Adams vs Hydra: Man 0.5 – Machine 5.5

Subject: Rise of the Machines

The Adams-Hydra match signals the approaching end of man-machine contests. Already, last year's event in Bilbao was a sign that things were looking bleak for the humans. In Bilbao, it was not so much the performance of Hydra that was so impressive, but Fritz's score of 3.5/4 against Ponomariov, Topalov and Kariakin (twice). Hydra, which made the same score, was running on its special-purpose hardware but Fritz was running on a laptop computer from the local department store.

John Nunn, Grandmaster, author, computer expert

Mickey was always going to have a tough job against Hydra, and before the match I thought he would make 1.5 points. In the event he fared worse than I expected. Hydra proved that if you have enough computing power, you can play very well not only in wide-open positions but also in quiet, semi-closed positions.

There has been some criticism of Mickey for poor preparation, which has been compared unfavourably with that of Kasparov and Kramnik in their man-machine contests. But there is a big difference. Kramnik, for example, insisted on being given a copy of the program he was to play several months before the match. Of course, this makes it much easier to prepare. Hydra hasn't played very much chess, and as it is a machine rather than a program, Mickey could not have a 'copy' to put under the microscope. Thus Mickey's task was much harder.

I really can't see much point in further man-machine contests under the present rules (in which the computer is allowed an unlimited opening book and access to endgame tablebases). However, even changing the rules would probably only delay the inevitable dominance of the machines. Let's get back to humans playing humans, which I for one find more interesting than man-machine contests.

John Nunn

Following Hydra's crushing victory over Micky Adams, I would like to add something to John Nunn's [previous] comments on what will now happen in human vs computer Chess.

Firstly, I feel that John's estimate of when his son's Logo brick will defeat a strong Grandmaster is somewhat optimistic – I believe that it will be several years later than John does before a "simple" (at that point in time) microprocessor has the necessary capability.

But be that as it may, what is more important for the Chess world is the question: Where does man vs machine Chess go from here? How can we continue to create interest in man vs machine matches? Is there any point in future contests of this ilk?

I am convinced that man vs machine Chess still has a long future, full of human interest. First, of course, we must hope that the Pal Group can somehow convince Kasparov that he should play a match against Hydra, a match for which there should be a rematch clause in the contract. This would truly be the mother of all Chess matches – the strongest ever human player against the strongest ever computer. The Chess world would be agog.

But no matter what the result, and no matter what the result of a rematch, we are clearly facing, very soon, a situation in which man vs machine Chess, as we currently know it, is no longer of any spectator interest, because soon the time will come, if it has not already arrived, when the gladiator will always be eaten by the lion. What then?

In my opinion the answer is simple – odds games. When the strongest human players have no chance at even games, let us give the human pawn odds. At the present time this would allow the very strongest human players to make a plus score against the programs, but this could perhaps be mitigated by speeding up the games. There is, undoubtedly, some rate of play, whether it is an average of 2 minutes per move, or 1 minute, or 30 seconds, at which pawn odds would be a fair match. As programs become stronger still, the rate of play could be slowed down, eventually reaching, say, 3 minutes per move (on average). When the best programs of the day can give the world's strongest human player pawn odds at 3 minutes per move, we simply increase the odds to two pawns and reduce the rate of play again.

This idea could, perhaps, also be employed in a new form of human vs human Chess – the handicap tournament. They have them in golf, why not in Chess?

Incidentally the comments I have seen thus far on Adams' performance in the match all appear to omit to mention how well Hydra played. To my mind Hydra played like the Bobby Fischer we knew and loved in the 1960s and early 1970s. Hydra's style was as clear as crystal, its moves were direct, to the point, and rather devastating. Amidst all the negatives being uttered about this match, should we not be fair in our praise of the victor?

David

I'm sure that we will soon be reading about how Hydra has performed at a 3000+ level in its match against Michael Adams. From a raw mathematical standpoint this is true - Hydra's 92% score against such a top grandmaster is well above a 3000 raw performance rating - but traditional performance rating calculations are quite flawed because they do not consider the number of games played. If Hydra had scored 55/60 against Adams, in a sixty-game match rather than a six-game match, it would receive the same raw performance rating as scoring 5.5/6. But the shorter performance is surely not as impressive. My calculations try to take this into account.

Through my recent Chessmetrics work, I have developed a different notion of performance rating. It represents the rating we would assign to the player, if we knew of no other results for them besides their score in that one event. So if we asked, how strong do we think Hydra is, based only on its results against Michael Adams, that is what this calculation tries to answer. Remember that six games is not that many.

The answer? Hydra has a Chessmetrics performance rating of about 2850 for this result. It's well below the best-ever performances, which were Anatoly Karpov's 2899 performance at Linares 1994 and Bobby Fischer's 2887 performance in scoring 6-0 against Bent Larsen in 1971, but it is almost certainly the best-ever result for a computer. By comparison, Deep Blue's 3.5-2.5 score against Garry Kasparov in 1997 was a 2806 Chessmetrics performance. Hydra's results would rank as the 25th-strongest chess performance of all time by anyone, the best since Vladimir Kramnik's defeat of Kasparov in 2000. I haven't updated my computer-human results database in a while, so I don't have too much else to say about it at this point.

Subject: Business proposition

I have a great business proposition for you: I give you $10,000, or $20,000 if you are really, really good. You get on a stage in London and allow yourself to be humiliated by a machine for six days. How does it sound?

The dreadful final score was dramatic, and we could even add that Adams had the worse of the one draw. Unfortunately for chess, Adams, and Hydra, the machine didn't have to show anything special to put up this amazing score. Adams tried to play "normal" chess without any apparent special preparation and we've known for a while that this is on the foolish side of bravery. Not being in his best form, as seen from his recent events, certainly didn't help.

I remember seeing the impressive amount of preparation done by Kasparov and Kramnik for their matches against Junior and Fritz. A team of seconds helped analyze the programs' games, looking for weaknesses and tendencies. This went on for months in advance. Months! Adams has been playing a lot lately and of course he didn't have Hydra or its previous incarnations available for training. With his full tournament schedule it's hard to imagine he did anything near the sort of preparation required to battle a machine this tough.

The problem with such a one-sided result is that we really don't know how strong Hydra is, other than "damn strong." From a cursory examination of the games, there were no amazing moves on its part, nothing extraordinary or out of reach of strong software. It simply wasn't required. I don't doubt Hydra is the strongest chess machine on the planet and probably ever, but this didn't turn out to be the test to prove its worth. A narrow victory against an in-form and well-prepared Adams (and he isn't one to make such excuses himself), would have been more convincing than this blowout. To his credit, Mickey kept playing hard.

So we have to wonder what's next. Hydra got smacked around in the Advanced Chess ("freestyle") tournament by humans with laptops, so they know it's far from perfect. But they don't have much incentive to keep plowing money into the machine to make it faster if they are already putting up such massive scores against top-ten players. One hopes their next opponent, if there is one, takes the contest more seriously and puts together a team that does the same.

One useful thing this match showed is that strong humans can lose to machines consistently without blundering. This is relatively new. Technical play and grinds with equal material of this sort were relatively rare in machine wins over Grandmasters. Sure, they would grab a pawn and win all the time, but here Hydra never really had a strategically inferior position. It made a few odd "computer" moves, but for the most part it held the balance positionally. When that happens, it's almost inevitable that the computer's error-free play will eventually triumph. To win, or even draw, a human needs a considerable head-start by reaching positions the computer doesn't understand.

It's time to stop wondering how strong chess computers are, and about time we start enjoying chess as a sport between humans. Watching Adams get his a** kicked by Hydra is like watching Carl Lewis get his a** kicked by a Ferrari. The greatest sprinters top out at about ten miles per hour and Garry Kasparov peaked at about 2850. These are just the facts about being human; there is nothing more we can do.

In the past, some people have tried to perpetuate the idea that chess computers continue to struggle with artificial intelligence issues such as intuition and prediction. This is far from the truth, however, because chess computers are single purpose machines, programmed for a specific purpose: to play chess.

Now that Adams has been trounced by what looks like a 3000 Elo behemoth, let' move on.