RealTechNews.com - Independent Tech

Comments on this Item:

IBM using a team of programmers and their grand chess master consultant (Joel Benjamin) to beat Kasparov is like using oxygen to climb Mount Everest. Hey, they do use oxygen when climbing Everest, although it is more impressive if they don't.

# posted by georgef 7:16 PM

Chessboard ... $50.00

A group of programmers, a consulting chess master and a souped up computer ... $1.5 Million

Ganging up on Garry Kasparov, kicking sand in his face, and making the world believe (at least for a moment) that Garry was beat by a computer .... priceless!

*****

As Gus (the ship's engineer in Tripping The Rift) said as he was beaten with an aluminum bat " well, that's fair. "

# posted by Anonymous 7:24 PM

George, I think it would have been more like using a snow cat.

# posted by Bill 7:31 PM

I'm not sure it's cheating; a lot depends on how the IBM team tweaked the programming between rounds.

For example, many computer chess programs depend heavily on databases of board positions and the corresponding correct responses. If the IBM team had Benjamin analyze the first loss to determine what the program's correct response should have been, and then added that game (and it's solution) to Deep Blue's database, then I think that's fair. That is enough to keep Kasparov from simply replaying the first game over and over for an easy win.

I think you approach a grey area when the programming team selects a different set of analysis algorithms to better counter Kasparov's game. Computer programs tend to evaluate many more possible moves than human players do, and (in addition to more processing speed) improvement in the program's ability tends to come from making better choices about which moves to evaluate. It might be possible to observe Kasparov and have Benjamin make comments on his style of play: threats and opportunities Kasparov didn't seem to notice, moves he seems to favor or disfavor, and the like. Deep Blue's analysis algorithm could then be tweaked to look for the things that Kasparov tends to overlook or disfavor, and exploit them. This is in effect making the program better at playing chess against Kasparov, but not necessarily better at playing chess. On the other hand, if Kasparov is the greatest chess player in the world, I'm not sure how much difference that really is, and would be hard put to decide if it is cheating or not.

It would have definitely been cheating to have Joel Benjamin pick moves for the computer, either on his own or from a list provided by the program's analysis, then have Deep Blue relay Benjamin's choice to the board. That would be Benjamin playing against Kasparov with computer assistance, not a Deep Blue versus Kasparov match.

# posted by D.J.Wildstar 5:34 AM

Mr. Wildstar, you miss the basic point and are bordering dangerously close to relativism. ;-)

Once the first match begins, the computer must be out of bounds to anyone. To be fair, the computer must either have the skills to succeed ot be able to learn on its own, not be "taught" by third-parties along the way. As far as I remember, Kasparov didn't have a team of computer scientists in the wings waiting to tell him how he should expect the computer to react during the next match. It was his intelligence (CPU/memory) utilizing his existing experience (database) --and that was it.

Anyone can build a wall. When the wall builds itself, that's an achievement.

# posted by Bill 7:50 AM

I'm not sure I think it's cheating. But it is certainly not a match. As Bill said, the computer should be untouched, so it is more like a series of single game matches, each with an upgraded computer.

But I must disagree with an earlier statement that the same game could simply be replayed. Software is no longer necessarily predictably deterministic. The cheap and easy example of this is Microsoft Windows. :) When an algorythm becomes over-specialized one of the first things I'd do to loosen it up is throw in a couple random variables. Is it cheating to use nearly random inputs to the computer without controlling the initial state?

# posted by Tom 8:04 AM

But Tom, this occurred in 1997. We weren't quite that savvy 7 years ago.

# posted by Bill 8:08 AM

So if Deep Blue was able to log its game with Kasparov, add it to its database on its own, and mark it as "bad", that's okay, but if a group of scientists do the work on its own on the backend, that's not? Just curious.

Personally, I think even if IBM rewrote the program after ever game, that's still amazing. Think about it- they made not have come up with a computer system to beat the world champion, but they did come up with an algorithm to do so (Step one- play game; If loss, Step two- adjust; Step three- Goto Step one). The fact that we got that far is simply amazing.

Software development is continual. Think of it this way- imagine if Microsoft refines the Windows system to one day be bug free and impervious to security problems. Will you be more amazed at the fact that MS evolved to that point or frustrated that they didn't do it out of the box?

# posted by Ray 8:41 AM

Ray, try thinking of Deep Blue as a medical diagnostic program... Patient dies, adjust program...? IBM touted Deep Blue as being able to beat the man. It didn't. It couldn't --not without being modified every time it made a mistake.

# posted by Bill 9:29 AM

To be fair, the computer must either have the skills to succeed or be able to learn on its own, not be "taught" by third-parties along the way. As far as I remember, Kasparov didn't have a team of computer scientists in the wings waiting to tell him how he should expect the computer to react during the next match.Certainly, some people do learn from making mistakes, being shown their error and a correct solution, and proceed to bigger and better things (or at least brand new errors). This is a valid learning process for both humans and computers. Perhaps not Kasparov, but certainly for many other chess players learn this way: loose a game, have an expert explain their mistake to them, and learn from the experience.

I know that in many other "fair" contests that consist of multiple matches over time, the participants are allowed to have coaches and other support staff advise them between individual matches. In a chess match, would contestants be allowed to dissect a game afterwards, and perhaps do research in books on similar matches or players, or discuss it with other experts and adjust their strategy and tactics before the next game?

If this wasn't allowed - in other words, if the rules of a chess match forbid the use of any external help - then certainly Deep Blue cheated. If Deep Blue were a human contestant, would it have been allowed to consult with another chess grandmaster or a psychologist to improve its chances of winning the next time? Would Kasparov have been allowed to place a call from his hotel room after his first loss, and talk to a computer scientist about likely ways to beat the program?

# posted by D.J.Wildstar 11:09 AM

I have to side with Bill. Using a consultant to assist in making the initial program is cool, but using your team and the consultant to 'tweak' (reprogram) the system between matches is not fair. Either you have a system that can analyze the opponents moves and learn from its own mistakes, or you didn't. Any human intervention means you didn't. Even if you have a system that can analyze human strategy, you'll find great difficulty in attempting to counter the unpredictable nature of the human mind.
It should be possible to make an unbeatable computer opponent. I'll wait until it happens.

# posted by metermax 11:19 AM

Wildstar, it's one thing to coach someone. All that means is that you give the person advice and that person has the responsibility to remember it and must use judegement to apply it properly. When you program a computer, there's no danger that it will forget what you said and you're actually telling it, in real time, when it's appropriate to apply the advice. To my mind, that's a huge difference.

# posted by Bill 11:30 AM

Your premise about cheating is flawed. How is it cheating to intervene between each match? Did the rules state that no programming changes could be made? Absolutely not! The rules stated that no human intervention could be done *during* each match. So was it cheating? How could it possibly be cheating if no rules were broken? IBM never made the claim that Deep Blue could defeat Kasparov with no changes between matches. IBM never made the claim that Deep Blue could defeat Kasparov without expert help between matches.

Was it have been more impressive if Deep Blue could learn on its own and not depend on human intervention to win? I will grant you that. But being less impressive is a far less serious charge than "IBM cheated."

# posted by Anonymous 2:39 PM

Todd, if we're in a boxing match and inbetween rounds I get a shot of "juice" that lets me beat the hell out of you, isn't that cheating? Did I have to say I was going to beat the hell out f you? Did I have to say I wasn't going to get shot up?

It's the same thing... except, a chess match has some basic rules like you can't change people after each match. Essentially that's what IBM did.

# posted by Bill 5:50 PM

Didn't Kasparov beat "Big Blue" hands down the previous year? So the programmers,engineers and chessgamers built a deeper "Blue." So if a rowing team beats a motorboat and the engineers bolt on a bigger motor, what does that prove?
These bouts are just publicity stunts anyway. I'll remember that John Henry won the first match with that steam drill, not that the drill was easier to upgrade.
Bill R.

# posted by Anonymous 10:01 AM

Kasparov prevailed against a computer I think was called Blue Jr. IBM then went back and built a better cattrap for the next round.

# posted by Bill 11:37 AM

There is not such a thing like tweaking an algorithm to beat a grand master strategy. These are the kind of algorithms that are most complex and that cannot be done from one day to another. A dramatic prove of how bad played deep blue was the first game and even worst the error made during the second, just before Kasparov resigned which would allowed him to perpetually check forcing draw, not even the most faulty chess engine does a mistake like that, mathematically impossible for a machine but perfectly natural in a human. Why isn’t IBM doing more research on this? What a single match proves? In any case Kasparov wasn’t defeated except psychologically.

# posted by Anonymous 2:54 PM

Post a Comment