Tuesday
Dec162014
Silman on Chess Engines
Tuesday, December 16, 2014 at 7:33PM
A chess engine is not the non-master's friend, says IM Jeremy Silman. (HT: Ross Hytnen) He moderates that conclusion slightly by the end of the article, but his general point that most players would be far better served trying to understand things themselves, in human categories, and would benefit from human teaching seems right to me - at least or especially for those who are interested in improving.
I'm inclined to agree with the drawbacks he notes of using an engine - especially (but not only!) for non-masters, but suspect he's neglecting some positive ways of using engines. What are your thoughts, readers, especially on this last point?
tagged Jeremy Silman, chess engines
Reader Comments (19)
Silman is a brilliant and insightful author, and I've found his books extremely instructive. But his pedagogy (and approach to chess) is conceptual and strategic in orientation -- which isn't to imply that his approach isn't dynamic, but rather that it's fundamentally at odds with the purely tactical, concrete approach that engines embody (more or less). An engine doesn't care about "imbalance" as a concept, doesn't make plans, doesn't do any of that except ruthlessly calculate. And several authors, John Watson among them, have made trenchant observations about the general irrelevance of long-term plans in high-level chess -- the positions are too dynamic, the situation too fluid. Only when the position has crystallized and one side has a passive or "dead" position do plans usually come into play.
So it seems to me natural that Silman would be hostile to engines, and rightly so when it comes to self-teaching and armchair warriors: I've never found that playing a computer is viable training for playing against humans, and there's nothing as tiresome as watching real-time kibitzing from 1200 players armed with Fritz or Stockfish.
But it's also true that if there are tactically unsound lines in How to Reassess Your Chess, it undermines the whole conceptual framework those books advocate. It's well and good to make a plan, or to ask (as he does) “What does this move do?”, but at the risk of sounding Orwellian, at a certain level the point of 422. Rfb8!! is 422. Rfb8. In other words, as we've seen in advanced tablebase endgames, "perfect" chess is often unparaphrasable, unconceptualizable: it simply is itself.
Or to put it another way, as I've long thought (and occasionally said, no doubt to the irritation of some), chess is 100% tactics, not 99%. Strategy is an incredibly useful shorthand, but it has no objective existence, even within the structure of the game.
In general I agree with IM Jeremy Silman - recently I find that I learn a lot more by not using engines. However, I feel that while many people do misuse engines, they can be very useful.
Regarding the point about the positive aspects of using engines (for a non-master) as a non-master, I think there is definitely room to use the engine for positive improvement. Recently I have not been using engines, except when following games on chess24 (I wish the TWIC live feed would have clock times too!). I used to use the engine for preparing openings especially after a game I played, and sometimes to play endgames out against. I think the engine was a positive influence for both reasons. I can think of another positive use for the engine: One can take any position and try to calculate one's assessment and variations and then compare with a computer - especially for the tactical points. In this sense, it is a great training tool - as one gets better, maybe one could play a few moves from the position until the position clarifies or takes a different character. This could be very useful for master games from the past. In this way, one does not need to buy books/dvds of great games in order to train against annotations from them (especially if one doesn't have the means or the space!).
I think it is important to stress that it is easy to fall in to the trap of using the "engine evaluation" as truth (as Silman implies). While it may be closer to the truth than what we can assess, it is not going to be part of our brains (ignoring futuristic applications for the present!) or thinking. In other words, one needs to think on one's own as opposed to relying on the engine "crutch". To give a different analogy, regardless of having a car/horse in one's possession, it helps one's independence to be able to walk without the help of either.
Silman raises some valid points, and the people harshly criticizing him in the opening story were certainly out of line, but I have some disagreements.
His
is far too dismissive. If an engine is right and its moves refute some suggested plan, then that plan is simply wrong in that precise position and to suggest otherwise is at best naive and at worst deceptive.Although I haven't read The Amateur's Mind, I feel that Silman's
defense of mistakes smacks of rationalization -- if his examples don't accurately demonstrate his themes, then the fault is entirely his and, if he couldn't find sound example, maybe it's because none exist. Perhaps most positions are simply more complicated than teachers would like to admit, and ignoring tactical (or even just good defensive) difficulties does the student a great disservice.Silman's reductio ad absurdum
seems out of place. Tossing everything would be ridiculous, but disregarding incorrect analysis (and replacing it with accurate lines) would be quite reasonable. Everything depends on what the author is trying to achieve. If the goal is to show how a given plan leads to victory in a given position, but the computer demonstrates that it does not, then sorry, the author has already lost. Of course, not every book has that as a goal.Commenting on Black's missed defense in Fischer - Reshevsky, 1966, Silman writes
I find this hard to understand. Sure, all of White's moves were good tries and he established something of an advantage, but if Black had found this defense then it all would have been for naught. White's play wasn't good enough to win by force, even near the very end. That has to matter! Had Reshevsky found that line, we would have had to praise his play, barely (and perhaps precisely) hanging on against Fischer's pressure.There are plenty of desirable strategic plans that are "declined" because of tactical issues. If one player misses (or ignores) the tactics and gets away with it, is (s)he demonstrating some great skill that transcends computers' understanding? Or is (s)he just getting lucky? What about players that see the tactics, choose lines to avoid them, and don't do quite as well? Similar issues hold for interesting tactical lines -- at some point you have to accept that people who see through them and therefore decline to play them are playing better than those that don't, regardless of how beautiful the lines would be if they worked.
The danger of overuse of engines by those who are inexperienced is real. Just as an example, in working on endgames with a young student, he was repeatedly confused by evaluations of positions as either winning (falsely, because of fortresses) or drawn (not false, but misleading given how many tricks there are to try to make progress and how the side to draw in many cases must repeatedly make the one single move that draws). On the other hand, I find it very instructive to use the engine to reveal the "truth" of the messiness of human games. For example, my student has sometimes observed that in a model game where one side lost, actually that side was fine all along up till a critical point, so the narrative of the one-sided game in some old annotations was a lie. [DM: A mistake, not a lie, unless you think the old annotators actually knew better and meant to deceive.] I think it's an art form to decide when to overlook some detail in order to learn a bigger lesson and when that detail is critical and teaches about, say, how to defend a particular kind of position that might seem bad. It all depends on one's readiness to hold many ideas at once in one's head.
I agree with much in this article but it is too extreme.
One example:
'All in all, a chess engine is almost useless for the amateur player. '
If used correctly chess engines can be of immense help to amateurs. I'd recommend the amateur first spend time analyzing the position without the computer. Once he is satisfied with his analysis see where the computer disagrees and try to understand why. How is this useless? [DM: He does gesture in this direction with his third bullet point under "Recommendations".]
IM Jeremy Silman is one of the best authors at relating to amateur players and yet among the most condescending.
I use engines mostly to check for blunders and turning points in my games. And I think they are extremely useful for that.
This master / non-master division is a bit irritating because there definitely is no shortage of masters who rely (too) heavily on engine assistance while writing or commentating. You don't need to be a 1200 player to be derisive of some super gm move because the engine evaluation jumped, I've seen (normal) gms do that. And there is no shortage of books that are basically just engine dumps or at least drown every insight in a flood of computer generated variations.
Regarding engine evaluations while watching games: Using your own brain is certainly preferable, but it is very interesting to watch a post mortem interview after you have seen a game with engine assistance. It gives you a much deeper insight in what these top players actually can see and how accurately they evaluate positions (Amazing: Carlsen's precise and objective evaluations. Equally amazing: Kramnik's constant over evaluation of his own position ...).
In a certain sense I agree with IM Silman when it comes to chess engines. I agree that if you rely on them you will not improve your play. The key to improvement is using your own brain. I have watched several recent "Super" GM tournaments where the live feed includes engine analysis of the games. My take from it seems to be somewhat different than most of the kibitzers and commentators. What I have learned from seeing the moves of the GM's and engine analysis is that I really don't understand chess, at least at the level of a master or especially a 2700+ GM. During the commentary, there is often an obvious difference in understanding between an IM or "average" GM and a 2700+ level GM.
I'm amazed at how often the engine recommends a particular line of play and the GM plays the same way, while I have no understanding of why that is the right line of play. As I've written to a fellow class A level friend of mine, we simply don't understand chess at that level. It's not a matter of the GM being a better calculator, or knowing reams of opening theory, it's simply the fact that they understand the game in a way that we don't. I've often pointed out to fellow chess players that just because a chess engine recommends a move or variation doesn't mean I, or they, would ever see it or consider it. It's simply beyond our grasp. I've also noted times when an engine will show that a game is totally winning for one player but a Super GM will miss the correct move and end up losing or drawing the game. That reminds me that even the 2800 players of the world are not infallible and in fact, are human.
I tend to agree with some authors (notably John Watson, Willy Hendricks and Axel Smith) that Silman's approach, and the teaching methods of most instructive books in the past, are only useful to a certain point. Chess is a game of good moves not good principles.
[DM: I'll chime in here with two quick responses. First, I think Silman would agree with the first sentence as well. He's not trying to offer an infallible algorithm and would hardly deny the importance of tactical skill. Second, I think your second sentence offers a false dichotomy. Good principles typically lead us to good moves. If we didn't have these principles we'd be hopeless at chess. Almost all of the moves we find are generated subconsciously by appeal to one of the zillions of "chunks" we've internalized over the years. (Examples of chunks include tactical patterns, ideas about structure, pawn breaks, typical plans, views about material, standard maneuvers and so on.) Club players know many of these heuristics, and the higher one goes the more chunks/heuristics the player knows and is better able to weigh and apply them. Try to play a game without using principles and it will be complete nonsense. Why not play 1.g4 followed by 2.a4, 3.Ra3 and then bringing out the queen to sac it for a pawn? To say that the center is important is to offer a principle; it's not that we see that 1.g4 followed by 2.a4 allows Black to mate in 120 moves with best play on both sides. Likewise, if 3.Ra3 involves an exchange sacrifice, the view that it is a sacrifice and not a fair or even advantageous trade is again a principle: we don't calculate the result out until mate. Likewise even with the queen sac.]
So when Silman uses an example position to teach a principle and it contains a tactical refutation, his principle loses credibility. And that is how it should be. As for chess engines, they are exceptional at pointing out tactical errors and that is where they provide the most value. When I have come up with a plan during a game, executed it successfully and won the game, I find it useful when an engine points out that both my opponent and myself missed an obvious tactic that would have changed the result. I've also noticed when reviewing games with an engine that there were times when I debated over two moves for a long period of time and the engine passed right over that moment as inconsequential. I think that is useful as well.
So, in my opinion, Silman is partially correct, I think reliance on an engine is not a replacement for thinking and improving your chess skill. He's especially correct when it comes to kibitzers online, but that's a different issue that has more to do with psychology than chess or chess improvement. On the other hand, as I mentioned before, chess is a game of good moves, not principles, and a chess engine can point out good and bad moves that a player may never have found on his own.
I disagree with the idea that an example loses its educational value because it's tactically flawed.
It's of course unfortunate, but if the idea has general value (eg. counter in the center against an attack on the flank), and moreover, if the example stems from a game between strong players and they both missed the tactical refutation (meaning it's not an easy one), then I think the concept has a lot of practical value.
Chess being already a draw with best play, it becomes a human game, rather than a scientific research field (at least in my eyes). Any idea that works up to a certain level have value, whatever the 3200 overlords think :-)
Now, there are certainly many ways engines can be useful :
- of course to spot various tactical patterns and correct analysis calculation mistakes
- to develop our feel for dynamic play
- to train how to convert a winning an advantage or hold your ground in an equal endgame
- to tune our own evaluation function when we compare it in some positions with the engine's
- to develop one's defensive skills by seeing how an engine defends against a "fearsome" attack
But where I agree with Silman, is that all of this is pretty much useless if we don't actively think how to integrate the engine findings into our own thought process (an exception might be the new tactical patterns we spot when analyzing, as they can probably be stored 'as is'). For this we often need to try and refute the engine's conception, in order to understand why it's superior to ours in the first place (and conclude if this is due to a tactical detail, or a plain misevaluation on our part). So some active thinking is required before we can transform the engine's insights and feed our own thought process. This is this last part that is completely left out by many engine users, and that's why I think Silman's observations make sense.
For example,
Great topic Dennis. The topic should not be just be tied to IM Silman, it could be have be made by any great chess trainer and still have the same relevance. Chess engines are like any tool that we humans create. It has both benefits and limitations. The engine's ability to find tactical opportunities in a position are outstanding. But engines still have blind spots even when they are playing at the +3000 level. A couple of years ago I found a position in a engine match where the top engines ( Houdini. Critter and Rybka for example) at the time rated it as a +2 win for white. Well the position was a dead draw even though White was up two pawns. I was able to easily prove it and posted the analysis on TCEC. How did a 1900 player (me) know it was a draw? Well I had a good understanding of chess, both through study (IM Silman was one the authors that helped me) and experience (I have played chess for over 55 years.)
I use this only as an example, many amateurs find these weaknesses. The engines can find amazing tactical tricks but they still have trouble understanding certain general principles. Do we forget that without an opening book engines can often serious misplay the opening. If calculation was the only thing engines do well, then how do they lose to another engine of similar rating? One last thought. Engines are really just programs create by humans and the reason they continue to improve is that humans have understood chess better and can translate chess principles into the code.
People seem to take this topic personally - I guess because they take their chess personally.
Really all these comments about whether or not a variation loses value because of corrections are missing the point - especially because most of these points don't have anything to do with amateurs playing too much with engines. Engines do not do the heavy lifting of building up your capacity to do hard work - in fact they train you not to work. Nor are you teaching yourself to evaluate a position on your own. I can work through engine variations and understand a bit about a chess game. But I didn't practice coming up with impressive chessic ideas at all. I didn't force myself to sit on my hands and visualize, or refute a move and find a new continuation or anything else that looks like chess.
You can complain that Silman is caustic or that his books are terrible now (they aren't and you'll need to throw out other classics like Art of Attack and Fischers 60 etc) or you can find excuses for why engines are good. The bottom line is, engines are bad for you because they let you trick yourself that understanding the engine evaluation is the same as playing chess. They are not the same, they aren't even close to the same. This is also why blitz is of limited value for improvement. You want to get good at chess, you have to practice playing chess. Playing a chess substitute (like blitz or evaluating with engines) isn't going to help you.
Engines have a place. They are great for correcting opening mistakes. They are an awful lot of fun to play with. But they are no substitute for building a capacity for hard work and generating ideas on your own. They never will be.
I guess everybody's in agreement that one should analyze a position/game without an engine first, and only then check with the engine, if one is wishing to improve his game (Silman's third point, partially), as obviously, "effort" is necessary for improvement. What surprises me, is that Silman (and everybody else in this forum, if I didn't miss something) seems to suggest that engines are only good for tactics, and that "checking with an engine" means checking for tactical errors (Silman's third and fourth points). Sorry, but that is simply untrue, and is a misrepresentation of the capabilities of a chess engine. I agree totally with Silman that some crazy tactical refutation shown by a computer does not take anything away from any of his books, or any of the classic annotated game collections (in terms of instructive value). But that is not even interesting or relevant for me. What is more interesting to me is that the engine can actually suggest alternative and maybe stronger human-like plans in positions that are rich in imbalances. Just increase the number of the lines given by the engine to four or five when analyzing an interesting (but "non-tactical") position and have a look at the evaluation values of each line. Even though Silman ridicules these numbers ("0.21" etc), their absolute meaning is sometimes less important that the fact that very often they will indicate that more than one line is playable in a position. And not only that, some of the lines suggested by the engine will resemble very much a human-like plan. If some of the lines are too computer-like, don't bother with them, just look at the ones which "make sense". Sorry, but I find this very instructive, once I make the connection between the position and the suggested line by the engine. I learn something, and next time I get a similar position, hopefully I remember this mini "plan pattern".
Silman is right that engines are often abused. Of course taking an engine evaluation of +0.21 literally is meaningless for practical play, and even with mutually best play almost like firmly believing a weather forecast predicting temperatures of 17 degrees (while a few degrees give or take are within uncertainty of the model). And patzers with engines criticizng GMs can indeed be annoying.
Silman is also right that it is best to analyze own games thoroughly without an engine, BUT: how many amateurs have the time and energy to do this? After this introduction, some words on how I use engines: My team just promoted to a higher amateur league, and I suggested to have a joint look at games from team competitions (Saturday afternoon) during the next club evening on Monday. Then my task became to prepare the analysis session. With such limited time (partly spent on deciphering the handwriting of my teammates ...), I have no choice but using engines to get a quick(!) overview of eight games - if only to find out at a glance which games deserve further analysis: some were decided by simple blunders, some were rather eventless draws. I consider this way at least better than no way.
Engine improvements fall into two categories: 1) moves/plans that a player with Elo 1800-2100 could have found himself. It's too late for this particular game, but might be useful in the future in similar structures. 2) "crazy" engine moves [this occurred at least twice in so far 32 games]. Even here, one can a) appreciate the beauty or complexity of chess, b) learn something from trying to understand why such resources exist. Basically I am turning things around: Silman says "first own analysis, then consult engines" - I am/we are doing it the other way around.
[Personally, I might be close to a player who ("at least 2100 rated") is, according to Silman, "allowed" to use an engine. My national rating was 2100ish in the past, currently it is roughly 1900-2000 (but European national and FIDE Elo ratings tend to be lower than USCF ratings)]
I really like Silman when he's writing about chess principles, but his style seems more than a little condescending at times.
That said, I think he's basically correct about engines being a crutch that keeps many players from improving. If you want to understand something or to learn a skill, there's no substitute for thinking it through for yourself in *ANY* area of human endeavor. If you use engines to help give you ideas as part of thinking a position or game through yourself, that has to be goodness. If you let the engine do your thinking for you, then there will probably be little practical gain. And that's only badness if you care about improving! Chess is just a game and hobby for 99.99% of us, after all... ;)
Playing chess with another human is a human endeavor. Human beings are finite with limited resources, and thus make mistakes -- even mistakes they would not have made if they played the same position on a different day. The recent back-to-back blunders in game 6 of the recent WCC match show it happens to super-GMs too. No one is perfect.
Practically speaking, if I'm playing someone of comparable strength (I'm 1702 USCF) the game is likely to be riddled with mistakes and missed opportunities. Sometimes one of those will decide the game, and sometimes not. If not, then the game will be decided by whoever did the better job of planning and executing. And that's where things like books come into play. If Silman (or anyone else) illustrates a principle with a flawed example, that doesn't diminish the value of the principle to me in trying to figure out what I might do while playing a chess game. I trust Silman as an IM to know what he's talking about if such-and-such type of plan can be successful in a position with a particular pawn structure or a particular piece placement. That there's a tactic in the position that an IM missed while writing the book is of little concern to me since (1) I'm never likely to play that exact same position and I am likely to play many similar positions where I have to try and figure out how to win, and (2) there's a good chance that if I played that same position against a comparably strong opponent we'd *BOTH* miss it and it wouldn't matter in the slightest -- just another one of those mistakes and missed opportunities under the bridge.
People aren't computers, so people shouldn't strive to understand chess in the same way a computer understands it! To the extent that we can play like computers, computer advice is beneficial. But if in tournament play you can't find all the ideas the computer finds, then in the comfort of your own home you should challenge the notion that the computer's preferred move should be your preferred move.
For the longest time I was of the belief "chess is 99% tactics" and I had no interest in learning opening tabia or positional chess of any kind, since I could outcalculate most opponents, and against other opponents I'd either get checkmated in the middlegame or survive to a worse endgame. I found endgames and tactics fascinating and didn't appreciate opening systems. It took many losses to stronger players to understand how chess is appreciable beyond endgames and calculation that I found so fascinating.
Among many great, insightful comments I want to cite the first one which made an interesting point:
“..as we've seen in advanced tablebase endgames, "perfect" chess is often unparaphrasable, unconceptualizable: it simply is itself. Or to put it another way, as I've long thought, chess is 100% tactics, not 99%. Strategy is an incredibly useful shorthand, but it has no objective existence, even within the structure of the game.“
My own chess experience and skills are fare too poor, to make my own assessment on that point. (I am also limited in expressing my thoughts in English) I just wonder whether a real good strategical thinking chess player would be able to find patterns, or even concepts spending some time with seemingly senseless tablebase sequences, which would allow him to understand this special x-piece endgame to much better degree. Whether this effort would be worth his time is another question. He might never be able to use his insights in real game. In that sense I would actually put (a small) question mark behind the statement above.
Ralph, thanks for the kind words!
"I just wonder whether a real good strategical thinking chess player would be able to find patterns, or even concepts spending some time with seemingly senseless tablebase sequences, which would allow him to understand this special x-piece endgame to much better degree."
Well, Pal Benko and John Nunn have both made attempts to explain tablebase endgames, and you'd be hard-pressed to find deeper endgame thinkers than those two. :-) From what I remember Benko basically gave up in bafflement when confronted with some of the earliest tablebases in the 1980s and early 1990s -- there was even a Chess Life where he questioned one defensive move as a blunder, and had to issue a good-natured correction when his "improvement" was refuted with one of those unforeseeable waiting moves that computers excel at finding. I don't know what he may have written on the topic more recently.
Nunn has made more comprehensive attempts to explain the simpler tablebase endings, and from what I know has done an excellent job of it. I don't own his "Secrets of..." books, though I probably should -- they looked great when I browsed them in the bookstore, many years ago -- so I can't address the extent to which he's managed to extract viable general principles from these endings.
[DM: Do mid-to-upper-level club players and up still buy endgame books by authors other than Dvoretsky, Mueller and (to a slightly lesser extent) Aagaard and Shereshevsky? If so, why? That's a tiny bit snide, but while there are some decent books on practical endgames by other authors (e.g. Mednis) the quartet just named pretty much blows everyone else out of the water for both theoretical and practical endgames.]
Dennis, are you arguing that Benko and Nunn aren't on that worthy level?
[DM: It was more of an assertion than an argument, but yes.]
Surprising if so, as I've found they're two of the most rewarding authors out there, and thought they were very well-regarded. My copy of Benko's collection of Chess Life columns is so well-loved as to be falling apart (his column was far and away the best thing in that publication), and several of my Nunn books are nearing that stage. Or are you merely saying their work has been superseded?
[DM: I have the Benko book and it's very good. (As is his autobiography, though in a very different way.) He was a very strong practical player in his younger days, making the Candidates in 1959 and 1962, and a real endgame aficionado. Certainly one will benefit from his book. But much of his work was done before tablebases and before powerful engines came along, and even the best works of the pre-computer era have plenty of errors. Additionally, he isn't writing a systematically didactic work like Dvoretsky or Mueller. So I have no problem with his work - it's very good - but I don't see any reason to point someone towards it rather than Dvoretsky or Mueller.
With Nunn it's more a matter of taste. He is a very fine analyst who was in the vanguard when it came to using computers in general and tablebases in particular. I find him a far less effective pedagogue than either Dvoretsky or Mueller.]
I know Dvoretsky and Mueller's writing from years of reading their columns on Chess Cafe, though I don't own any of their books. They're both obviously brilliant, but I haven't been engaged by their work the way I have Nunn's or Benko's. (Maybe that's because they're so concrete!)
[DM: Maybe that's the issue. Have a look at their books - Dvoretsky's Endgame Manual in particular. I think you'll find it far more engaging than his columns.]
Jeremy,
what free chess program do you recommend?
[DM: This isn't Silman's blog, he hasn't contributed to the thread, he's suggesting that most people shouldn't use chess engines, and the post is over a year old.]