Stockfish 8: The Tal of AI

Most AI fights are pretty boring to watch, at least from an aesthetic standpoint. But you gotta see this, man.

Neato cabritto.

So is this to imply that StockFish 8 plays similar to Tal (more than other engines)??   I really don't know, am asking.  

Some very uncomputer like moves in there

Cool to watch.  Thanks for sharing!

It's not Tal until it can express witty sentiments in interviews!

34. Kg1, instead of 34. Rxf7+, is a very counter-intuitive move, I doubt Tal would have made it (he would probably have resorted to something like 34. Kf1, if he was to go with the idea of involving the queen immediately).  But I agree, the game is beautiful, even if a bit computer-like!

Yes, sir! Compared to the other engines, Stockfish has an almost comically aggressive style. Tal would be proud.

13.g4 is a novelty which will surely not appeal to many. It shouldn't work against anyone who understands the intricacies of this particular position (and thus prepare a quick ...b5), but engines aren't bothered by such "subtleties"...

Is it actually a novelty if it's been played in 28 computer matches? Apparently they "discovered" it back in 2015.

So, what they have discovered? Neither the correspondence, nor the OTB players cared so much about this "discovery". Black can transpose if he wishes to the mainline after 13...a6 14.Na4 Qxd2+ 15.Kxd2 Nd7 and now there is nothing better than the transpositional 16.Be2, but he may well have something different like 13..Rab8.

The game makes no real sense- e.g. why not 14...Qxd2+ 15.Kxd2 Bxd5 16.cxd5 Rxc1 (white has to recaprture with king, since his bishop is still at f1) 17.Kxf1 Rc8+ 18.Kd1  e6!? when the "novelty" just helped Black to solve his space issues.

Well, Stockfish is super strong, but it only does the most efficient thing. It's no Tal.

Take Chess System Tal II...Now THAT's a Tal AI!

(A couple of disadvantages of CSTAL are that it's a commercial program, and it has to be run in its own GUI - It's not Winboard or UCI.)

https://chessprogramming.wikispaces.com/Chess+System+Tal

Yea Pfren's right, the computer blundered. Pfren should play in the computer championships, he understands the positions so much better than all of them.

Based on seeing much of Pfren's analysis, I think this is often true.

Actually I do play serious CC chess at LSS since 2012. My current record is +139 =202 -2. The big majority of my wins are due to- you know what? The computer blundered, or misevaluated, or fell into a deceptively good position. It's quite common. If you think your computer is faultless, just play actively in a major correspondence server. Your results against strong opposition will be disheartening, at best.

That's great, but I'm not saying the computers are perfect. Just that your rating is under 2500 and Komodo 11s rating is 3393. Saying you have a snowball's chance in hell against komodo is disingenuous to the snowball, because at least the laws of physics could break down.

Furthermore, Nakamura recently played a series of classical games against Komodo, but he was unable to win or draw any of the 4 games even with advantages like f pawn odds, exchange odds, and tempi odds. Are you better than Nakamura? Would you have beaten deep blue? I'm not going to argue with you over any position in chess because you are so much unbelievably more knowledgeable than me about chess, but the idea that computers are better than humans at chess seems utterly beyond debate at this point. Computer's don't blunder, show me a position where a 3300+ computer blundered, I dare you, and prove it blundered without using another stronger computer. It must be a real, achieved, game position and not a specifically anti-computer position that takes advantage of incredibly specific weaknesses like how they evaluate draws. It has to go from a won position to a draw or a drawn position to a loss, not a position where it's at +4 and goes to +3 or -5 and it goes to -8. These discrepancies are not the computers blundering, but just quirks in the formulas they use to spit out the numbers combined with a horizon effect. A true blunder means the position is fundamentally changed between won, drawn, or lost.

If I could enter into the Sinequfield cup right now with even an engine like Houdini 5 I would win every game with no losses or draws and sweep Carlsen off the board.

Correspondence chess is even more hopeless than classical time control chess. I agree there may happen to be chances in this current year if humans get a correspondence time control while computers only get a few minutes to think and then have to play their move. But in a fair correspondence game, the time ratio is 1:1: the human player would get one week to think and then the responding computer would get one week to think, looking 50 - 60+ moves deep into the position. No human would stand a chance.

+1

I am really surprised you did not understand anything at all... oh well.

That 3300-something Komodo rating you have mentioned is from a pool of other engines playing at fast time controls. In a pool like ICCF or LSS, a sole engines' rating would be around 2000, at the very best.

I think it is pretty clear that top engines nowadays can easily destroy top chess players in ICCF that do not use engines themselves. However, a top engine vs a strong ICCF player using a top engine in a clever way will be absolutely demolished. Engines still lack in positional understanding, they just make up for it by their sheer calculation capability. But take Magnus with impeccable positional understanding, give him Komodo that will supplement it with insane calculation power - and as a result, you will get a super-player that will destroy both top engines and top players on their own with ease.

pfren is correct, a player just letting Stockfish or Komodo play the game for them won't get far in serious correspondence chess. The situation will change when the engines start working via complex neural networks, which will give them an analogue of humans' positional intuition - then correspondence chess may be in peril, and the rules will probably have to be changed significantly, to avoid "battle of machines" - but as it is now, engines lack in many departments, and the raw strength of their play doesn't mean they always make great moves.