I don't get the value of this game at all. It looks like Bronstein playing black, intentionally sacrificing material to throw off the computer evaluations.

First the game. At first sight you might think that Black does something that humans have been doing for centuries and computers usually find difficult. Black sacrifices pawns for positional compensation. For me the key position is this one, after 18... Na4.

click for larger view

Black is two pawns down but has better coordinated pieces and strong pressure against the white position. Black has the bishop pair and open files and diagonals. White has a badly congested queenside. Just how is he going to untangle his pieces? If the Bc1 can't move than the Ra1 never comes out of its hidey hole and black is effectively a rook and a bishop ahead.

The $64,000 question is whether that is worth the two pawns. If white can untangle himself he has an easy win on material. If white can't get out of the bind, black will increase the pressure and recover the material.

How do we answer this basic question - "is the sacrifice sound?" I think there are two answers - gut/ experience and calculation. And that is why, up to now, humans have been better at these sorts of sacrifices than computers. Our gut instincts and experience are very hard to reproduce in silicon so the computer has to resort to a brute force analysis (which we could not hope to replicate).

There is a way to inject human style thinking into computers and that is to tweak the evaluations to "reward" certain positional pluses. Having the bishop pair might be worth a certain proportion of a pawn, controlling an open file another chunk of eval, and so on.

So the big unknown here is whether Houdini calculated the sacrifices through to a clear advantage, or whether it was influenced by these human tweaks - the computer version of playing on gut instincts.

That may make the game a little less remarkable than we think. Throw enough computing power at a position and engines will find moves that seem initially startling. It's not magic - it's just the computer looking quite a bit further than we can, not making mistakes in its analysis and examining far more moves then humanly possible.

Does this mean that we are entering a new era of computers playing romantic chess? Possibly, but I don't think such an era will last for very long or that attacking games like this will be common. Improvements in computer power and programming will mean that engines will get stronger at both attack and defence. So the chances of computer #1 finding a crazy sacrificial line will be outweighed most of the time by its opponent finding a successful defence.

And, sad to say, a sacrificial line is more likely to have a hidden defence than a more solid line.

Computers will take chess into new territory. They will find sacrifices that we would not have spotted, and they will also find satisfactory defences to them. Every now and again one computer will outply another and we will get a game like this. But a return to the play of Morphy and Tal? I don't think so.

The second remarkable thing about this game has been the polarised reaction - either to over-praise the winner or over-criticise the loser. I suppose that's a natural human reaction. The level of analysis may be way beyond what we are capable of, but that doesn't mean that genius played duffer or that chess as we know it is about to change.

<26.Nb4> is the equivalent of a self-imposed strait jacket leading to the forced loss of a piece

BTW: Harry Houdini was never known as "The Amazing Houdini"

Still it seems an interesting line.

Many of us have seen this q-side traffic jam <QSTJ>, of the q-side B locking in the q-side R over and over again, from both sides of the board. Happens much more with the Black pieces.

Bronstein vs A Zaitsev, 1968
Petrosian vs Bronstein, 1956
Spassky vs Fischer, 1972

I imagine most strong FM and above human players would prefer Black after 18..Na4. The question is, would they see that position and go for it with 16...Nd3.

<Bonuses>:

1. Piece mobility in the middle and endgame (and different bonuses for each phase)

2. Outpost squares for knights and bishops (and an increased bonus if they are supported by a pawn).

3. Lower-valued piece (i.e. knight and bishop) attacking a higher-valued piece (i.e. rook or queen).

4. Number of squares within the area between the c- and f-files and the 2nd – 7th rank which are safe and available for each side's minor pieces.

5. Safe checks (patzer sees a check, patzer gives a check?)

6. If bishop can pin a piece or give discovered check via x-ray attack.

7. Major piece on 7th (White) or 2nd (Black) rank and enemy king trapped on 8th (White) or 1st (Black) rank.

8. Rook on open or semi-open file.

9. Advanced passed pawns, and/or pawns that are free to advance, increased if they have a free path to the queening square and/or they are supported by friendly pawn.

10. No pieces under attack by enemy pieces.

11. Have rook pawns and opponent has only a knight.

<Penalties>:

1. Higher-valued piece being attacked by a lower-valued piece or pawns.

2. Position of the defending king and threats against it.

3. If position is more drawish than it appears based on the full evaluation.

4. Opposite colored bishops in endgame.

5. Number of pawns on the same colored square as bishop.

6. Opponent has few pawns and you only have a knight.

7. Undefended minor pieces, even if not under attack.

8. Have rook pawns and opponent has rook or queen.

Well, yes and no. All commercial chess engines evaluate positional factors - that much is well known.

But the relevance of positional factors - and the extent to which they influence the choice of move - will depend on the position being studied. To take an extreme example, if an engine finds it has a forced mate it will not need to take into account any positional factors.

In the case of today's game, we have a chess engine making a positional sacrifice. But every <successful> positional sacrifice has to result in a favourable tactical/ material outcome eventually. The player making the sacrifice expects or hopes to be able to win material, force mate or get a draw in an otherwise losing position.

Chess is more than one thing. As originally conceived it is a contest between two human beings. You might call it a sport or a game.

In such a contest, the game is testing the individuals' powers of concentration, memory, stamina, will to win, creativity, the ability to calculate accurately.

A game of chess between two computers or a computer versus a human is still a contest but it is a rather different contest. The computer doesn't forget, so it is no longer a test of memory. The computer doesn't get tired or have emotions, so it is no longer a test of stamina or concentration or the will to win.

What is more, the computer calculates accurately in part because it is storing its analysis in its memory. In effect, the computer "cheats" by making notes as it goes along - something that humans are not allowed to do by the laws of chess.

Eventually it gets to the point where there is not much value in humans playing against computers. It would be like Usain Bolt running the 100 metres against a man on a motorbike. Not an equal contest.

So at that point we could give up on computer vs computer matches. That's not chess. It's unrealistic. They're cheating.

Except ... isn't it fascinating to see what they come up with? It's still a contest - it's simply a different contest. We do it because it's interesting. We want to know how far we can push this. We also expect that computers will find things which will benefit human players.

Ultimately we may find that chess isn't a contest. It's really a puzzle. Chess may be "solved" when enough computing power has been thrown at it. Other games have already been solved in this way.

But regardless of that, people will continue to play chess, and it is in fact the very imperfecions of their games that make it interesting.

The mathematician John von Neumann who did maths on "game theory" didn't consider chess to be a true game, like tic tack toe, as it could (in theory) be solved.

But it is still is and probably will always be - de facto - a game precisely because of error.

If humans always played "perfect" chess like very advanced computers and say won or drew every game (the losers would have to either be weaker players or play more weakly on purpose) then the winners would lose interest in chess.

We could never recall the infinite lines the computers might "learn" and nor could we ever play with as much precision (even without time restraints). So the game remains.

Yes, of course. "Checkmate leaves no weaknesses in its wake" – Horowitz. But I also think that a game between 2 engines also tests the computers' memory (amount of information that they can story in either RAM or disk), creativity (the factors incorporated into its evaluation function by the engine's developers), and the ability to calculate accurately (the accuracy of its evaluation function and search tree pruning heuristics). And both the computer and the human store their analyses in their memories; it is no more "cheating" by the computer to "make notes" by storing the results of its calculations in its memory than it is "cheating" for the human to "make notes" by storing the results of its calculation in his/her memory. And both of them need to be selective when deciding which calculations should be discarded once the available information is larger than what their memories can store.

And computer vs computer matches <are> chess, but between different contestants, namely the developers of one engine vs. the developers of a different engine, and a test of how well they can utilize the hardware that the engines run in. And, at least for the foreseeable future, those contestants will be human. It's more like a motor race where the contestants are drivers and cars. If the cars are equal, then, luck aside, the better driver will likely win; certainly if the race is a long one and the course is challenging. If, however, the cars are not equal, then the better driver might not win if the performance difference between the cars is significant. In the later case we would be justified in praising the superior car's designers, builders, and mechanics. But it will still be a human (team) vs. human (team) contest.

One useful feature of computer analysis has been that 'annotation by result' is largely rendered as dead as the dodo bird.

13. dxe5 Nxe5 14. Na3 h6 15. Ne4 Bd7 16. Bf4 Bg7 17. Rad1 and white has developed the a1 rook (Fruit 2.1)