One-on-one with Joe Hoane

Q: Looking at last year and looking at this year, if you were to compare where we've been and where we're going now… how would today's Deep Blue compare with last year's Deep Blue?

JOE HOANE: Well, the first thing is that the SP will be twice as fast, you know, one year goes by and the computer is twice as fast. And that's sort of just a fact of life in this way we live today. We're going to be using the new Part Two Super Chip, which has just started chipping. They're giving us a few -- they're letting us borrow a couple -- about thirty for the match. So, that's the first thing. What does that give you? Well, it's hard to quantify, but to use Murray's line, you don't sneeze at twice the speed.

One of the big things is that we had this hunk of technology that we brought to the table last year. And it was just a hunk of technology we had, you know, a lot of work we put in as a piece of chess software, a lot of work on parallelism, a lot of work just to bring this thing to the table – a tremendous effort just to create it. But it was just something that wasn't mature. So over the past year, we've had the chance to make it mature.

So, what does that give you? I think it gives you the world. You know, you take something and the difference between making it work and making it all that it can be is a long road. I guess I need an example to make that more interesting.

Well, talk about chess knowledge. That's one of the things. You have this computer that knows a few things about chess. You take how the computer plays chess -- it really doesn't need to know that much about chess – it's how we play chess with technology. We do computer science, we do algorithms, and we do parallel processing, and we use the best technology available, and so we created a big parallel system that plays chess. And the reason it plays chess well is because it computes quickly.

We do fast computing and therefore it plays chess well. But, course, you do have to understand the problem you're trying to solve and we're trying to, quote, "solve" chess. I don't mean we're trying to solve chess, but, you have to know something about chess to play chess well -- I think that's obvious.

There are topics in understanding chess and research topics [going out] that we've worked on this year. Because, you know, we've been using chess experts. You get actually almost more into, what I would call, the knowledge bottleneck. We brought something to the table that didn't have a lot of knowledge and we've been using experts to put more knowledge into it over the past year. It's actually one of the big things. It's not just to do this thing well – we had to do that – it was in order to make a bigger better challenge this year.

Q: It's one thing to create a bigger and better challenge, but what are the implications beyond that challenge, once Kasparov folds up his chessboard and you go back to Yorktown?

J.H.: There's actually a really big thing happening here. Besides just technology there's something to do with society here because we're crossing a threshold where we can solve problems better than any person dreamed of solving problems.

Maybe we don't play chess better than the world champion this year, but it's going to be this year or maybe it was last year and we just didn't happen to win – I mean, right now. No, really, you think that's a joke, but we could have won last year, it could've happened. And so, you know, sometime this year we will play chess with a computer better than the best human. I mean there's something going on there.

SOUND BITE: WAV(644k) | Bamba(33k)

Just recently, when I was trying to put together this talk for Illinois, I understood in my mind what it is: We're going to be able to do different things with computers tomorrow, literally next year, than we could do last year. And the difference is qualitative in the sense that we can now solve problems that people couldn't dream of solving. And I think, you know, yes, we've contributed to this by doing Big Blue. We've pulled in what you can do with a super-computer by about five years. I mean, that's actually one of the things that Deep Blue is good for.

We say: OK here's this problem. And in order to get a certain performance level on this problem on a general-purpose super-computer, well, it's gonna be five years, I mean, before you could get a billion positions per second on a super-computer without the techniques we used. But, I mean, so we contributed to the advancement of computing, but actually, I think the match -- Deep Blue, Kasparov -- I think that one of the big things is that it called people's attention to the fact that we are here. O.K.? Get the point?

If you go back to the HAL thing, 1968, 2001 came out and a lot of people were introduced to the idea that you could have a relationship with a computer. HAL in the movie had a personality and in1968, people started to realize that computers were getting interesting, and maybe we've reached another milestone where they're getting really interesting and they can solve really interesting problems that we couldn't otherwise solve.

Q: Along those lines, since you brought up HAL, and since you are an integral part of the development of Deep Blue, did you enjoy going back to Urbana for the HAL birthday celebration? You know HAL actually was a chess player, a very good one at that, and there are parallels that people are drawing to…

J.H.: Well, you know, Murray wrote a chapter about that…. What was the question?

Q: Did you enjoy going to Urbana for the HAL celebration? What were you thinking while you were there?

J.H.: Well, what I was thinking about that whole time was how people tend to anthropomorphize the computer. And it says more about our need to have a relationship than about the fact that there's something there to have a relationship with. And this is what I was thinking about the whole time.

You have this match, Deep Blue, Kasparov -- and people want to envision some sort of personality or some sort of human-like thing there. But there isn't, I mean, it's just a tool, it's just a thing that computes very fast and plays chess well. Deep Blue can compute a certain calculation so fast. It plays chess well, but there's no intelligence there, the way we would define intelligence.

That's what I was thinking about, going back to Illinois, thinking about HAL. Because HAL as a chess player is a human chess player, OK? He's intelligent, but you can trust that with Deep Blue, [the intelligence] is not there. Would Deep Blue beat "HAL"? Probably, I mean, yeah. Maybe. But, you know, it's certainly plausible to think that it would – at chess, I mean.

Q: As long as we're talking 2001, let's move forward five years: it's the year 2001. What does this all mean now to society? What are the applications that are going to be gained from this year's chess match in five years?

J.H.: Well, you know, there are two answers. One is a direct follow on; What can you do with the techniques, specific techniques, that we use in Deep Blue? And then the more general question is, what do you do with more computing? And I think the exciting thing is that we have more computing now and that there's going to be qualitatively different things you can do.

The best example that I can think of is -- we always come back to drug design for some reason, but it's a good example. I mean, let's say you bring in the cycle time for making some anticancer drug. You bring it in 10 years, from 20 years to 10 years. OK, how many people did you just save?

Computing goes times two, and let's say that contributes to drug design, so now you get it in five years, and now you get computing times a hundred and you can design a drug in a doctor's office, you know? You walk in, you have some specific need, [the doctor] solves the problem with the computer, right there. You walk out with the drug that cures cancer. I mean, it's not inconceivable.

It's just so different from anything that you can imagine now. The other thing is, you know, I look at my son, he's two-and-a-half, and there's so much information going into his head that could not possibly have gone into my head at that age because these computer teaching tools are here. The computer is patient, the computer will go do something a hundred times…

Computing is going to be a part of our lives and super-computing is going to contribute to the quality of our lives in a positive way because we're using these tools to think better and to solve problems better.

SOUND BITE: WAV(563k) | Bamba

Q: Let me follow up on that, talking about the pharmaceutical industry and maybe being able to, if not cure disease, maybe impacting on the lives of those who might have a particular affliction. Do you talk to people in the medical community or the pharmaceutical community and obtain any feedback about what they see on a future horizon in terms of how computing might impact their professions?

J.H.: I don't know what they have to say. I know a few of the specific problems. I mean, one is the multi-body problem, which is the same as in astrophysics. You've got all these objects and they all interact with each other and there's [unlimited], millions and millions of objects you're talking about in designing a drug. These are the atoms of something, and you've got to figure out the interaction that each atom has with each other atom and that takes a lot of computing power.

Now you get into a whole bunch of distinct problems in understanding how molecules work, but they're all … it's all simulation; you know, how does this molecule behave? Do you invent a molecule and say, well, what does it do? And you've got to simulate that with the computer. I know people who have been talking to the pharmaceutical industry about that problem, I don't know any more specifics other than that.

Q: What do you do on the team? What's your role?

J.H.: I do the software.

Q: What's your day-to-day like, now that you're leading up, let's just say since August, leading up to actual match day? What are some of the things you get yourself involved in?

J.H.: Well, I think of four things we do, that we put into doing chess and doing it well. And one is the technology. You know, we're using SP to the fullest of it's … I'm really very fully using SP in every possible way.

The second thing is, we're doing some exercise in architecture. We're designing a piece of the computer to solve [the chess] problem fast. That's why we're five years ahead of the game, because we have a special co-processor. The third thing is algorithms and software. You can do the search better because you do the search software better. And so, you play chess better because it's effectively faster.

The fourth thing is chess knowledge. If you look at these four things, three are just technology – how do I do computing better? And one of them is how do I play chess better? And if you look at the way Deep Blue plays chess, there's really just a little bit of chess knowledge and then when you amplify that with the computing ability, then you get a huge effective chess-playing ability.

It plays a good move. I mean, you don't know – from a tourney chess point of view, you don't know where that move came from – but it plays a good move. And where it comes from is just a little bit of chess knowledge, and then through a big search, you sift through and you find all the good positions.

Q: Is this a huge database of possible moves?

J.H.: No, no database. It's all live software. In three minutes, from ground zero it computes everything it knows about the current position from scratch. And that's the software that I do, the real-time software.

Q: Does it take a strategic look at the board?

J.H.: Yes. You look at the board, you try to integrate what you know about chess, what kinds of things are good in a certain position. There's a program that runs that does this type of analysis, sets up all the parameters for the search, then the search runs. That's what you use the SP for.

I mean, you break down the search into pieces of the search on each computer and the RS/6000 (link to http://www.rs6000.ibm.com/) and the SP, and you make those guys work together as best they can to search as fast as you can. That's the essence of the problem, when you've solved search, you've just solved every problem.

If you solved search, you would solve every problem that's in front of anyone because you can always reduce any difficult problem to just search through all the possibilities. In some sense, the fact that we play chess well is just that. We just search through all the possibilities and come up with the best answer we see, out of all the possibilities; it's a very simple technique.

Q: How many moves a second can the computer evaluate?

J.H.: I think the peak speed right now is probably about 1,000,000,000 positions per second. But this is peak speed, which when you do parallel computing you redefine that as the speed you're guaranteed never to exceed. And, in fact, you never get close. I mean, when you're doing parallel computing you look for certain problems. Point-three efficiency is really good. So, we're shooting for about 300,000,000 positions per second. If you translate that into MIPS, it's about two terra-ops.

Q: MIPS, terra-ops are?

J.H.: Millions of instructions per second. If you had to run this algorithm on a general-purpose computer at speed, you'd get about, you'd have about two Terra-Ops, which are a trillion instructions per second. That's the equivalent speed on another computer.

Q: And Garry's speed is just a little slower?

J.H.: Yeah, but this is where human strengths come in, because he has a lot of chess knowledge and a little calculation ability. I mean, he has very precise calculation ability, but it's sort of the opposite. You know, what I said about computers playing chess is that they have a little bit of chess knowledge and a lot of calculation power.

With Garry it's weighted the other way. He has tremendous feeling and intuition and knowledge about chess, which is a very human thing. And he's a tremendous performer in terms of taking that knowledge and doing something with it and calculating what he needs to calculate, which is – if you look at the studies – about 200 positions per second, no, 200 positions in three minutes. And he knows so much. He can look at a chess position and say, "Oh, I know, this is the move." And he's almost always right.
SOUND BITE: WAV(630k) | Bamba(33k)

Q: Do you play chess?

J.H.: I'm a bad chess player, so no, I don't.

Q: Do you find that ironic?

J.H.: Yeah, I mean, when I think about it, yeah.

Q: Have you played Deep Blue?

J.H.: Have I? Yeah, I did once, Deep Thought, I think.

Q: How badly did you lose? How many moves?

J.H.: I don't remember, it was fast. But I don't have to play Deep Blue to lose badly to a computer. But that just shows you what you can do with a computer, because I'm vicariously playing chess at a world champion level. I'm not an expert, certainly less than an expert actually. But if I'm the one programming the computer and the computer can play at that level, that shows you what you can do with a computer.

If I were trying to work on some other problem and if I weren't a grandmaster molecule designer, if I were just an expert molecule designer, and I had a good tool, I can be a grandmaster molecule designer.

Q: So do you define the computer as extending human's capabilities?

J.H.: That's exactly the way I define it. I think that Deep Blue just shows us that we can think a little bit better than we ever did before, and that's what's meaningful to me, really, truly.

Q: Has your vision of the implications of Deep Blue changed over the last year?

J.H.: No. No, I think maybe I can speak for some of the other people, but for me, you have to have faith that -- a sort of faith -- based on what I see in front of me when I look at the program… I see something that could win a match against the world champion. Or, you know it's potentially possible, I see that possibility. Obviously, you have to prove it. So…

Q: And you got that back in college then?

J.H.: No, no, no. We could see, maybe two years ago, that it was going to come together. Does that make any sense? I mean, you know, in the middle of it you can see what's going to happen, it doesn't necessarily surprise you that much when it happens. But it certainly makes you happy to prove it out.

Q: Did you guys all go out to a bar after you won game one last year, and celebrate?

J.H.: I don't think hardly anybody in this whole building [The Thomas J. Watson Research Center] goes to bars.

Q: Last year's match was a historical event, there was so much media coverage and you guys did, literally, make history by winning the first game. It was the first time it had ever happened against a world champion. As a member of the development team, how was the event for you? What was the atmosphere like, was it just very exciting? Were you able to sleep throughout the entire event?

J.H.: I could have slept the whole week. This maybe isn't the interesting answer for you but, you know, when I come to that event, I've done all my work. If I'm not totally used up, then I haven't done everything I can possibly do. So, I came to the first game, first day, used up.

Q: It was exciting for you though, I would imagine.

J.H.: Absolutely. No, so how do I feel about the event? You know, it had more impact later. I saw this interview with the guy who lit the torch for the Barcelona Olympics. He was an archer, with polio, so he didn't run and light the torch, he shot an arrow. And what he said was, "You know, when I was shooting the arrow, all I was thinking about was shooting the arrow. And it was only later, when people talked to me, that I realized just what it meant to people to see this happen" – and it's a little bit like that for me.

You know it's not until months after when you talk to a bunch of high school kids and ask them what they were interested in that you get some of that feedback. That's actually what's makes more of an impact for me. Because when you're doing it, you're just doing it.

Q: That makes a lot of sense. An analogy could be an athlete who is participating a sport. He doesn't really realize it because he's concentrating so much on doing it, but afterward he realizes the magnitude of what he's doing.

J.H.: Yeah, because you really are just used up, doing everything that you can do. So the excitement doesn't have that much impact at the time, truly.

Q: When and how did you get interested in computers?

J.H.: Solving and doing software just seemed obvious, like an obvious choice, sort of like solving puzzles. I don't know. It's stimulating.

Q: Can you think of the first computer you ever saw?

J.H.: There was a Radio Shack TRS model 80 in my library in high school, right? And you could sit down and write some basic programs on it and I remember looking at the directions and figuring out how to make the cursor bounce around the screen, like in Pong; it would bounce up and hit the top. I think that's the first program I ever wrote. That was senior year in high school.

And that's the other thing. What's going to happen… I'm good at doing software, but kids who start doing software at age 5, how much better are they going to be by the time they're 35? You know? It's like chess players. They consider it way too old if you start trying to play chess at age 15.

SOUND BITE: WAV(693k) | Bamba(36k)

Q: Are you going to encourage your son to get into programming?

J.H.: I'll encourage him to do whatever he's interested in.

Q: What programming language do you use?

J.H.: It's all in C.

Q: And is there one big Deep Blue program?

J.H.: Yeah.

Q: What operating system does the Deep Blue computer run on?

J.H.: Well, the basic software, it's all in AIX, but the IBM SP Parallel System is called MPI. It's a message-passing system. So the search is just all control logic. You're just passing control messages back and forth to say, well, what am I doing? Did you finish this? OK, here's your next job. That kind of thing. At the SP level. At the search level, you're saying, OK, here's a position, I need to search all the moves. And you go search all the moves, all at the same time, preferably. On a bunch of different computers.

Q: Do you have a prediction for the match?

J.H.: I have very high hopes for at least even. I mean it very easily could go three-three. You know chess is so variable that it could go back four-two again, or it could be better. I think we have a much better chance of doing better, I think our upside chances are much higher, but Mr. Kasparov has taken this match seriously and he's preparing even now. And he prepared maybe a week last time; he didn't take it as seriously last time. This time he's taking it very seriously.

Often you may sort of flippantly say that computers are getting better all the time and humans are staying about the same, but that's not true in this case. Kasparov's going to be better for playing the computer this year, much better. He's going to have thought about what it can do and what it can't do, and have prepared his openings, which are his strength, his pre-game preparation.

You know, you can just lose the game because of the opening, and that's what he is good at doing. Making people lose because of all the preparation he's done. So he's not staying basically the same. He's going to be much stronger; we're going to be much stronger – much stronger. But, as for a prediction, it's sort of like last year; there are great strengths on each side that are going to juxtapose… And who knows what strengths are going to play off what weaknesses? It's very difficult to say for sure.

Q: It will be an exciting match, nonetheless.

J.H.: No doubt about it. You'll see different things than last time.