Well, obviously I picked an extreme case. But there are always blurry edges: If a computer uses a particular opening and loses 60% of the time, does that mean that that opening is sub-par, or just that the computer isn’t very good at playing it?
(N.B. I edited that quote slightly to save space)
The normal scoring is indeed 1 for win and 0.5 for draw (i.e. one win is worth two draws.)
The scoring of one win = three draws is to encourage players to play more decisive chess.
Similarly there is usually an associated rule that players cannot agree a draw in under 30 moves (this was introduced in the Sofia tournament) and also that the referee has to agree too.
The difference comes about because there are three main types of event featuring world-class players:
- championships (where the scoring is as usual)
- long-lasting traditional events like the recent Tal Memorial in Russia (where the scoring is as usual)
- newer shorter events, like London (which use the dynamic scoring)
You are correct about the time limit (introduced by Fischer, I believe), which avoids both slow adjournments and ‘ruthless’ time limits (where one or both players has only seconds to make all their remaining moves)
Note that the Tal Memorial had a** lot **of draws - not so interesting for spectators.
The organiser of London, Malcom Pein (top bloke and long-standing friend of mine) wants to make London:
- exciting (e.g. there’s an extra pot of money divided amongst players who win games)
- good for spectators (live coverage plus live commentary plus players visit the commentary room)
- helpful to schools’ chess (there’s an associated charity to do just that)
- a real opportunity for English players to meet the best in the World (achieved!)
I know what you mean, but a 60% loss would literally never happen.
All computer chess programs have ‘opening books’, i.e. massive preparation.
And all this comes from thousands of Grandmaster games.
So computers are relying on decades of human practice for their initial opening preparation, which means they will not do badly in the opening. (After all, their memory is perfect compared to us carbon-based life-forms!)
And even if the computer gets a poor position out of the opening (perhaps because a new move has recently been discovered), they might still outplay their human opponent in the rest of the game.
I hope that makes sense!
Glee, some of the computer programs on fairly cheap PCs are rated over 3000 now, which is a couple hundred over the best human.
Do you think those ratings are accurate? Have we reached the point that a free program (like Houdini) on a <$1000 PC would be favored to beat the best human under tournament conditions?
Well, this is considered suboptimal (by White):
- f3 e5
- g4 Qh4#
I don’t think you could call it a longstanding opening, though :).
I would assume that, for the purposes of analyzing the strength of various openings, you’d pit the computer against itself. So both sides would have the advantage of perfect recall, no blatant mistakes, etc.
Both of these deviations from the norm serve the same purpose- to limit draws. See, in the olden days, when chess was popular, people would willingly put up lots of money to watch the best of the best go at it with each other. Today, if we wanted that, we’d just put Ribka up against itself. And because of computers, chess has advanced so fast that we’re nearing perfect play. And so Grandmasters are drawing each other more and more.
The problem with this is that it’s boring. People don’t want to pay to see draws. They want action. So tournament organizers have come up with all sorts of ways to discourage draws. One of the ways is to disallow the so-called “strategic draws”, where players agree to a draw early in the game in order to, say, clinch a tournament victory. That’s why the arbiter is necessary to declare draws. Another way to do it is to offer more points for a win, so players not only have more incentive to go for the gold, but also to make “I can clinch the prize with a draw” scenarios less likely in the first place.
That’s correct.
On the other hand, limiting draws this way would have nullified at least one piece of brilliant strategy I’ve seen, which did in fact lead to a win. In one of the Kasparov/Deep Blue matches, one game he was under serious time pressure, and so offered a draw. Now, Kasparov is an absolute master at the psychological aspect of the game, which of course wasn’t much use against a computer. But at the time, Deep Blue wasn’t programmed to be able to offer or accept a draw, and so the rules were that the human IBM team would make that decision. And Kasparov, master of psychology that he is, knew that the humans would decline the offer, but that it would take them a long time to come to that decision. Thus taking the time pressure off of him, and giving him the breathing space to rally and win the game.
My understanding is that those ratings used to come mainly from games v other computers, so they’re weren’t necessarily accurate in that sense.
However Hydra (for example) scored some great results v top Grandmasters before being ‘retired’:
Another point is that top players are starting to set conditions before they play matches v computers.
For example on Kramnik v Deep Fritz, it was agreed that while Deep Fritz was in its opening book Kramnik was allowed to see Fritz’s display.
I think also Kramnik could claim a draw if the position was in Deep Fritz’s engame tablebase (since the computer would then not need to analyse but would still play perfect chess.)
The Wiki article reckons that computers can now beat the best humans and I have no evidence to refute that claim (although there have been few recent man v machine matches):
That’s the only 2 move mate - I believe this is the only 3 move mate:
- e4 e5
- Qh5 Ke7?
- Qxe5 mate
My quickest checkmate in a serious game was:
- e4 e5
2.Nf3 Nc6 - Bc4 Nf6
- Ng5 Nxe4?
- Bxf7+ Ke7
- Nxe4 Kxf7
- Qf3+ Kg8??
- Ng5! Qxg5
- Qd5 mate
I haven’t discussed it with any chess programmers recently, but I don’t think two computer programs would be any use at analysing openings!
I expect they would each come up with just a single ‘best move’ each time*, so you simply wouldn’t have the wide range of moves played by humans.
*I think you can program a computer to play the ‘second-best’ move sometimes, but it’s still not helping.
I’m sorry, but I don’t think this is either brilliant (rather sharp practice in my opinion), nor any practical use.
(I’m sure that modern programs can agree draws using their assessment function.)
Well, you can make the computer do whatever you want. There are a finite number of openings and you can have the computer try each one.
Finding fool’s mates by computer would be trivial for instance. Harder but imaginable is finding a short opening which–while it doesn’t immediately lead to checkmate–is nevertheless a guaranteed loss. Of course we know from game theory that every opening must be a guaranteed win/loss/draw, but the computer can discover the perfect play that enables this, assuming it’s not too many moves in the future.
Is it useful? I couldn’t say, but adding humans makes the problem harder. A particular opening may be known to be a guaranteed loss, but it could be that the perfect play required to accomplish this is fraught with so many chances for error that it’s actually a very successful opening in practice. It all depends on who’s playing.
Hang on!
You certainly can’t make the computer ‘do whatever you want’ (construct tablebases for 10 piece positions comes to mind…)
After just 6 moves by each side, there are over 9 million positions for a computer to analyse.
Given that human analysis goes way beyond that (e.g. I’ve given a couple of opening lines here of around 10 moves), it is not possible for the computer to ‘analyse each one’.
Similarly ‘the computer can discover the perfect play that enables this, assuming it’s not too many moves in the future’ is not true with current computer power. (It may never be true, given that the longest human game lasted 269 moves…)
I also don’t see what practical relevance ‘finding a short opening which–while it doesn’t immediately lead to checkmate–is nevertheless a guaranteed loss is’.
In a school match long ago, my opponent as Black tried:
- e4 e5
- Nf3 Qh4??
- Nxh4
This opening will not be repeated by him - but any computer analysing openings would have to study it.
Given that all computer opening books are based on previous human research, ‘adding humans makes the problem harder’ is not correct either.
Heh, I had to laugh when I noticed that it was Dr. Strangelove saying that “you can make the computer do whatever you want”…
A moment, please, Mr. President!.. To avoid the problem of “human error” you have to use it…
You can make the computer do it. It just won’t finish anytime soon :-).
All true. But 3 moves isn’t much, for instance, and we’ve already established that there are some 3-move-or-less openings which are very bad. 9M boards is also not much at all for a computer, if we’re looking for a checkmate on the next move, and there are probably thousands of checkmates in that space.
That’s why I was careful to qualify the statement with “not too many moves in the future”. 269 certainly exceeds this!
You agree, I hope, that some positions lead to inevitable checkmate in some short number of moves. It’s not hard to imagine a 5-move opening that allows the other player to force a checkmate in another 20 moves.
I should point out that we are very far away from doing this in reverse. If there were a short opening that guaranteed a win (instead of a loss), the game of chess would be instantly very boring. Fortunately, it doesn’t seem like we’ll solve this problem anytime soon.
Now we’re getting into the fuzzy details of chess programs. Any board evaluation function will decide that losing a queen that early is really not a great strategy, and mostly likely ignore it in its search space. Still, one might imagine a very fast mate that depends on sacrificing a queen, so if you’re trying to be comprehensive you might want to keep looking.
Well, that’s kinda my point. One could create an opening book by simply pitting one computer on another and finding the “best” one. But the best one against a program is not necessarily the best one against a particular human. Humans make particular kinds of mistakes, and different humans have different specialties. Very crudely, computers are better at tactics than strategy, so if some openings lead to more tactical games, a programmer would be wise to have the program prefer them.
That hardly seems fair to me, because even if there’s no “thinking” in the endgame itself, there was still “thinking” to get to that point. Or do you mean just situations where the tablebase showed that Kramnik could force a draw with perfect play, effectively giving the human access to the tables, too?
Of course, you could also have something with the basic structure of the Fool’s Mate, but with a few extraneous moves tossed in on the side. That gives you a very large number of 3-move (and 4-move and 5-move…) mates. I once won against a computer (on a handicapped difficulty setting where it would occasionally make deliberate mistakes) in six moves, using a fool’s-mate-plus-irrelevant-moves.
Well, not generally applicable, of course. But in that very specific situation, it allowed Kasparov to make use of one of his particular strengths, where it naively seemed impossible to do so. I have to applaud that.
Talking of “basic structure of the Fool’s Mate”, I once beat the kid next door in 3:
- e4, g5
- e5, f6
- Qh5+
Are there any other regular tournaments similar to the London Classic in terms of live online coverage/commentary?
OK, I amend my statement to ‘there are many things computers would take far too long to do, even in the foreseeable future’.
Chess has a vast number of possibilities and the computer simply can’t cope with some chess tasks. I firmly believe opening analysis from scratch is one of them.
I honestly don’t see the relevance of this.
Yes, there are openings which are very bad. If you play two computers against each other they have spend huge amounts of time to analyse all of this junk.
And sensible openings which include a snap checkmate are few and far between
(1. d4 Nf6 2. c4 e5 3. dxe5 Ng4 4. Bf4 Nc6 5. Nf3 Bb4+ 6. Nd2 Qe7 7. a3 Ngxe5 8. axb4?? Nd3 mate for example. All such mates have been discovered by humans.)
Almost all openings take time to reveal who’s better. The brute force approach of computers simply doesn’t work.
Yes, there are forced checkmates. (I’ve played some. 
)
However it’s easy to imagine anything - have you got an example of a 5 move opening that leads to checkmate in 20 more moves?
Even if one did exist, how would a computer discover it?
And most importantly, how does that help in opening analysis?
And now we come to the crux of why I disagree with your idea of getting computers to make up openings.
When do you stop the computer analysing? A Queen up? A pawn up?
How far ahead do you look before you announce this is a good opening? 10 moves? 20 moves? 30 moves?
How would a computer do with the Benko gambit (1. d4 Nf6 2. c4 c5 3. d5 b5), where Black sacrifices a pwan for long-term compensation?
Wouldn’t it simply say the opening loses a pawn?
Chess players have been using computers for decades. Companies have poured millions into chess programs. None of them have ever suggested using computers the way you have…
I know all top Grandmaster events provide the moves live.
However I’m not sure about Internet commentary (there’s usually an on-site commentary room, but it may not be broadcast.)
Here’s a long-running event in the Netherlands that uses computer analysis as commentary:
http://games.chessdom.com/tata-steel-chess-2011-live
OK, the Tal Memorial (incredibly strong in depth!) had live commentary:
However having such a tough field meant lots of draws (10 decisive games, but 35 draws 
)
Linares had commentary last year (sadly it didn’t happen this year.):