks. Coming home, it took about maybe 15 minutes to walk to our apartment, and when we got there and took off our breathing masks they were yellow.

UBIQUITY: Well, that must be quite an interesting experience.

HSU: It doesn't happen every day — maybe once a month during winter time or something. There aren't that many days like that, but there are some, and you know that you should stay indoors if you can.

UBIQUITY: How does the work culture compare to that in the U.S.?

HSU: The two work cultures are dramatically different. When Microsoft chose a location in Beijing it made a really smart decision — it picked a spot within a 15-minute bike ride from the major university in Beijing, so that students could come in fairly easily. I think nowadays we have close to 180 full-time employees and something like 250 students year-round in the lab. So it is quite different from the typical industrial lab in the States, where you only have students coming in the summer, whereas here in our Beijing lab they're with us almost semi-permanently. So that's one big difference in terms of atmosphere. We love having young kids around — you feel differently, and that's quite obvious the moment you walk in. And also the way people do things here is quite dramatically different from what's happening in the States. In the States, if you're an industrial lab researcher you typically start out as a lone ranger: you start some project on your own and then you try to sweet-talk people to join your project. Whereas here, you have more people and lots of students you can access, so instead of recruiting other people in the lab to help you, you can recruit students to start your projects from blank pages. Potentially you can be up to scale much faster, and you can build a fairly large team. I think one manager here actually has something like 60 students working with him. Of course, he also has second-level people in-between, but he was able to develop a huge group within just a year doing things that way. So if you have some project that can benefit from a large amount of brain power, China is one place to do that.

UBIQUITY: Then if you had a brand new project that was essentially a clean slate and nobody knew anything about it, and you had to start from scratch, you'd be more likely to go to China than to the U.S.?

HSU: It depends of course on the nature of the project. If it is a company in a new field (which means there's no experts anyway), and you know you might need a lot of brain power, and it happens to be a project where you can parcel pieces of the project out to multiple people, I think China would be a good place. But if you're dealing with things that only a few people in the world know how to do, and you happen to have those people near you in the U.S. for instance, then you might be better being in the U.S. because then you have access to the right people where you can do things much faster. So it depends. Not all projects can be done best in China, but China does offer some opportunities that you don't have in the U.S.

UBIQUITY: What could the U.S. learn from the Chinese educational system — and what could the Chinese learn from the U.S. educational system?

HSU: That's a good question, but I don't have a ready answer. Let me first give my impression of the difference between students in China versus students in the U.S., particularly in the hardware area. In terms of hardware guys, I think the U.S. has more hands-on experience. In China, the students tend to work only on very small pieces of projects, and they don't have the big picture. They may not even know some of the basic things that you would know when you build the whole system by yourself, which is more typical for students in the U.S.

UBIQUITY: Why so?

HSU: What happened is that in China, especially in graduate school, the students are effectively like employees of a small start-up owned by the teacher. Professor salaries in China are actually quite low, so professors get the bulk of their earnings from projects. They got those projects from the university, from industry, from the government, whatever, and in order for them to get a project working, they use their students as (to put it bluntly) slave laborers to work on various pieces of the projects. So they partition a project and piece it out, and each student gets to do a small piece of the project, even Ph.D. students. So the students are fine if you just want them as some laborers doing things they already know how to do, but they're often inadequate as researchers. Typically, when we hire somebody, we have to retrain them because of that.

UBIQUITY: Why is that?

HSU: Because it's more likely to be the case that they can't think on their own initially. They tend to need to be told what to do, because that's how they work in the university. They haven't been trained to think independently and think about the bigger picture. But different universities in China can be quite different. When I visit various universities to give talks, I find really different behavior from the students. Some are just after you like crazy, and others don't want to say anything unless somebody else starts asking questions first — and then gradually they build up their own nerve. They're not as talkative as the students in the States.

UBIQUITY: Do you have any interest yourself in going back into academia at some point?

HSU: Not really. I'm actually a horrible teacher.

UBIQUITY: Are you really?

HSU: My wife would tell you that. I think my problem is that, when I was learning, everything was too easy for me. So sometimes I have no empathy with what people are suffering when they have problems with certain things. I hope I'm improving in that area, but I think being a teacher is not really my calling, and it's not something I'm really interested in doing. Of course even a research manager has to do some coaching — that's a necessary part of the job. But as for actually going back to university and teaching, no, that's not really what I'm interested in doing.

UBIQUITY: At Microsoft, do you supervise a big group?

HSU: My crew is really small right now. I have six full-time employees, including myself. Counting students, we have about 10 people total. It's really small compared to the guy with 60 students.

UBIQUITY: Do you play any chess?

HSU: I do play chess a little bit, but I'm not good at it. And I pretty much haven't played chess since we played Garry Kasparov. Well, my nephew does actually force me to play him every once in a while. I won't say who wins.

COMMENTS

POST A COMMENT
Leave this field empty