1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. f3 e5 6. Nb3 Be7 7. c4 a5 8. Be3 a4 9. Nc1 O-O 10. Na3 Nc6 11. Ne2 Nd7 12. Nc3 Nc5 13. Be2 f5 14. Nab5 fxe4 15. O-O exf3 16. Rxf3 Bf5 17. Qd5+ Be6 18. Rxf8+ Bxf8 19. Qf3 Kh8 20. Rd1 Qa5 21. Nd5 Rb8 22. h3 Bg8 23. Qf2 Qd8 24. Qg3 b6 25. Qg4 Be6 26. Qf3 Na5 27. Kh2 Rc8 28. Bf2 Rb8 29. Bg3 Kg8 30. Nbc7 Bf7 31. Rf1 Bg6 32. Nb5 Qe8 33. Bh4 Be4 34. Qg4 Qg6 35. Qxg6 hxg6 36. Rd1 Bc2 37. Rd2 Bf5 38. Nbc3 Kf7 39. Bf2 Be7 40. Rd1 Be6 41. Be3 Bxd5 42. Nxd5 Bd8 43. g3 e4 44. Bf4 Nab7 45. Rd2 Ne6 46. Be3 Nec5 47. Kg2 Na5 48. Nc3 Bf6 49. Rxd6 a3 50. Rxf6+ gxf6 51. b4 Nd3 52. bxa5 bxa5 53. Nb5 Ke7 54. h4 Nb4 55. Bf4 Rc8 56. Bd6+ Kd7 57. c5 Nxa2 58. Bc4 Nb4 59. Nxa3 Nd3 60. Bb5+ Ke6 61. Nc4 Nxc5 62. Nb6 Kxd6 63. Nxc8+ Ke6 64. Bc4+ Kd7 65. Nb6+ Ke7 66. Nd5+ Kf7 67. g4 f5 68. g5 Kg7 69. Kf2 Nd7 70. Bb5 Ne5 71. Nf4 Nf3 72. Kg3 Nd4 73. Ba4 Kf7 74. Bd1 Kg7 75. Kf2 Nc6 76. Ke3 Ne5 77. Ne6+ Kf7 78. Nc5 Ke7 79. Kf4 Nc6 80. h5 gxh5 81. Bxh5 Kd6 82. Nb7+ Ke7 83. g6 Kf6 84. Nd6 Ne7 85. Ne8+ Ke6 86. g7 Nd5+ 87. Kg5 Ne7 88. Nc7+ Kd6 89. Nb5+ Kd5 90. Kf6 Ng8+ 91. Kxf5 Ne7+ 92. Kf4 Ke6 93. Nc7+ Kd7 94. Ne8 e3 95. Kxe3 Nf5+ 96. Kf4 1-0

Four of the participants qualified from Stage 2a, and four from 2b.

The top four will qualify for Stage 4.

It is played on a 16-core Xeon server.

Houdini vs Stockfish.

nTCEC is using 150m + 60s per move for the whole game, running on a dual Intel Xeon E5-2689 @ 3300 MHz. 16 cores will be used for each engine.

In any case, I would love to see that kind of match ... Another possibility would be a small team of the best human players ... maybe the top 6-10 human players vs. Houdini 3 in a normal time classical time control game ... and maybe with an analysis board ... assuming such a team would not end up squabbling and bickering like little kids, I suspect the human team would win.

<One synapse, by itself, is more like a microprocessor -- with both memory-storage and information-processing elements -- than a mere on/off switch. In fact, one synapse may contain on the order of 1,000 molecular-scale switches. A single human brain has more switches than all the computers and routers and Internet connections on Earth.>

http://www.mindpowernews.com/BrainV...

<Kurzweil estimates the computing capacity of the human brain at 20 million billion FLOPS – or 20 (here it comes) quadrillion FLOPS. It’s not easy to precisely calculate the computational speed of the human brain, but if we were assume he is correct then based on the historical data above we should see supercomputers reaching human-like speeds within the next 6 years – so by 2013. But due to the Law of Accelerating Returns, even if he is off by a factor of 10 or 100, that will only add a couple of years to the estimate. So we can very safely predict that by 2015 we will have machines with the computing capacity of the human brain.>

I personally programmed the computer for chess months ago. I gave the machine an understanding of the game equal to my own. The computer cannot make an error. And, assuming that I do not either, the best that could normally be hoped for would be stalemate after stalemate.

I'll post my response here since it seems more appropriate. This is basically a repost with some modifications and additions.

If knight odds is on average worth about 600 points (in some cases maybe 800 in others 400, etc), a human player has absolutely no chance of beating a PP (perfect player). As mentioned before, modern engines already may be 3400-3500+ relative to humans; and they are obviously very far from perfect play. Since PP has to be way higher than engines with 3500 Elo, it follows that 600 Elo odds is definitely not enough to beat a PP.

To simplify and make things a bit more primitive, lets take a rather conservative estimate for a PP of 4500+ Elo (a difference of 1000 points relative to the best engine seems reasonable). How is a 2900 human supposed to beat an entity which is some 1000+ Elo stronger with knight odds?

An accident? Well since the PP is defined as a perfect player, he does not make mistakes! So how exactly is human able to beat the PP? I don't see any possible way.

Of course, the story changes dramatically when the odds are increased: eg, Queen odds is so overwhelming that the Elo difference now is in favor of the human and even if he is playing the PP, he should still win (apart from any blunders).

Now, one issue here may be the Elo value of knight odds. Obviously it depends on the opponents, but relative to this discussion, an average of 600 seems reasonable. However, even if one is extremely generous and gives it 1000 Elo, it still doesn't change the conclusion relative to PP, since now the human is not 1000 Elo weaker but 600 Elo than PP with knight odds. Still a huge difference.

Well, I am pretty sure you are wrong on this one.

The point is that when you reach some level of proficiency, you simply <know> how to convert totally winning positions, no matter who is playing against you (that's why strong players simply resign when they are a piece down without compensation (or time trouble)).

Carlsen would easily know how to take full advantage of the huge deficit of his opponent, without giving him (it) any chance, no matter who he is. In fact I'd say that even an average GM or a strong IM would be able to do that.

---

Normally I easily beat chess amateurs that never studied the principle of chess, while giving them Rook odds (and even queen odds in many cases).

Yet I recall that when I was asked whether Kasparov would have beaten me with rook odds, it was obvious to me that I would easily beat him (and also with knight odds, but less clearly). I explained that while he is expected to beat me 50-0 in normal games, I just have that proficiency level, that he simply wouldn't be able to stop me from converting my advantage.

And another example from a different field:

Martial-arts professional can beat a complete amateur (who never practiced the main principles), with one of his arm tied behind his back.

No, actually I convinced myself that the result of that match (human vs best engine) is not completely clear. So I would insist that we can't say for sure how the match would end. However, I think with knight and two pawns the human would be the favorite for sure.

However, the real argument was not about an engine but about the divine perfect player (call it PDP or PP) playing a 2900 human, when I think one knight is absolutely not enough to win (unless by accident but even then it gets complicated as I explained above). That's what my post here was about.

<The point is that when you reach some level of proficiency, you simply <know> how to convert totally winning positions>

We have to understand that <totally winning> is a relative term, it depends on the opposition. Eg, pawn odds is <theoretically> winning, but even the best human will struggle against the best engine. Of course, there are tablebase winning positions which will always lose if the winning side knows the theory; however, knight odds is simply not such a position as compared to some tablebase endgame position since there are still trillions of possible ways to go wrong.

Certainly it could be true in some games that one knight is enough against an engine, but I simply can't imagine it being enough against the PDP to win a match.

I noticed that in your previous post you mentioned perfect play, but I didn't address it in my response since I wasn't sure what it means in that context.

I mean, perfect play a knight down is probably just to resign, but if we don't allow resignation, than how do you define PP when you are totally lost? Choose the move that results in win/draw in most variations? My guess is that it wouldn't be the best practical move, since all those wins/draws are based on bad decisions from the other side, and hence they are quite irrelevant.

Perfect play is efficient when there is a clear path not to lose, but if such path doesn't exist than <Practical Play> is probably preferable.

And if we go in that direction, than I'd say that even if 30 yrs from now, we'll have engines that are 500 elo stronger than today's best engines, Carlsen would still beat them with knight odds, for the same reasons I mentioned before.

Those extra 500 elo points will probably mean: Much deeper opening book, deeper calculations, deeper table bases, and improved practical moves selections algorithms, but I don't think that by then there will be any real new revelations about the game, that would change my original assumptions.

<shach: We have to understand that <totally winning> is a relative term, it depends on the opposition. Eg, pawn odds is <theoretically> winning, but even the best human will struggle against the best engine>

Well actually pawn odds is not totally winning. The material advantage is not that great, and moreover, changing the pawns structure really creates unbalances in the position, since it opens files and diagonals that can be used as potential compensation.

But what kind of compensation one gets from giving knight odds? that he can castle one move sooner?

There is no real imbalance with knight odds (other than that missing knight..), and so the start position really is totally winning for the one having 2 knights.

We probably don't understand each other: PDP is defined as the player who sees every possible resulting position of every possible move, sees every line to the end of the game; and the game starts from the start, so even-though <theoretically> the position may be won between two PDP's (knight odds, pawn odds, etc), in practice whether it won or lost still depends on the players.

And <perfect play> is refereed to the PDP and not to the human player, who will still usually make inaccuracies and blunders. I additionally define PDP as the entity for which chess has been solved, the whole of chess is like one big tablebase for it.

So, the issue is a match between PDP vs 2900 human.

<And if we go in that direction, than I'd say that even if 30 yrs from now, we'll have engines that are 500 elo stronger than today's best engines, Carlsen would still beat them with knight odds, for the same reasons I mentioned before>

This is your opinion and that's fine; though, again, I wasn't talking about engines (which still play imperfect chess), but a perfect player. If today's engines are 3400+, and are still very far away from being perfect and clearly haven't solved chess, the PDP is perhaps 4500-5000+ Elo; though this is a bit artificial but still seems reasonable.

So how exactly will a 2900 human be able to beat an entity which is some 2000 Elo stronger? Knight odds on average may be worth about 1000 Elo, so the PDP will still be another 1000Elo stronger (including knight odds). Hwo is the human able to beat the PDP who is so much stronger?

<There is no real imbalance with knight odds (other than that missing knight..), and so the start position really is totally winning for the one having 2 knights>

Again, it's only totally winning theoretically, in practice it depends on the players. Eg, a 1800 Elo will usually not be able to beat a 2900 with knight odds. So it is clearly winning only in the relative sense, depending on the players.

Of course, we can talk about the various imbalances (though on knight odds clearly does create imbalances, material imbalance for instance, and as you mentioned castling), but the point i tried to make is that: theoretically one knight odds is winning if the two players are the same strength, but when one player is so much superior than another, one knight odds is far from clear (and I believe, on the bases of the ratings considerations and the complexity of chess in general, that one knight is not enough against PDP).