The arrival of strong chess engines as Stockfish, Houdini or Rybka allow top players to deeply prepare lines in openings or to study better endgames.

I would expect a rating increase of top players. However, Kasparov's top score from 1999 is still the second best rating score of all time and Karpov and Fischer are still in the top twenty historical rated players.

Enter image description here

Source: Wikipedia

If those engines have a 3600 Elo, shouldn't top players have profited from it to leave Kasparov's 2851 score in 1999 far away?

And shouldn't more than 18 players surpass Fischer's 2785 in 1972?

  • I will just pitch in - while the answers below are of course informative - that the simplest conclusion doesn't require all that much explanation: a glance at the rankings posted is the clearest possible evidence of top players' ELO inflation. Or should one believe that there are 15 players active today more worthy of pages in the chess annals than Karpov and Fischer? The simple truth is that Kasparov was an extreme anomaly, and so was Fischer's peak rating. Apr 18, 2021 at 3:52

5 Answers 5


I think they certainly have increased their ELO, but more importantly, their overall chess strength. ELO is only a rating relative to others in the pool so it may tend to go up more slowly if everyone in the pool gets better, which they have collectively.

First, you need to take an average of the top players, rather than look at just two incredibly special players, Kasparov, and Carlsen, and their peak ratings. Fischer was a total anomaly for his time, and again, you cannot base a lot on one person. Humans probably have a limit mentally much the way runners will eventually hit a speed wall that they cannot break. It is the Bell Curve in action: Easy to make progress in the middle, but harder on the end.

ChessBase, and the first strong engine, KnightStalker (aka Fritz 1.0, renamed at version 2.0) came out in 1990, and the Internet was still in its infancy. In January of 1990, there were only two players over 2700, Karpov at 2730, and Kasparov at 2800. Today, there are 37 players over 2700, and of those, three are over 2800.

I have been playing for a long time, and anyone, who knows my handle, has seen me watching relays of the top tournaments on ICC for more than 20 years, and I can say that it is not just inflation from having actually watched the level of play improve over the years. Players at the Master-level, including even normal Masters, are MUCH better now due to the tools available today than the tools available when I first became a Master.

When I first started playing, all we had was books, and a local club. Today, I can logon virtually any day, and play a GM. I was also a ranked tennis player as a teenager, and I can tell you from that experience, playing better competition makes you better, so being able to logon and play stronger players than ever has increased the overall level of play. In addition, we now have ChessBase, and a database of about 8.3M games that can be sorted efficiently and studied. We also have engines, as noted in the question, and they certainly have also added to our chess knowledge, and fundamentally changed how the top players prepare. Lastly, I mentioned books, and as an avid collector (I had 1800 at my peak), I can tell you that even the quality of the material in books is on another level since when I started.

I cannot tell you what proportion of all these advances have played toward X-number of rating points increase, but I can tell you that without a doubt, they all have lead to players at the top levels, and even down some, being much better overall.

  • Of course it has improved play for many people. The real question is why are not the top GMs even higher if they have these better engines and databases to help them.
    – yobamamama
    Dec 26, 2019 at 17:00
  • 11
    I answered that in the very first paragraph: "ELO is only a rating relative to others in the pool so it may tend to go up more slowly if everyone in the pool gets better, which they have collectively." You would only see huge disparities if only one, or a few, players had access to these tools. Dec 26, 2019 at 17:31
  • 1
    The bell curve actually implies the opposite. A small change in means does not change the midbell populations much, but can double the respective tail population... Dec 27, 2019 at 10:00
  • 1
    We are either not talking about the same thing, or I did not describe it well enough, or you do not understand my point; but I am not talking about small changes. What I meant is that it is easier to make progress in the middle of the curve than at the upper end, in particular. It is easier to go from a 50-60 than to go from 99-100. I hope that clarifies what I meant. Dec 27, 2019 at 10:38
  • 1
    I know what you meant (diminishing returns / getting asymptotically closer to an absolute limit), but that's not how a Bell curve distribution works. With sufficient luck, you can get arbitrarily far from the mean: it implies there is no hard limit. (This is separate what @StianYttervik was pointing out.) A Bell curve probably only models things well up to a few std deviations away from the mean, and probably isn't useful for predicting how much harder it will be to get closer to human performance limits. It's a distribution function, not an effort vs. result graph. Dec 28, 2019 at 1:09

The question is apparently based on a misunderstanding of how ELO ratings work. There is absolutely no mechanism by which the overall increase in players' strength would lead to increase in their ELO. The actual value of the ELO rating bears no meaning; nor does the comparison of ELO at distant times. The only thing that has direct relevance is the difference in simultaneous rating between players.

If tomorrow FIDE decided to add a constant, say, 500 points, to every player's rating, essentially nothing would change - the outcomes of games would lead to the same increments in ratings, so the rating dynamics would be the same, just at +500 points level. And the computers will provide no basepoint - their ELO is relative to human players, so they will play at 3600 rather than 3100.

The same applies if such a constant were somehow added or subtracted gradually due to natural processes - e. g., expansion of rated players' base, players becoming inactive with higher rating than they started with, etc. And this, apparently, did happen as we witnessed rating inflation until recently.

Similarly, if all the players suddenly, or gradually, start playing stronger, without a change of their relative strength, there is no reason to expect a change in rating dynamics.

  • One way to observe this is "rating pockets"; if there's a boom in an isolated playing community, and many players start improving rapidly, their ratings will deflate, because in rated tournaments, they'll be mostly facing each other. Their isolation means that even if some of the players participate in tournaments "outside" the pocket, not enough Elo points are added into the rating pocket's total pool to compensate for the overall increase in skill level.
    – Bass
    Dec 26, 2019 at 23:03
  • 1
    I think the OP assumes that top players use computer programs more or more effectively so that they should improve more than the main field, so that their Elo rating rises. This could also affect their rating relative to past players. I think Brian addresses this argument (by refuting it). Dec 27, 2019 at 11:31

If those engines have a 3600 Elo, shouldn't top players have profited from it to leave Kasparov's 2851 score in 1999 far away?

No. The main effect of engines and also the internet has been to democratise chess. The top players have always had access to top level evaluation and knowledge. For lower level players that kind of knowledge, analysis, position evaluation, endgame knowledge was very difficult to acquire.

This is important because higher level players gain rating points by beating lower level players. If those lower level players are harder to beat because of increased access to the analysis and knowledge that engines and the internet provide then it actually becomes harder to reach higher levels as measured by rating.

  • So Kasparov had it a bit easier. But he hadn't those tools neither, wich is a point for him. I guess the resume is if Carlsen goes back in time with his knowledge he would score >2900. So engines have increased chess level.
    – user18196
    Dec 26, 2019 at 19:57
  • I anyhow don't understand how is measured Stockfish 3100 ELO. Those engines don't play against humans. If it is short to answer on comments perfect. If not I will post this question soon.
    – user18196
    Dec 26, 2019 at 20:00
  • 1
    @Universal_learner I think that is a new question. Would also be interesting what is the correlation/connection (if any) between engine ELO and human.
    – Brian Towers
    Dec 26, 2019 at 20:02
  • @Universal_learner well, they do sometimes play against humans, and you don't need that much games against rated humans to statistically evaluate where they lie on the current ELO scale.
    – Peteris
    Dec 26, 2019 at 20:04
  • 1
    @Peteris: A 2800 never draws Stockfish, so the limit if you try to compare it playing against humans is when a victory means +0 points earned. I guess it is evaluated against weaker engines that have played against humans (and sometimes GM drew them).
    – user18196
    Dec 26, 2019 at 20:07

A few things:

1) Ratings overall tend to rise due to inflation of more players entering the pool, rather than the average playing strength of players increasing. If everyone becomes better by some factor due to engines, why would top players' ratings increase? They're now playing opponents who have also gotten stronger by roughly the same amount.

2) Kasparov, Karpov, and Fischer are anomalies of a sort (Kasparov especially), considering they are all world champions. And still, they only occupy 3/20 spots on the list - all other ratings were achieved in the modern era.

3) Engines are making people stronger, but not to a huge extent. People aren't able to extract all the strength out of an engine, we just see the moves it recommends.

In summary, engines make people somewhat stronger, but in any case this wouldn't be reflected in the ratings. The pool of players is a closed system: everyone getting stronger doesn't benefit one player/group of players in particular. The way the ratings rise (at all levels) is from more people joining the system.

EDIT: one way you would see some increase purely due to engines is in a hypothetical scenario where we played against another group of people who didn't have access to our engines.

EDIT 2: Point 1 may not be entirely correct (see the discussion in the comments).

  • 1
    1) That's a common misconception: pure ELO rating actually deflates due to new players entering the pool. They enter at low rating, then progress, taking the points away from the others. As they become inactive, a vast majority of players are higher rated than at entry. This means they take points away from the system.
    – Kostya_I
    Dec 27, 2019 at 11:20
  • @Kostya_I But new players bring more points into the system via their initial rating. Even if one's starting rating is only 1000, they still brought 1000 points in by basically taking nothing away. Dec 27, 2019 at 14:08
  • @Inertial_Ignorance, imagine a new player who gets initial rating of 1000 points and never plays again. By your logic, he or she "brought 1000 points in", but in fact nothing changed for other players. And if they do play, they will typically progress - say, to 1500. That's 500 points taken away from other players, so, their rating will decrease by that amount while their strength has not diminished. That's deflation. Of course, someone could drop below their inital level, which is the only case they really "bring points in", but that's a rare case
    – Kostya_I
    Dec 27, 2019 at 14:36
  • @Kostya_I That's a good point. But inflation is happening with ratings, so how do you explain that? Dec 28, 2019 at 2:52
  • @Inertial_Ignorance, the inflation is no longer happening - the top-N averages are steady for about 7-10 years. The inflation in 1990s and 2000s might have been due to lowering the rating floor, where a lot of established players were added, who no longer progressed from that point on, but aged and declined in strength, giving away the rating points. Or it may be that the deflation counter-measures (different coefficient K for new players) were not properly adjusted.
    – Kostya_I
    Dec 28, 2019 at 9:21

People are not computers. They learn and memorize as much as possible but they can not do as much as a computer can nor as fast. The top players are doing everything they can but they can not do it all like the computer does.

  • Well I think Carlsen has sometimes played a 25 moves Stockfish line that wasn't on books. I agree with your answer and it gives a point but I was not asking why GM don't reach engines level, I was asking why they don't increase a bit more his rating, saying all top ten players at 2900.
    – user18196
    Dec 26, 2019 at 20:17
  • I suspect that all the top players are using as much as they can whether printed or online to improve. So if nobody improves more than the others and they only keep playing themselves then their ratings will stay the same. The top GMs need to play wider competition and take some of their points from lower rated players for the top to improve.
    – yobamamama
    Dec 26, 2019 at 21:01
  • That's the point of my question. Allbody have engines, but top rated players may prepare better games than said, the 2600. Caruana says his point is game preparation. Doesn't engines help 2800 to beat lazy 2600 more easy than Bobby, Gary or Anatoly faced to?
    – user18196
    Dec 26, 2019 at 21:34
  • Maybe, but that won't affect the ratings.
    – yobamamama
    Dec 26, 2019 at 21:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy