Kasparov vs Deep Blue

[Charlie Reams]

I was just watching the awesome BBC documentary about the famous match in which Deeper Blue became the first computer to defeat a chess world champion. In the documentary, Kasparov openly accuses IBM of cheating, and suggests that at some crucial points in the match the computer must have received human assistance, because it played such human-like moves.

I guess there must be some people around here who are interested in at least one of chess, AI and conspiracy theories so... what do you reckon?

[Kai Laddiman]

It's a term that Charlie has been associated with a lot, but does not tend to use himself: sour grapes.

[Gavin Chipper]

Don't really know anything about it, but couldn't he have challenged it again under more controlled circumstances, with James Randi keeping a watchful eye?

[Charlie Reams]

Don't really know anything about it, but couldn't he have challenged it again under more controlled circumstances, with James Randi keeping a watchful eye?

IBM declined a rematch and dismantled Deeper Blue immediately after the game. This was particularly weird because, after Kasparov beat the original Deep Blue a few years before, he immediately offered them a rematch.

[Junaid Mubeen]

I guess there must be some people around here who are interested in at least one of chess, AI and conspiracy theories so... what do you reckon?

Isn't there one particular forum member who thrives on all three?

[Dan Vanniasingham]

I think I watched this some time ago, does it end with a camera panning out of a room full of computer hard drives?

I'd probably need to see it again to refresh my memory, but it was a great watch. I'd consider the "optimum" chess player to have the calculatative abilities of a computer with the intuition of a human, and whether IBM somehow succeeded in giving Deep Blue some of the latter is a mystery. I don't think many can on the planet can match Kasparov's insight, but at the same time he could simply be bitter at (eventually) underestimating the program.

This has turned out to be a disappointingly rubbish post. I think I'll Foe myself now.

[Charlie Reams]

I think I watched this some time ago, does it end with a camera panning out of a room full of computer hard drives?

I think it's a zoom in, but sounds like we're on the same page.

[Gary Male]

It's the nature of the elite chess players to have a dose of paranoia. Given that the programming was to "learn" as much as possible by studying thousands of top-class games and work out the positional qualities leading to its own move weightings instead of the old mangrabbing, I don't think it's a surprise it was able to avoid some of the traps that had previously worked on computers and play a more "human" style at times.

IBM then had nothing more to prove. Deep Blue was only going to get deeper and deeper so after showing a computer could beat a world champion and getting all the kudos possible there was no value in a rematch for them.

Bring on quantum computing.

[David O'Donnell]

It's the nature of the elite chess players to have a dose of paranoia. Given that the programming was to "learn" as much as possible by studying thousands of top-class games and work out the positional qualities leading to its own move weightings instead of the old mangrabbing, I don't think it's a surprise it was able to avoid some of the traps that had previously worked on computers and play a more "human" style at times.

IBM then had nothing more to prove. Deep Blue was only going to get deeper and deeper so after showing a computer could beat a world champion and getting all the kudos possible there was no value in a rematch for them.

Bring on quantum computing.

Kasparov had a point:

1. For a start, Deep Blue's opening books are human generated;

2. Deep Blue didn't seem to make any 'bug moves' which is strange since it basically is a number cruncher. Deep Fritz makes 'bug moves' like refusing repititions in drawn games, because it reads the position as say plus 0.14 (slightly over a tenth of a pawn), and trying to force a win that doesn't exist only to end up losing. Deep Blue should be more prone to these errors yet strangely wasn't against 'the Beast of Baku.'

Kasparov didn't have a point:

1. At times he tried to go toe-to-toe with the machine which seems like an outrageously silly strategy to adopt against a machine that can calculate 6 billion positions per second. Most anti-computer players will opt for a closed system that computers tend to not understand and one that minimises the tactical superiority of the computer. Kasparov was a little arrogant in thinking that he could out-calculate Deep Blue in tactical play.

Kasparov may have had a point but it would have been better made had he not started the game in such an arrogant fashion.

[Dinos Sfyris]

All in favour of a new KILLER bot Apterous Blue?

[Gary Male]

But over 6 games, one of which Kasparov made a blunder effectively ending the game before it even got going, is that really enough games to show Deep Blue wasn't going to make any "bug moves"? Maybe over 12 or 18 games it would have shown some but there's no real reason the various weights involved especially if it was geared up specifically to play Kasparov it could have avoided those situations over a very short sample.

[Ben Wilson]

All in favour of a new KILLER bot Apterous Blue?

Played Rex lately? :p

[Kevin Thurlow]

Kasparov doesn't really have the right style to play a computer. He does (or did!) calculate incredibly well, but does play "interesting" moves - taking risks, which may not be 100 % correct but are effectively beyond human calculating depth. The computer using "brute force" (i.e. He does that, I do this, he does...) can nowadays work this all out. For years now I have been setting up the critical position in a postal game I had where I found a very good winning idea (it helps being able to write notes and move the pieces of course). Fritz 4 and Fritz 5 could not find the idea as it was many moves deep, Fritz 8 found it in a few minutes, but Fritz 10 found it in less than a second....

Computers are playing more humanly now - some years ago you could just let the machine take a pawn or even a piece on the wrong side of the board (which it always did), and you won routinely on the other side as its pieces were misplaced. That doesn't happen now.

Having said that, the computer's team did have several grandmasters around (as you would expect) so you can understand Garry being a bit suspicious that they were feeding in good human moves. Remember too he wasn't used to losing, which is going to make him a bit ratty. Also he probably intimidated his opponents just with his physical presence and his intense aura of power - that doesn't work on computers. I interviewed him at a press reception (in 1984? for the London match vs Smyslov) and just felt the energy he was giving off - he frowned slightly at one question he maybe didn't quite understand, and I felt like taking a pace backwards. Having said that, he was world champion because he was the best player. I would expect an on-form Kramnik (who doesn't make many mistakes) to do better against a computer than Kasparov did.....

The fact that the computer was shrouded in secrecy and dismantled so quickly afterwards is interesting, but maybe they were worried about industrial espionage. Of course Kasparov was fantastic at assessing the character and style of his opponents, so if the computer suddenly acted out of "character", he would notice.

Having said all that, I guess the answer is "I don't know if the computer had human help".

Kevin

[Gavin Chipper]

Obviously the whole point of this was to get a computer to beat a human chess player. But all this was several years ago, so things have presumably moved on, so hasn't anyone else since made a computer that can definitively beat any human? If so, wouldn't that make the whole episode a footnote in history? I could probably look this up easily enough though.

[Charlie Reams]

Obviously the whole point of this was to get a computer to beat a human chess player. But all this was several years ago, so things have presumably moved on, so hasn't anyone else since made a computer that can definitively beat any human? If so, wouldn't that make the whole episode a footnote in history? I could probably look this up easily enough though.

Yes, modern programs (which run on commodity hardware rather than some special machine like Deep Blue) are slightly stronger than humans, although since Kasparov's retirement the standard of human chess has fallen slightly too. It's likely that within 50 years the idea of a human vs computer chess game would be about as interesting as a human vs car road race. However it's like putting men on the moon; the question is not so much whether we can do it now (indisputable) but whether they did it then (mildly disputable.)

[Gavin Chipper]

Obviously the whole point of this was to get a computer to beat a human chess player. But all this was several years ago, so things have presumably moved on, so hasn't anyone else since made a computer that can definitively beat any human? If so, wouldn't that make the whole episode a footnote in history? I could probably look this up easily enough though.

Yes, modern programs (which run on commodity hardware rather than some special machine like Deep Blue) are slightly stronger than humans, although since Kasparov's retirement the standard of human chess has fallen slightly too. It's likely that within 50 years the idea of a human vs computer chess game would be about as interesting as a human vs car road race. However it's like putting men on the moon; the question is not so much whether we can do it now (indisputable) but whether they did it then (mildly disputable.)

Over to Richard Brittain?

[Martin Gardner]

A friend who used to play chess told me that Kasparov was ill.

[Matt Morrison]

A friend who used to play chess told me that Kasparov was ill.

ill as in 'down with the kids' or ill as in 'not very well' ?

[Michael Wallace]

A friend who used to play chess told me that Kasparov was ill.

And it was from lines like this that conspiracy theories are born.

[Martin Gardner]

I like chess but I'm really bad at it (well, it depends who I compare myself too, I suppose). I played once at a club and got beat very quickly, and my second game was a draw against another inexperienced player. After that, I started Uni and I couldn't be bothered playing anymore.

I can't see why a human player would ever beat a computer - surely the pure processing power of the computer should mean it can 'foresee' any trap a human player can come up with? Obviously not, I'm just saying I don't understand.

[Jimmy Gough]

A friend who used to play chess told me that Kasparov was ill.

ill as in 'down with the kids' or ill as in 'not very well' ?

"Wicked", "Sick", "Insane". "Ill" is a new one to me though.
(this is blatantly the most interesting, intelligent and useful post on this thread - like all my posts really)

[Paul Howe]

I can't see why a human player would ever beat a computer - surely the pure processing power of the computer should mean it can 'foresee' any trap a human player can come up with? Obviously not, I'm just saying I don't understand.

Not necessarily. Say its white's turn and there are 50 (just pulling a number out of my arse here) possible moves, and on average black has 50 possible responses to each of these moves. That means just to look 1 turn ahead there are 2,500 possible games to examine, and 50^(2n) possible games n moves ahead. This is a very big number, even for small n, and even for a computer. The upshot of this is that a computer can only look a certain number of turns ahead. A human can lay a trap if they can find a position that looks tempting to a computer (as judged by an "evaluation function", which assesses the value of a given position) for the number of moves it can look ahead, but actually turns out to be a losing position beyond the limits of the computer's vision. Knowing very little about chess, I have no idea how easy this is for a human, but it certainly seems the top players are capable of it through some hard won chess intuition that bears little resemblance to how a computer plays.

That's a grossly simplified explanation, but captures the gist of it. Designing a good chess program is about coming up with a good evaluation function, and deciding which positions the computer should look at, given that it can only look at a small fraction of them. The same basic principles apply to other board games. An unbeatable computer draughts player was created decades ago, but the best computer go players can only match an intermediate human as there are many more possible moves per turn than in chess, so the computer has a very limited horizon.

[Paul Howe]

Regarding dismantling Deep Blue after the match, designing a computer that defeated the world chess champion was a great PR coup for IBM, but the match was close (3.5-2.5, and Kasparov could've forced a draw from the game he lost), so I doubt IBM would want their achievement tarnished by possible defeat in a rematch. A shame from an intellectual perspective, but sadly perfectly sensible from a PR viewpoint, without the need to invoke conspiracy theories.

[Jon O'Neill]

BUT WHERE ARE THE PRINTOUTS? THE LOGS??

[Michael Wallace]

BUT WHERE ARE THE PRINTOUTS? THE LOGS??

Exactly. You know what else we don't have printouts for? The moon landing.

QED, biatches.

[Dinos Sfyris]

"Wicked", "Sick", "Insane". "Ill" is a new one to me though.
(this is blatantly the most interesting, intelligent and useful post on this thread - like all my posts really)

I don't think any self-respecting homie says Illin' any more. A bit like diss. It died out.

[Gavin Chipper]

Obviously the whole point of this was to get a computer to beat a human chess player. But all this was several years ago, so things have presumably moved on, so hasn't anyone else since made a computer that can definitively beat any human? If so, wouldn't that make the whole episode a footnote in history? I could probably look this up easily enough though.

Yes, modern programs (which run on commodity hardware rather than some special machine like Deep Blue) are slightly stronger than humans, although since Kasparov's retirement the standard of human chess has fallen slightly too. It's likely that within 50 years the idea of a human vs computer chess game would be about as interesting as a human vs car road race. However it's like putting men on the moon; the question is not so much whether we can do it now (indisputable) but whether they did it then (mildly disputable.)

I wouldn't compare it to landing on the moon myself. Everyone knows the year that happened (1969 for those that don't know ) and if it turned out to be a fake then it would be a big shock to most people even if they sent people to the moon tomorrow to make up for it. But if someone said "We actually cheated to get a computer to beat Kasparov back in 199-whatever but a computer can definitively beat any human now and we can openly show this" most people would probably just say, and I quote, "Yeah fine - we can draw a line under that now."

[Gavin Chipper]

Don't really know anything about it, but couldn't he have challenged it again under more controlled circumstances, with James Randi keeping a watchful eye?

IBM declined a rematch and dismantled Deeper Blue immediately after the game. This was particularly weird because, after Kasparov beat the original Deep Blue a few years before, he immediately offered them a rematch.

I heard that he was initially going to dismantle himself.

[Charlie Reams]

An unbeatable computer draughts player was created decades ago

Draughts was solved completely quite recently, in the sense that there now exists a program which plays literally perfectly, forcing at least a draw against optimal play and obviously wiping the floor with most human opposition. Unfortunately the Team Humanity peaked rather too soon, since by far the greatest player of all time died long before the program was completed.

[Michael Wallace]

An unbeatable computer draughts player was created decades ago

Draughts was solved completely quite recently, in the sense that there now exists a program which plays literally perfectly, forcing at least a draw against optimal play and obviously wiping the floor with most human opposition. Unfortunately the Team Humanity peaked rather too soon, since by far the greatest player of all time died long before the program was completed.

I remember reading about them solving draughts, and the implication was that the Draughts world were really pissed off about the guys who made the program claiming that it could beat Tinsley (even though he was dead). I'm not really sure where I'm going with this, since it turns out I can't actually remember the story very well at all, other than that it was about them solving draughts, and about some draughts players getting a bit annoyed.