Greengard's ChessNinja.com

Hydra vs Adams

| Permalink | 31 comments

I heard about this a month ago but either forgot about it or preferred to wait until the event was closer. But now it's in the media (here, here, and here, with a cool promo photo of Adams at the bottom) and everyone is sending me links, so here we are, a month early. From June 21-26 English GM Mickey Adams will play a six-game match against the chess computer Hydra of the UAE in London. (BBC has a cute article here, with the writer playing a (horrible) game against Hydra.)

Hydra is a hardware-based system; that is, it's a not a software program that runs on a general-purpose CPU like Fritz and the rest. Hydra was formerly the ChessBase Brutus project. [Search here for Hydra for more on that.] It's not clear to me how much UAE technology is involved. They are using Xeon processors, FPGA cards and the programming wizardry of Austrian Chrilly Donninger, who created the program Nimzo. But it's a promising method for a sort of off-the-shelf supercomputer and the technology has potential beyond dominating chess. Kudos to the sponsors and organizers.

I do wish these press reports wouldn't act as though we don't already have computers much faster than this thing. They are just very, very expensive and not specialized in chess. There also are other clustered machines that are among the fastest in the world.

There is little doubt that Hydra can beat the top PC programs. Search depth is critical in comp-comp games. But a few million more moves here or there isn't decisive against a human, where the type of position out of the opening is everything. (That and human blunders, which any computer will punish instantly.)

One thing that is long past due is that the Hydra team seem content to let their monster off its leash very early instead of getting into an opening book war. At least that has been their policy against other computers. There are the usual errors in the media reports (Adams #3 in 2002? Can boil an egg?), and also some confusion. The Hydra site, which continues its strange tradition of copy-pasting content from other sites, says the new version uses 32 FPGA cards. The press reports, no doubt all based on the same press release, say 64. The Times story says the upgrade to 64 will come later in the year.

Another number they all latch on to is that Adams sees "1.4 positions per second." The old made-up number, much used in the Kasparov matches, was 2-3. As if we really know, or could ever know. The human brain doesn't work in convenient numbers like nodes per second when it comes to chess thinking.


There are no break periods so Adams has a chance!
Kidding apart, do we have any records of Adams games vs computers?


Ooh, I'm terribly excited about this you know, this is the most important chess event to have been held in London since 2000. I just hope Mickey doesn't play like he has done recently, i.e. building up massive advantages and then throwing it all away. God, this is going to be frickin' excellent...

Boil an egg I can believe, but make tea? Come on, he's super-GM, not a super-human.

Anyone know how we go about getting tickets for the event? Can't see anything on the official Hydra site...

I hope Adams will have plenty of time to make moves in this match.
(I mean no quick finish, a reasonable time increment, in a word all to put Adams in a confortable situation where he can give in games his real best)

I mean, the interest in this kind of matches is to know who can play BETTER CHESS, a top human GM or a machine.

If we REALLY want to know this, we need always to set the optimal time control for the human side.
(not time advantage for human, the same for both but choosing the optimal for the human brain)

OK do not reply me 2 hours for one move please , I mean the optimum a human can play a whole chess game without any interruption.
May be something like 1h plus a increment 3-4 minutes for a move or similar would be good.

This becouse the human's brain chess approach is more sensitive to the time than machines.

The level difference between a Kasparov's move after 3 minutes or after 30 seconds could often be huge: the GM could even blunder or, in more common situations, if he has time he often plays a move who improves his original planned choose.

Instead if the move is from a machine, in the most cases, the real value of the move is nearly the same (most of times the move choosen by machine is even the same in both cases)

So, I mean, is an error to set the clock in human-machine in the same way of the human-human games. That time fits becouse the two humans are in the same conditions and not becouse that is the optimal for their brains.

So I would like to see human and machine in the same condition in the match, the best possible for both...
The human likes more time, play the game at the time control he likes as the machine likes 2 CPU, tablebases and all the chess history on its hard disk ok give it all, no problem we want to know the true answer to the question WHO CAN PLAY BETTER CHESS, MAN OR MACHINE ?...
But please don't let Adams to play a beautiful position faster than he needs becouse we like the blood of the spectacular zeitnots.

Engines are still much weaker compared to a human at long time control.
Just few days ago ICCF GM Arno Nickel won in his corr. match vs. Hydra with score of 3:1. And if you look at games, the score should be 4:0. His only loss was on time in very promising position.

Human positions per second can be quantified, I claim, in a roundabout manner. Mig is right that human "PPS" is not directly quantifiable since we don't really understand yet how to think about how we think... so define Equivalent Positions Per Second to be the PPS of the computer a given player can draw a match with.

This gives us our due credit for having a good tree-pruning algorithm in our rather impressive onboard computers.

Please see my short interview with Nasir Ali of team Hydra, at the end of my review of the GAME OVER DVD, at ChessCafe. I think his answer to question #1 was the most interesting. I asked how Hydra's evaluation function compared to that of DB's. Supposedly (from their website) Hydra "only" calculates at about 40M pos/s.


"Engines are still much weaker compared to a human at long time control. Just few days ago ICCF GM Arno Nickel won in his corr. match vs. Hydra with score of 3:1."

You referred to a correspondence match. Have there been enough human vs. computer corr. matches from which we could draw such conclusions?

In over-the-board competition at classical time controls, I would disupte the statement that computers are "much weaker" than humans. Indeed, part of the reasons such matches continue to be interesting, is that the outcome is in doubt. Just ask Garry Kasparov or Vladimir Kramnik whether computers are "much weaker" opposition. The results haven't borne that out.

What *is* true, is that computers are poorer strategists, but superior tacticians. Computers, therefore, come with an opening book geared to maximize tactical complexity, while the human players try to steer into positional openings. It's also true, as Mig noted, that computers punish blunders instantly. It seems to me that a fair assessment needs to recognize computers' weaknesses, but also give credit to computers' strengths---just as you'd do when assessing a human player.

My basic postulate, written years ago and still roughly true, is that the top computers play 2400-2500 level chess but NEVER BLUNDER. That's huge, and enough to play at a 2800 level on a results basis against humans. That baseline level of play is improving slowly, but only slowly. Even if Hydra has pushed it up to 2600 it's an opponent that a top-10 GM like Adams would outscore consistently if he too never blundered. There are precious few examples of elite players losing classical time control games to computers without a ?? move somewhere near resignation.

I completely disagree with Mr. Fatland in his post above. Assigning a number of positions per second that a human can see by equalling it to the positions per second seen by a computer of roughly equivalent strength still doesn't make any sense because humans simply don't think in this type of manner! How many positions do you see when you think "those doubled isolated pawns on c3 and c4 will be severely punished when I dissipate my opponent's weak-sauce attack."?

Ashton Anderson

"There are precious few examples of elite players losing classical time control games to computers without a ?? move somewhere near resignation."

Ok, but what about all the draws? Kramnik and Kasparov would hardly let a 2400-2500 player get a draw that easily. And almost all of those games are without blunders.

Remove the human blunders from Kasparov-X3D Fritz and Kramnik-Deep Fritz and the computers would still get a performance rating way above that.

This made the Metro today, the UKs free newspaper (in the big cities). Even had a photo of Adams!!!!

I reckon that for every 1000 people who have heard of Nigel Short in Britain, only 1 has heard of Adams. If that! So it's all good...

hi, first, off topic but great piece by Jeff Sonas about greatest ever up on Chessbase. first three were boring, but this part really captures the essence of his analysis.
second, Mig, great introductory piece to this post. I rarely see mention of Hydra without some BS about FPGAs or something, which is quite annoying. Your post cuts through most of the crap and gets to the crux of the matter.

Wait, he lost on time in a correspondence match?

"1.4 positions per second"

Refutation: In a bullet game (1 min. o inc) between 2 GMs, a game that lasted 60 moves would be made out of absolute random chess moves. It can easily be proved ( I hope :) ), that the moves played have a lot of purpose, or else everybody would have the same rating in icc for bullet.

"Ok, but what about all the draws? Kramnik and Kasparov would hardly let a 2400-2500 player get a draw that easily. And almost all of those games are without blunders."

I wonder whether the draws are the result of the human player deliberately steering clear of tactically complex positions when facing a computer. Against a 2500-rated human, Adams would probably be delighted to invite complications, because he knows his opponent can't out-calculate him, and will probably blunder before he does.

Does Adams have access to hydra before the match ( a la Kramnik)? I guees not, since hydra is hardware based...

I truly believe Hydra will win the match.
Anybody else?


I'm also rooting for the machine here.

I am pretty sure 2400-2500 is exaggerated but maybe 2600-2700 after all I recall Ilya Smirin easily taking down a computer in a match once. But my question is about the correspondence matches--- I presume that the match allowed for the GM to make use of a computer or not?

It seems that the top humans blunder against computers more frequently than they do against other humans. Isn't it possible that they can cut done on their blunders and start winning some of these matches?

Juan: Why in hell would you pull for the computer?

"The computer" is just a way to say it. I'm rooting for the team that makes Hydra, all of them humans i suppose.

Nigel Short once (maybe about 10 years back) said something like: "I have done some work (analysis) with most of the top players including Kasparov, but Vishy really has the fastest version (!). While the rest of us are just absorbing the characteristics of a position, Vishy is already 8 moves down the mainline."

I wonder how many positions the computer need to see to reach 'eight moves down the mainline'? Since every program would have to use brute force and hence consider every legal move in every position it reached...most of the positions it sees would be truly ridiculous..

It would be interesting if Adams does not get psyched by the numbers showcasing the computers calculating ability and play his natural game.

Another quote:
"If only some one had convinced/told him (Kasparov) that I could see 2 million positions per second..." Anand commenting on one of the Kasparov - Deep Blue matches (first one?) where Kasparov played a lot of 1)d3 kind of stuff...

It would also be very interesting to see how 'Drawcula' (Leko) would fare against the super aggressive Hydra...or maybe...we can have a short survey of opinions here - Which of the following players would do best against a machine:

a) Topalov (would he sacrifice and win)?
b) Morozevich (1)d4 d5 2)c4 Nc6/e5)?
c) Leko (will he be able to suck the initiative (life) out of the positions)?
d) Anand (super-classical-fast-versatile)

e) Kramnik?? - nah....he blunders too much, too often

Ok then, why would you pull for the Hydra team?

You should see the HYDRA "documentary":
That was a barrel of laughs... "Destroy the bishop!"

(Download is very slow, so I'd "Right Click, Save (Link) As".)

not that I want to hijack the thread but Spassky and Lautier landed in Reykjavik today, along with business man Alex Titomirov. They met with Fischer today in what was supposed to be a secret visit but word spread to the media. Apparently Titomirov, who is a wealthy russian businessman, is interested in sponsoring a(fischer random) match between Fischer and anonymous top rated GM.

I was trawling the Web for Adams stuff and came across the following funny (perhaps apocryphal) tidbit:

At nine, he entered the Cornish championships at under-11, under-14 and under-18 level, moving from room to room so he could play three games simultaneously. He won all three titles - on the same day.

Vlad, the correspondence games were with the "Chimera" version of Hydra, an old weak version.

Adams has a rapid match against Leko coming up before this, starting June 2nd. As signature "azaris" on chessgames.com put it:

"He has to face a cold, heartless, machine that won't give him an inch of remorse and will grind down any mistakes he makes on the board. And then he'll play Hydra!"

heh.. good one..

In response to previous post, the tickets for the event are free, and there's no need to book - you can just turn up on the day.

There are games at 3pm every day - 21, 22, 23, 25, 26, 27, June at Wembley Convention Centre, in the Greenwich room.

Twitter Updates

    Follow me on Twitter



    About this Entry

    This page contains a single entry by Mig published on May 25, 2005 4:52 AM.

    2006 US Championship Site was the previous entry in this blog.

    Post-Sofia Kramnik Interview is the next entry in this blog.

    Find recent content on the main index or look in the archives to find all content.