Kasparov: Chess is the Drosophila of reasoning

by Kasparov Chess Foundation
12/13/2018 – The Kasparov Chess Foundation (KCF) and The Chess Club and Scholastic Center of Saint Louis (CCSCSL) have run the very successful Young Stars program for 'Team USA' for the past six years. Meet the latest crop of talented kids from the recent New York training session. Plus, In a new article, Garry Kasparov reminds us that the recent World Championship match did not represent "a contest between the two strongest chess players on the planet, only the strongest humans." Brute force programs have long exceeded human playing strength. But that does not end the historical role of chess as a laboratory of cognition. A new pure AI system has superseded even those programs. Kasparov evaluates AlphaZero's play in Science Magazine and gives a political interview in the prestigious magazine The New Yorker. | Photo DeepMind Technologies, Igor Khodzinskiy

ChessBase 15 - Mega package ChessBase 15 - Mega package

Find the right combination! ChessBase 15 program + new Mega Database 2020 with 8 million games and more than 80,000 master analyses. Plus ChessBase Magazine (DVD + magazine) and CB Premium membership for 1 year!

More...

The latest "young stars"

The Young Stars program in the USA has helped develop several American chess prodigies from across the country, including GMs Jeffery Xiong and GM Sam Sevian. Students have been treated to individualized training sessions with Garry Kasparov’s coaching staff and special training sessions with Kasparov himself. Below are photos from the latest instalment of the Global Young Stars program, and you can find applications for future sessions here.

The most recent Young Stars camp brought forward a talented new generation. This KCF camp was the first serious training session for some of the players in attendance, and it gives them new opportunities to jump-start their game. Held at the KCF offices in Manhattan, students came from as far as California and even Russia! We anticipate more students from other continents to be a part of the program in future editions.

Mishra and Chasin

World’s number one nine-year-old player Abhimanyu Mishra from New Jersey and 2018 World Cadets Championship under-12 bronze medalist Nico Chasin from New York 

Khodarkovsky and Kasparov

Kasparov Chess Foundation director and FIDE Vice President Michael Khodarkovsky along with Garry Kasparov

Group photo with Kasparov

This year’s participants: Left: Rochelle Wu (California), Abhimanyu Mishra (New Jersey), Christopher Yoo (California), Brandon Jacobson, (New Jersey) Right: Anastassia Matus (Minnesota), Nico Chasin (New York), Bibisara Assaubayeva (Russia), Andrew Hong (California) Center: Garry Kasparov and Michael Khodarkovsky


Kasparov's other recent moves

by Frederic Friedel

In 1997 Kasparov famously lost a rematch against IBM's Deep Blue. There were a few human-computer matches after that, but soon "the short window of human-machine chess competition slammed shut forever" (Kasparov). This was due to machines getting faster and programs better. A modern day smartphone is clearly stronger than Deep Blue was. And to the dismay of the Artificial Intelligence community, it became clear that it was not the emulation of human pattern recognition and visualisation that produce a machine that could challenge the world champion. Brute force and clever programming tricks were enough.

Then came AlphaZero, a neural network program developed by the Google company DeepMind. It required no human help, but it generated its own chess knowledge by playing more games against itself than have been recorded in human chess history, as Kasparov notes. In doing so AlphaZero became stronger than any chess engine — and by extension any human player.

So how does the legendary eleventh World Champion view this development, how does he evaluate AlphaZero's play? Kasparov:

I admit that I was pleased to see that AlphaZero had a dynamic, open style like my own. The conventional wisdom was that machines would approach perfection with endless dry maneuvering, usually leading to drawn games. But in my observation, AlphaZero prioritizes piece activity over material, preferring positions that to my eye looked risky and aggressive. Programs usually reflect priorities and prejudices of programmers, but because AlphaZero programs itself, I would say that its style reflects the truth.


There is another piece with Garry Kasparov that has caught international attention. It appeared in the prestigious magazine The New Yorker which is famous for its in-depth reporting.

The piece is a dozen pages long and comes in the form of an interview conducted by Masha Gessen, a visiting professor at Amherst College and the recipient of numerous fellowships and awards (most recently won the National Book Award). After more than twenty years as a journalist and editor in Moscow, Gessen has been living in New York since 2013. As a prelude to her interview with Kasparov she writes:

Born in Baku, the capital of Soviet Azerbaijan, to an Armenian mother and a Jewish father, Kasparov fled that city when anti-Armenian pogroms broke out in 1990. He was twenty-seven years old and had held the world-champion title in chess for five years; he was famous and, by Soviet standards, wealthy. He chartered a plane to Moscow and took nearly seventy people with him. Kasparov announced his retirement from chess in 2005, when he was still ranked No. 1 in the world, and declared that he would devote himself to politics. He started a movement called The Other Russia, a broad coalition united in its opposition to President Vladimir Putin. After a series of street protests and a failed attempt to put Kasparov on the ballot for the 2008 Presidential election, the movement sputtered along until the mass protests of 2011–12, in which the movement’s activists—and Kasparov personally—played a key role. In the political crackdown that followed the protests, Kasparov was forced to leave Russia. He now lives in Manhattan.

Kasparov sat down with Gessen for the interview in a coffee shop in Manhattan. You can read it on this New Yorker page, which is free and not posted behind a paywall.




The vision of the Kasparov Chess Foundation is primarily to use the many benefits of chess to help children worldwide. KCF does this through a variety of different programs, which you can find at KasparovChessFoundation.org
Discussion and Feedback Join the public discussion or submit your feedback to the editors


Discuss

Rules for reader comments

 
 

Not registered yet? Register

Frederic Frederic 12/16/2018 10:58
@SermadShah: "Just disable computer's opening book reference and see how it plays without it. You would find defects in strategy. Note and then use them in middle game to beat computers." More than a decade ago there was a player who pretended to be Bobby Fischer on Playchess (and convinced a couple of GM opponents that he was indeed Fischer). He played stuff like 1.f3 and 2.Kf2 and still crushed GMs in blitz. Perhaps it would be worth a try, with today's 3500 Elo engines: don't just disable the openings book, make the computer play rubbish for a start. I predict they will still win.
Keshava Keshava 12/16/2018 10:40
@SermadShah, another site has tried matches with top GM's like Nakamura vs one of the top engines without the opening book. The engine will find 'good enough' moves to get to an equal middlegame. Capablanca played the openings that way also.
SermadShah SermadShah 12/15/2018 05:44
I believe Humans can still beat computers.
Just disable computer's opening book reference and see how it plays without it.
You would find defects in strategy. Note and then use them in middle game to beat computers.
Note: I did not tried it because I am a novice player (1800).
Keshava Keshava 12/15/2018 04:29
@Frederic,
re: "With regard to the diminishing returns displayed in NN training, there is a reason for that (which I will reveal, at some stage, in an article on our news page). In fact I believe that no chess program can go higher than about 4000 Elo. You guys can guess why."
o.k. I assume the point of the shogi programmers was that Alpha Zero is slowing down in it's shogi improvement and the human programmers feel that they will catch it (and I presume surpass it). This is interesting to me regardless of if their is an upper limit to shogi level. So assuming your 4000 elo limit for chess - do you think that human programmers will surpass the level of Alpha Zero before either of them hit that limit? Using the hammering of SF 8 as the control I understand a program can already score better against SF8 than Alpha Zero did in their 1000 game test.
Petrarlsen Petrarlsen 12/14/2018 07:39
@ rasagyan:

"Engines running on quantum computing will cross Elo 4000 and much more, whenever they become ready, hopefully in not too long."

Frederic's core idea (explained in the "How God plays chess" article he cites) is linked to chess itself, and not to the possibilities of computers.
Frederic Frederic 12/14/2018 04:58
@dumkof: There are 10¹²⁸ different possible sequences of moves (until move 40), but there are "only" 10^35 different positions in all of chess. That is vaguely analogous to there being many more different stories than words in a language. In both cases chess is practically infinite. I wrote about this in this article: https://en.chessbase.com/post/how-god-plays-chess.
rasagyan rasagyan 12/14/2018 04:54
@Frederic: Engines running on quantum computing will cross Elo 4000 and much more, whenever they become ready, hopefully in not too long.
Marselos Marselos 12/14/2018 08:25
@dumkof
ἄτομος, (atom). Ah, ah,ah it should be indivisible.
"In reality we do not know anything, because the truth lies deeply."
Democritus.
fgkdjlkag fgkdjlkag 12/14/2018 06:28
Obviously there is an upper limit on the attainment of elo for chess, and there will be diminishing returns in approaching that.

4 hours is very misleading because it is based on substantial computing power; this was pointed out by GM Anand. Keshava mentions $25 million but I wonder if that expense on hardware for stockfish would not be comparable because no doubt the alphazero algorithms are designed for TPUs.

If it's true that stockfish 10 would beat stockfish 8 by a bigger margin than alphazero-stockfish 8, then that is very interesting. But also the 155 wins by alphazero was partially over a handicapped stockfish because stockfish did not have access to its opening book in all the games, while alphazero had access to its millions of self-play games in its training.
truthadjustr truthadjustr 12/14/2018 05:47
Sometimes, I felt that the reasons used in chess is not of human. We humans have certain foundational primitives of which becomes the basis of our reasons. Most obvious of which could be space and control. But sometimes, you see a computer chess move going against these primitives in order to find the more optimal. Thus, begets the first question: what is reason for? And who is the owner of reason?
dumkof dumkof 12/14/2018 03:42
@Marselos,

Chess will never be solved. There are 10^120 possible moves in chess, which is far above the number of atoms in the observable universe. Nothing can store and compute such an enormous amount of data. It's above universal capabilities. Most probably, we won't even see a complete 8 piece endgame tablebase until the end of this century. Not to mention that chess is an endgame tablebase of 32 pieces.

Your "hypothesis" is nonsense, and so are your "consequence" statements A, B and C.
Marselos Marselos 12/13/2018 10:51
Hypothesis: a normal PC solves chess (XXII century, end XXI)
Consequences:
A-The game of chess lose charm and mystery.
B- (first phase): professionals use memory techniques.
It is a memory match. Sponsors fast lose interest.
C-(last phase): Return to the origin,only amateurs play.

That's our road. Chess will be only a game.
If the FischerRandom does not change the substance. The reasoning remains.

  I immediatly should stop the development of chess engines.
Frederic Frederic 12/13/2018 09:49
@IntensityInsanity: There is a reason I believe that no AI or Brute Force program will cross 4000 Elo, at least not substantially. There is a specific reason for that, which I discussed with the Deep Mind people. They agreed that my reasoning was plausible. I am going to keep further elaboration for an article I am planning for later this month. Sorry for being enigmatic. Incidentally can you elaborate on "IntensityInsanity"?
IntensityInsanity IntensityInsanity 12/13/2018 05:13
Hi Frederic, can you please elaborate on your last sentence, "In fact I believe that no chess program can go higher than about 4000 Elo. You guys can guess why." I'm not sure I can guess (or I am no sure my guess is correct).

Thanks!
Frederic Frederic 12/13/2018 02:04
@Keshava: "Yes, it is exciting if it can keep making improvements." Look, I gave this six-year-old boy a chess set and showed him the rules -- how the pieces move. Then left him to figure out the game all by himself. After a week he said: "I can play now." So I invited him to play a match against my good friend Vishy Anand. The boy won. Everyone was excited, but some people said: "Okay, but can he beat Carlsen? And how do we know he will keep getting stronger?" With regard to the diminishing returns displayed in NN training, there is a reason for that (which I will reveal, at some stage, in an article on our news page). In fact I believe that no chess program can go higher than about 4000 Elo. You guys can guess why.
Keshava Keshava 12/13/2018 01:30
@Frederic: Yes, it is exciting if it can keep making improvements. The reason I wonder about this is that I was reading that Alpha Zero may have a problem with diminishing returns in shogi. "Motohiro Isozaki, the author of YaneuraOu, noted that although AlphaZero did comprehensively beat elmo, the rating of AlphaZero in shogi stopped growing at a point which is at most 100~200 higher than elmo. This gap is not that high, and elmo and other shogi software should be able to catch up in 1~2 years." https://en.wikipedia.org/wiki/AlphaZero#cite_note-18
Of course in chess the engines make radical improvements very quickly with a big gap in the estimated rating between SF8 and SF10. Can the neural network improve at as fast a pace? That will be interesting!
Frederic Frederic 12/13/2018 11:32
@Keshava: The news that excited experts was that a neural network that was trained by playing 44 million games against itself, in a few hours, with ZERO human intervention (therefore the name), without a specialized openings book (A0: "What should I play? 1.e4 looks pretty good"), beats one of the strongest traditional brute force programs and could thus be plausibly considered the strongest chess playing entity in history. That is what everyone was celebrating. It did not take decades of hand programming, just four hours of self-play. Don't you feel a shiver? Now comes the task of determining if A0 is consistently stronger than all other programs, how much stronger, and under what precise conditions that is the case. This is something we are currently undertaking.
Keshava Keshava 12/13/2018 10:18
re: "A new pure AI system has superseded even those programs."
I'm sorry but that statement does not have an empirical basis. It may indeed be the case that a pure AI system can today supersede the best chess programs but without scientific controls we do not yet know.
Why not encourage a team of independent experts to determine what hardware is equivalent to the $25 million dollar hardware used by Alpha Zero in it's tests vs. Stockfish 8 (I have heard that Stockfish 10 can score better against Stockfish 8 than Alpha Zero did in it's 1000 game test). Were the stellar results of the Hydra chess machine in the early 2000's due to superior software or architecture? Without proper controls we are setting this point in history that a pure AI system is surpassing human chess programming efforts when in fact it may be a time in the future. When did chess playing computers surpass the best player in the world? Was it the year when Deep Blue defeated Kasparov in their match? I think the time actually came later (because of the lack of transparency of IBM). However it is easier to just go with the prevailing narrative than ask the hard questions that the philosophy of science demands.
Frederic Frederic 12/13/2018 09:53
@goeland: That's what Google is for. Garry himself explained it in the article: "Much as the Drosophila melanogaster fruit fly became a model organism for geneticists, chess became a Drosophila of reasoning.” Drosophilae are small pale yellow or reddish brown flies, with red eyes. Leave a fruit to decay in your kitchen and they will somehow appear. Different drosophila species have been extensively used as model organisms in genetics, and their genomes, which are fairly simple, have been fully sequenced. The same applies to artificial intelligence: chess is a restricted model where scientist can experiment.
Petrarlsen Petrarlsen 12/13/2018 09:38
@ goeland:

Yes, at first view, this "drosophila of reasoning" comparison can appear as being slightly mystifying.

But if you follow the "Read the full article in Science Magazine" link on this page (which gives access to the original Kasparov article), you will have Kasparov's explanations on this comparison, and, in my opinion, it is in fact a quite interesting comparison.
goeland goeland 12/13/2018 08:32
The drosophila of reasoning ?
When Gary played chess, I switched on an engine to understand,
Now, when he speaks English, I switch on Google to understand
1