|
||||||||||||||||
*
There is a book in my room. Like an uninvited guest, every morning
half past six it is
Some reviewers got mad about the exotic concoction of frivolous variations, home-cooked poetry and Dylaneske citations. In possession of free copies of New-In-Chess Yearbooks after acceptance of two Letters-to-the-Editor (nr.82, page 22, nr. 87, page 19) on the subject of MAMS-computer chess and seeing my name emerge from anonymity amidst those of high ranked grandmasters, we rearmed ourselves with a 2-Ghz processor, a Fritz-10 CD with book and immediately embarked on a sequel. Here it is! by Albert
H. Alberts This is the cover:
The main reason is that the majority of tournament players play in the tradition of the notorious *small-but-enduring advantage* primarily in the opening phase. Usually one waits for weaker human continuations - or two or three of those - from the opponent. The key issue in tournament chess is to prove that the adversary somewhere along the line has made a mistake. If so, one proves it, if not, the game is a draw and draws abound in tournaments. These people play for money too, you know! They have to bring at least some bacon home in one piece, worry about invitations/reputations and ELO-ratings. BUT: the small-but-enduring advantage principle is ineffective in computer chess. In fact: it is a sin! Almost every solid and sound position with a small advantage within the draw margin can be defended without effort by the modern programs. Playing computers, one has to get restless if one is NOT at a disadvantage say around move 10. Don't just sit there! Do something. As in HTTF-1, in HTTF-2 I maintain the no-risk no-win principle. Alekhine's Parrot correctly noted that we are talking CALCULATED risk. Man-Assisted-Machine Chess is played on the compass of QUANTITATIVE numerically assessed risk calculation.
The classic qualitative judgment
of positions, White-is-slightly-better, Black-is-good, etc., arising from
sort of a carpenter's eye judgment or *gut feeling*, is replaced by a scale
of numbers. From 0 to + 0.5 White has moderate advantage, from +0.5-+1
White is good, between +1 -+2 very good and over +2 on the way to a win. Over +3 Black can resign although I have seen stubborn people like Kramnik continue to play on even over +4. Michael Tal played positions for Black from as far as suicidal -5/-6 and still was able to draw them in a tournament setting! But all that has to do with a pair of diabolical eyes distracting the opponent, time pressure, temporary blindness and other typical human factors. Against a machine, Tal would not have gotten away with Swindles of this magnitude. In the process to estimate the risk margins for white and for black in HTTF-2, I developed two rules-of-thumb:
However, no brute-force piece sacrifices before move 8 for White (the Cochran Gambit 1.e4 e5 2.Nf3 Nf6 3.Ne5 d6 4.Nf7!? nor the Leipziger Razzle Dazzle Variation 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Ne5!? or similar philosophies are good enough. It is better to design two or more risk-moves to unbalance a position repeatedly. We will get to an example shortly.
Now here is the trick: Suppose you play White, you are at an advantage of +0.5 in the opening phase. You spot 2 risk-moves, generally long term positional sacrificial moves; one will set you back at say -0.5, the other one at -1.5. The machine in principle has two options: it can play the response that sets you back at -0.5 and the response setting you back at -1.5. I think we all agree that a machine will ALWAYS choose the second option, gathering maximum advantage. BUT. The human task in this procedure is to note that the apparent large advantage will lead to disaster in the long run. The disaster in a later stage invariably arises from the creation of more PIECE MOBILITY for the risk-taking player. In human terms: by playing on the progression of the -1.5 (dis)-advantage we let the machine *think* it can win! Cautious human players, in Karpov or Euwe tradition, will smell a rat and opt for the more modest advantage of -0.5, but a computer can not smell rats BECAUSE IT HAS NO NOSE! That is how the main Fooling Fritz procedure works. One CAMOUFLAGES on his/her ship more piece mobility behind an optically weak position. One more rule of thumb before we proceed: ALL pieces of the risk-player can move or will be able to move on-call. And one never castles to bring the king to safety, but one castles to move a rook with gain of tempo to an open file.
The Keres Attack. Open Sicilian, Scheveningen Variation.
I switched off the book after 12...Qc7 13.Qd3 because both 13...dc3 and 13...de3 are prescribed from the literature. Without book Fritz takes the c3-knight but my human intuition tells me 13...de3 might be the better one. Is it?
Question is: did Keres know that taking out the bishop on e3 (and eliminating Bb6) is superior to bc3? I suspect he did. But he refused to play it safe with the advised 10.gf6 or 12.Bd4. Just keep a straight poker face after 13...de3 and gamble on the circumstance that the man-in-black cannot control all the variables of the conspiracy threatening the black king. The numerical evaluation of 15.Bf7!? goes way over -3 (in fact -3.87!) so this sacrifice is a Swindle born out of desperation. The excessive disadvantage will NOT reverse to + for White, in fact, it will be stuck at around -1 and Black can play for a win. The Keres Attack is refuted after 13...de3! Alas! Now for the standard computer-modes-of-defense. The machines have several tricks up their sleeves: FIRST: once the apparent disadvantage has tilted to an advantage for the human player, the program will try to steer to a ROOK ENDING. In practice rook endings are tricky, and so they are in computer chess. A healthy, say +1.5, advantage after queen exchange usually is enough for a win, but when the light pieces are gone this margin goes up. NEVER enter a rook ending below +2. And even then the master defender will sometimes succeed in deliberately collapsing your apparently overwhelming advantage to a draw! Here is an example from the Caro-Kann complex:
More Houdini-like escape tricks: SECOND: the machine will try to force endings with bishops running on unequal colors. THIRD: in case the human player sacrifices two or three pawns in the opening, the program at some point will snatch a third or fourth pawn for a light piece and draw the game that way. FOURTH: when worst comes to worst it will exchange the queen for pieces. Controlling a large number of variables is difficult for a human brain, for a machine it is not. FIFTH: a machine will switch on a MATEFINDER algorithm when its king is attacked via sacrificial attack on a castle. In some cases the king walks unharmed to the middle of the board! This can lead to hilarious situations: I call this scenario: DEAD MAN. [Based on the title of a classic Jim Jarmusch film. An innocent office clerk - Johnny Depp - dwelling in illiterate cowboy/Indian territory in the late 19th century with a bullet, caught in his chest, in a brothel bed with a pretty girl. All this within the very first 30 minutes after he got off the train, handing over letters of recommendation to his boss - Robert Mitchum!] Spasski-Bronstein King's Gambit. Those were the days.
To close out let me respond to the recent highly picturesque African Proverb-annotated game Bronstein-Petrosian by Brian Wall, with a saying in that tradition:
It is dead!? What do you
mean by: dead!?
Bronstein seemed completely dead
in this game. Not just a little bit. Fritz revived Him! Albert
Alberts Albert Alberts' Explorations in Man-Assisted Machine Chess
|
Chessville
The
Advertise to Single insert:
|
|||||||||||||||
|
||||||||||||||||
