Extremely cool. And I figured top chess players would know what was going on. You guys did a good job acting it up in the post-match interviews though 
What about the number of remaining pieces written on the paper? My guess is he pulls a switch at the last minute when the envelope is being torn open.
It wasn’t difficult to expect Mirror chess.
There is so little chess on TV (understandably - it’s not a spectator sport!) that we really wanted things to go well. And Derren was doing something very difficult.
We discussed his extraordinary memory afterwards - out of 700 moves, he only hesitated once when he nearly played Bf6? instead of Bd6.
Remember that one single mistake means he loses that game and the match.
Afterwards there were brief interviews. I confess that I was looking forward to these, whilst my fellow players were thinking about their games. So the dialogue went something like:
Producer: What did you think of the event?
Grandmaster A: I liked it.
Producer: Did you work out what Derren was doing?
Grandmaster A: Yes.
Producer: What did you think of the event?
Grandmaster B: It was fun.
Producer: Did you work out what Derren was doing?
Grandmaster B: Oh yes.
Producer: What did you think of the event?
Glee (shamelessly): I was on the edge of my seat the whole time!
Producer: Did you work out what Derren was doing?
Glee (shamelessly): It’s going to keep me up all night trying to work it out!!
I think you can guess whose interview was used all week beforehand in the promotion for the program…
(I had pupils at school saying “Sir! You were on TV last night!!” and used to casually answer “Yes - it just shows what can happen if you get your homework done.”
As for the envelope, it was indeed a switch. I was expecting it and watching Derren closely - but he cleverly distracted me for less than a second…
Seriously he is a great illusionist with a fantastic memory.
And when it was all over, he took the trouble to have a friendly chat with us and sign some autographs for my pupils.
That is how celebrities should behave!
Looks like Black’s knight is toast. Pawns are underrated, as basically Black’s position has a bunch of holes in it where he pawns used to be, offsetting his (about to be eliminated) +1 advantage in knights.
Pawns are important (when I was a junior I won a bunch of games by first winning a pawn, then trading off all the pieces and queening my extra pawn.)
When I played Nxf7, I knew I was winning the knight back.
glee, I’m still not clear on why this can’t work.
We know that, overall, White wins about 55% of the time, counting draws as half a win.
One thing computers can do is play a large number of games at a much greater speed than human players can.
Now say you want to test, say, the King’s Gambit (1 e4 e5 2 f4) to see how it stacks up. Have two of the best programs play each other a million times, from the position after White’s second move, with each program playing White 500,000 times. See what the win percentage for White is. If you do this for one opening, you know whether it’s an above-average or below-average opening for White (and vice versa for Black).
I can’t see why you couldn’t do this with all the major openings of significance, starting a million computer games from the position that defines that opening (e.g. 1 e4 e5 2 Nf3 Nc6 3 Bb5 for the Ruy, or 1.d4 Nf6 2. c4 e6 3. Nc3 Bb4 for the Nimzo-Indian) to see what the White win rate of each one was.
It would be interesting to see, for instance, which response to d4 had the best odds for Black - Queen’s Gambit Declined, King’s Indian, Nimzo-Indian, Grunfeld, Slav or Semi-Slav, or what.
What am I missing here?
As I said earlier, I’m a bit behind on current computing developments. But I see the main problems as:
- randomising the games (no point in repetition)
- avoiding the dross (how far ahead are you going to analyse each move?)
- checking if the opening was responsible for the result
Bear in mind that the top players all have computers with full opening databases on them and that they will run top programs overnight on key positions from their own (and their rivals) games.
And they still get difficult positions…
Randomizing: yeah, would need to do some of this. Can’t see that that’s a problem: if a program has a way of scoring different alternatives at each move, and picking the best one, you just have it do a random pick of the possibilities, with probability proportional to score (or score, squared, if that works better).
Avoiding the dross: we want a fair amount of speed, so we’ll have to limit the number of moves the program analyzes ahead at each move. That’ll mean we get some dross, but ISTM we can live with some of that.
Opening responsible for result: not individually. But the results of a million games between two top-flight chess programs, beginning with a particular opening, should give a fair degree of insight into the viability of that opening.
Sure, chess is like that. What I’d hope to find is that some openings have 58% chance of success for White, while others have only 52%. And some are rarely played anymore at high levels because they don’t even crack 50%. It goes without saying that an opening that White won with 58% of the time is still going to be no walk in the park, will present difficult positions, and is still far from a guaranteed win.
But there are pretty obvious reasons why you’d want to know these differences.
[QUOTE=glee]
The experiment above used around 1 million games between computers to confirm that White was slightly better.
It is a huge leap from that to analyse every possible opening move and them play many games from each position to establish opening theory.
[/QUOTE]
It won’t work because of how the problem was defined. Glee said “every possible opening move”, which is not equivalent to the problem you define. My post 49 quantifies something less than “every possible move”, and it is still far too hard to solve practically.
Say we define this as; run 1 million games for each ECO code (and pick whatever starting position you like from the ECO code). Then we have 500 ECO codes, and 500 million games to run; this is not outside the scope of practical computation.
Did you have some special knowledge when you wrote this? Because there are some interesting developments in the chess world:
The King’s Gambit is almost certainly a draw for White. White’s best move is B32, which draws; every other move is a loss.
The text initially claims that the opening is solved, though if you read further on it’s actually a statistical claim based on board scoring never getting too out of whack. Still, this is the kind of thing I had in mind when I made my suggestion, so it’s nice to see that people have been working on it.
Before we go any further into that, I want to note the date of that article: The next day after March 31.
Ha–I didn’t think about that. But pulling stunts the day after April Fools is highly frowned upon. Anyway, I don’t see anything immediately wrong with the article given that they didn’t claim to have a formal proof, but instead just put strong bounds on the strength of the resulting games. Plus, Bobby Fisher had already come to the same conclusions, but without the force of thousands of computers.
I don’t know about this, but in the old days during the New York Open International (a pretty big deal back in the 80’s. I’m sure you know that, but this part isn’t for you), at the end of a game that White won, the pieces would be reset but with the white king on e4, or the black one on e5 if Black won, or the two of them on e4 and e5 if the game was a draw. Simply for the benefit of the spectators who missed the end so they wouldn’t to have to walk to where the crosstable was posted.
True, but I think they covered themselves in the intro, where they relate that the programmer moved to Budapest March 31 and agreed to an interview “the next day”.
Excellent point! I may have been suckered. Reading around on the web, I see that a few others also think it’s a joke, but many sites take the claim seriously. The claim is currently still in the Wikipedia article on the subject, though the talk page also mentions that it may be a prank.
Well, I eagerly await glee (or other strong player’s) opinion here. I don’t see any obvious errors in the computer aspects, but it may be that there are some chess-related hints that are over my head.
The article doesn’t really have any holes in it. It’s not like they claim to have powered the tests with hamster wheels. The thing about the King’s Gambit is that it’s long been thought to be inaccurate. In the old days, material wasn’t that important. They’d throw the kitchen sink at the barricades and try to slip a piece through. Some call this the “Romantic Period” of chess, and the KG is the hallmark of that era. Indeed, it’s the opening from the most famous game in history.
But today we know better. Material is important. And so this edgy opening has had its reputation tarnished. To call it solved would not be a revelation, but a confirmation, and so I doubt it’s an April Fool’s Day prank.
I’ll go on record as calling “prank”. April Fool stories are something of a specialty at the chessbase site, and they put lot of work into them. Of course, it’s pretty hard to fool anybody these days, so usually they run an odd-but-true story along with a fabrication or two, and the game is to guess which is which. This time, they just varied the game a bit.
No, just a non-deterministic Turing Machine program.
[QUOTE=ChessBase]
Our algorithm works in an iterative manner – it first forms a hypothesis, and then it confirms or alters that hypothesis over a number of passes using a non-deterministic Turing Machine program running across the clusters.
[/QUOTE]
Upon reflection, that line is probably a giveaway. It’s not that such a thing is impossible, it’s that no programmer would ever word it that way. Every computer is (within certain limits) a Turing machine, so saying so is redundant. And non-deterministic is an overly fancy way of saying random or stochastic. Calling it a “non-deterministic Turing Machine” sounds ridiculous.
Honestly, what tipped me off was the supposedly huge savings from truncating the search at an advantage of 5.12 . Two problems with that: First, by the time you get to an advantage that extreme, the game’s not going to last all that much longer, and it’s going to be easy to find winning moves (you don’t need to find all of the moves the winning player has available, just one of them). You should be able to just force trades until you find yourself in an endgame tablebase. Second, the number 5.12 seems unjustifiably precise: Why that particular number, instead of just calling it 5, or 6? Surely a 5-point advantage is nearly as definitive as a 5.12 point advantage.
It’s hard to say for sure without knowing exactly what that number represents. My understanding is that it’s pawn equivalents–pawn=1, knight/bishop=3, rook=5, and queen=9. The extra fraction (which sounds precise, but might just be a rounded 1/8 point) might just be to require some extra positional strength on top of being (say) just a rook down. Also, they had to pick something. 5.0 is almost as definitive as 5.12, but then 4.9 is almost as good as 5.0, and so on down. You have to pick some kind of arbitrary threshold.
Anyway, even assuming the article is a fake, I think the general idea seems valid. I’ve no idea if the computing power is feasible for a 5.12 point difference, but it’s clear that there’s some difference for which it is feasible–maybe it’s some uninterestingly low number like 1.0, but such a number must exist.