What would happen in a 10-game match between one of the strongest human player (Carlsen, Caruana, Aronian, Kramnik or Anand) and one of the strongest engine (Komodo, Stockfish or Houdini), but with the following two conditions:

  • The human always has the White pieces in every single game.
  • The human is given draw odds: a draw is counted as a White win (i.e. in the advent of a draw White gets 1 point and Black gets 0 point).

The games are played at classical time control: 120 minutes for the first 40 moves, 60 minutes for the next 20 moves and then 15 minutes for the rest of the game with an increment of 30 seconds per move starting after move 61 has been made.

The engine is run on a powerful computer, gets 7-piece Lomonosov endgame tablebases, has permanent brain (ponder) on, and has a big opening book designed to avoid getting into drawish positions.

  • 4
    "Can" a human defeat an engine? Of course. I think the question should be about the probability of such a thing happening. – Wes Nov 16 '14 at 7:23
  • 3
    My guess is that someone like Kramnik who plays the Catalan almost like God can make a draw against the best engines by playing 1. d4. That shouldn't be too difficult for him. – Wes Nov 16 '14 at 7:25
  • 1
    I have little doubt that the engine would be the strong favorite in such a match. – dfan Nov 17 '14 at 0:27
  • 2
    May the engine's evaluation function be adjusted to avoid drawish positions (e.g. setting an absurdly high "contempt" factor to treat draws as losses, avoiding opposite colored bishops etc.)? – JiK Nov 17 '14 at 11:33

In contrast to Rauan Sagit, I'm going to offer some reasons I think any human should (assuming they don't get lucky and play some computer-level games themselves) lose such a match.

  • Elo differential: As of Nov.15, 2014, the highest-ranked engine on CCRL 40/40 is Komodo with an Elo of 3303 on relatively modest hardware of an Athlon 64 X2 4600+ 2.4 GHz (Engine ratings were initially based on human ones, but you could debate how relevant it is to compare the two. We have no other metrics, though, so I'm going with this). The all time peak rating was Carlsen's 2882 - let's use him as our example and assume he is able to play consistently at a 2882 rating level. This is a differential of 421, and the central observation of the Elo rating system is that a rating difference of 400 points should correlate to a 95% probability of the higher rated player winning (not counting the possibility of draws). Top GM games draw around 50% of the time, meaning that even if this rate stays the same, Carlsen will be fighting for draws with no practical possibility to push for a win. This means that, at best, you would be expecting him to stand a chance of just winning the match with him playing at a peak level consistently throughout the match.
  • Time control: If the time control is short, humans are likely to blunder when at best computers will make sub-optimal positional choices. If the games are long, it becomes very hard for the human player to remain consistent until the end of the match.
  • The engine (assuming the people setting it up actually want it to win) would have it's 'contempt for draws' set really high, meaning it would keep pushing on until the human cracked.
  • The psychology of such a match (assuming that there is enough of a prize/prestige for the human to be taking it seriously) can be really taxing on the human player. Kasparov-Deep Blue were 6-game matches and Kasparov was finding to hard to maintain his mindset and mental stamina well before the end of these. Of course, there were other factors at play at the time, but a 10-game classical control match is very taxing.

I'd also point out that Rauan's last paragraph suggest a modification that would likely make it too easy for a human player to win. keeping the computer's calculated advantage relatively low after 30 moves should not be too difficult with appropriate opening choices. I think a better adjustment would be that if the score is 0.00 +- 0.05 for at least 5 ply and the human offers a draw, the machine must accept. Either that, or make the match take place over a long period of time, with one game every 3 months or so (in amongst other tournaments) to prevent fatigue.

  • Just a note about the time controls: the settings of the computer affect a lot to their play, and I doubt their positional understanding amounts to something measurable. Check this game. – Pablo S. Ocal Nov 16 '14 at 20:37
  • @pablo I agree, but at short time controls I'd say that a human is still very likely to make a tactical error at some point, even with standard anti-computer tactics. A computer's positional understanding is also poor, but becomes exceedingly-well justified in a tactical way the more time they are given. Even if they didn't have a pre-programmed value for the benefit of open files (which many do) for example, they would quickly reach a ply-depth that would allow it to see the importance of preventing the opponent from taking control over it. – DTR Nov 16 '14 at 21:44
  • I agree that the strongest engine should win such a match. Still, the human player does have a chance to win the match, thus the human player can win. The unfortunate thing about the word "can" is that it can also mean "can, but most probably won't". Thus, my answer does not suggest that the human player will most probably win the match, it just says the player can win. Thus being a valid and still realistic answer to the question :) – Rauan Sagit Nov 16 '14 at 22:08
  • @RauanSagit Seems reasonable enough - The answers can't be much more than qualified opinion here anyway, so I figured it'd be worth having both points of view :). – DTR Nov 16 '14 at 22:15
  • @Dave opposite views are always great to have. With so strong engines, I don't see any human versus machine matches coming up in the future. Human plus machine matches are much more likely. – Rauan Sagit Nov 16 '14 at 22:18

Let's see what happens if we just look at the math.

According to the FIDE handbook, a rating difference of 422 points would mean the engine would get an average score of about 0.93 per game. Assuming that the human player is only going to draw and not win, that results in the computer winning about 86% of the time, and the human drawing 14%.

If the computer wins 86% of the time, the computer will win a 10 game match about 99.3% of the time. About 0.6% of the time, it will be a 5-5 tie, and 0.1% of the time, the human will win.

But this ignores the effect of giving the human the White pieces. This blog suggests the first-move advantage is worth about 35 rating points. That would put the computer's odds of winning a single game at perhaps 83% instead of 86%. Using those odds, the computer would win the 10-game match about 98.3% of the time, it would be a 5-5 tie about 1.4% of the time, and the human would win about 0.3% of the time.

  • I like this approach, if rating calculation was perfect, this would be true. Still I don't think ratings between humans and computers are connected much. Ratings are not connected very well even between central Europe and Russia, even less with India for example. Players with same ratings have different skill in different part of planet. Computers have their own playground. And there is drawing tendency that has some meaning with increasing ratings. Also computers are not designed to play for win against weak opposition. This hurts any calculation seriously. But computer should win of course. – hoacin May 28 '17 at 7:26
  • Worth pointing out that engine ratings don't translate to human ratings. They're anchored differently. – Allure Dec 7 at 8:20
  • I don't suppose there's an estimate of how they translate (or just an estimate of how far off they could be?) – D M Dec 7 at 23:10

Any top human would be annihilated under those conditions. Even a mobile phone can destroy any human in a ten game match even under those conditions. The only hope would be those conditions plus give the human an exchange odds in every game.

  • I forgot to give an example. Nakamura was unable to win a match vs a top engine with conditions in his favor. I think he played Komodo. (Edit: I should have been clearer. Nakamura did play Komodo in an odds match but I should have specified the exact match. There is a video on YouTube about it.) btw there are multiple videos one of them was badly edited unfortunately. Anyway you can watch separate videos on each game. – Danny May 25 at 21:20

No.

Chess computers are now fantastically strong. To put things in perspective:

In other words, Stockfish 10 crushes Stockfish 9 crushes Deep Rybka 4 crushes Deep Fritz which beat Kramnik. There's only one conclusion from this. Realistically, humans have no chance, and would be lucky to even draw a game.

Yes, a strong Grandmaster can beat the strongest chess engine given these conditions. Still, these conditions highlight the fact that today, the best engine is much stronger than the best human player. Thus, the human player needs considerable odds in order to agree to the next human versus machine match.

Yet, imagine a strong Grandmaster (e.g. Elo > 2700) agreeing to these terms (10 games with classical time controls, white in every game, draw means win for the human player). Then, imagine the human player losing the match. I think a human player would prefer losing a match without any odds than a match with odds.

In addition, perhaps more subtle conditions need to be offered to even out the gap in strength between the human and the machine. Since an engine is so excellent at pushing a small advantage to a victory, draw odds may turn out to be too rigid. For example, if the engine does not have more than 1.0 pawns advantage after 30 moves, the game is drawn.

Edit:

Clarification. While a human player can theoretically win such a match, I think that the engine is more likely to win such a match, because the engine is much stronger. The engine in this case would have an excellent opening tree, would most likely not blunder a single time and would most likely identify and use every single blunder made by its human opponent.

  • This answer does not back up the initial claim: "Yes, a strong Grandmaster can beat the strongest chess engine given these conditions". In the paragraph everything is opinion-based, so hard to make something out of it. – gented May 23 at 12:16

I think the human will struggle to score just 1 point.

Engine elos are not even compariable human elos. I estimate the strongest engines are already over 1000 elo points stronger than humans like in go, and the engines in chess are much more refined. I think currently chess engines are a full rook stronger than the top humans.

Engines in go are not even refined as well as chess engines and they can beat pros with 4 stone handicap, which is basically a rook in chess.

  • Engines are certainly not a full rook stronger than the top humans - see chess.com/news/view/… - MVL beat Komodo at odds of rook + pawn for knight. – Allure Dec 9 at 2:28

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.