Computer Moves

Try to find the best move in these two positions. I have reduced the problems to sort of candidate moves in order to make it easier for you. There is no need for deep calculation.

[fen size=”small”]r2qr1k1/1pp2pb1/3p2pp/1N5n/1PP1n3/p3B1PP/P4PBK/1R1QR3 w – – 0 19[/fen]

White to play

[fen size=”small”]3R4/4np1k/6pp/R3p3/1p2b3/5qP1/PP3P1P/4Q1K1 w – – 0 34[/fen]

White to play

Negativity

A few times recently I have come across a very negative mentioning of computer moves. With disdain people talk about things that can only be found by Houdini – at times with the idea of criticising an author (yours truly for example) or just believing that the tactics are outside human reach.

I would like to come with a different opinion.

There are only good moves and bad moves

Computers are better at chess than humans and will remain so. But we need to understand what the reason for this is. It is not because the programmers understand chess better than Magnus Carlsen or can write a simulation that can do so. It is also not because the computers calculate much further than Magnus; though they do so. This feature is an advantage, but Carlsen’s position understanding levels that playing field.

The computers are basically only better than Magnus (and to some extent the rest of us) because they do not miss things. Their bad day is very close to their good day. For the rest of us, even good days have a lot of bad moments in it.

But just because computers are better at chess than humans and at times see some fantastic combinations we would rarely find (in exchange for violating positional sense at other times), does not mean that there is anything special about the moves computers find. They are (most often) simply good chess moves. Some of them are beyond our horizon; some are not.

Learning from the computers

I frequently have to change the solutions to my exercises because my students see ideas that are better than those suggested by my engine. I did so just yesterday.

This is always very satisfactory. I am proud of the students and I am happy to have been saved from putting an incomplete exercise in a book. This still happens of course, but limiting the frequency is a big ambition of mine.

But the biggest additions to my annotations always come from the rich ideas my students have, especially when they are wrong. Because then I can add explanations and variations I had not thought of otherwise. And at times a real point to the exercise is revealed, which the computer did not tell me about.

It is very rare that my students complain about moves being Computer Moves. Most of the time they are annoyed when they miss something; or surprised, or amused. And so on. Very rarely do they complain about the source of the moves.

And yes, we often talk about the frequency of mistakes, more than the impossibility of solving things. For it is quite common that 2-3 solve an exercise that another finds entirely impossible to solve.

Zombie analysis and real analysis

If you use an engine as an oracle, you will not learn a lot from them. Their conclusions will be numeric and your ignorance will be total.

But if you learn to ask questions from the engine; try other options and debate them with the engine, you will have a much better understanding of chess. Especially if you have some principles to work from (which is what the Grandmaster Preparation series is all about).

I understand the criticism that there are not a lot of explanations in a book like Attack&Defence. There aren’t. It is based a lot on the feedback I got from the readers of the first few books in the series (especially Calculation, which is the book closest related to this book), but also because I did not want to repeat what I had already done over 700 pages with the Attacking Manuals. This is the work book, not the text book!

Look at this position (from the middle of a chapter in Attack&Defence):

[fen size=”small”]r2q4/pp2kb2/4p2Q/n2pNp1P/3P4/P7/P4PP1/1R4K1 w – – 0 26[/fen]

White could have reached this position in Hector-A. Matthiesen, Denmark 2012. The combination with 26.Rxb7+ fails to accurate defence with …Nd6, …Rg8 and …Bg6. But White can play 26.Rb5!! and there is no good defence against 27.Qg7. White wins in all lines.

Yes, this was pointed out to me by a computer. But is this really so difficult to understand? The example is from the chapter called Include all your pieces in the attack. Should I really explain that in the diagram position the rook does not have an active role and for that reason it needs to find a target.

I understand this is not an easy position (even though it is simple), but do not tell me it cannot be solved! I have had 2650s that have failed it and I have had 2300s that have solved it. Yes, this is a bit random and with most positions it would be the other way around. But there you go.

If you want to be a really strong player, it is very useful to train seeing this kind of ideas. They really matter. They are not impossible, but they are difficult.

Solutions

So are the three positions I gave at the beginning. The solutions are:

Wojtaszek – Sethuraman, Basel 2014

[fen size=”small”]r2qr1k1/1pp2pb1/3p2pp/1N5n/1PP1n3/p3B1PP/P4PBK/1R1QR3 w – – 0 19[/fen]

After 19.Bd4?! f5 20.Bxg7 Kxg7 21.Qd4+ Kh7 Black was more or less ok. All my students looking at this position found 19.Bb6!! cxb6 20.Bxe4 with a big advantage.

Debashis – Negi, Jalgaon 2013

[fen size=”small”]3R4/4np1k/6pp/R3p3/1p2b3/5qP1/PP3P1P/4Q1K1 w – – 0 34[/fen]

All my students went for this position with Black, thinking they were close to winning. For example 35.Kf1 Nc6 36.Raa8 Nd4 37.Rxd4 exd4 with a big advantage. Instead White equalises with 35.Qxe4! Qxe4 36.Ra7 and the knight is trapped.

To me 19.Bb6 is a far more surprising move than giving up the queen to avoid mate. I would even go so far as to call it a computer move; but who knows what this really means? In fact it turns out that the difficult thing to spot is the opposite of what I would have expected. Maybe we are starting to learn so much from the machines that we can go past this definitions?

Final words on this

In the book I am working on at the moment, we are very conscious about not giving lines where the computer found it, but my co-author was nowhere close to it. But there have been interesting situations where I have suggested something and he had not considered it; but I then give it as a puzzle in the office and 2-3 of the guys solve it in 2 minutes.

So, I really think that to a great degree, we simply miss stuff. To say that it is because the moves are unnatural or anything like this; is simply to reject our humanity (we blunder) and at the same time reject the real challenge – to concentrate more, to work more, to improve…

 

 

 

 

 

 

 

 

 

 

 

  1. TonyRo
    June 17th, 2014 at 15:07 | #1

    Great article, thanks for writing it! There are some interesting corollaries that are implied here but not specifically discussed. It seems to me that one of the core takeaways from this article is that in order to not miss moves like these in your calculations, you need a wider “search” for tactical ideas before settling into the normal, human examination of a few promising candidates. Your central takeaway here is that computers are better than humans because they never miss things, and this is definitely true. We also know they never miss things because they look at everything – you can turn on any engine and watch it scroll through the 40 or so legal moves in any position at a very fast rate. On the other hand, since human chess is based on understanding, experience, and intuition, it’s difficult to find such moves and never miss things. So perhaps a second, more important takeaway is that most players aren’t looking sufficiently at the wide array of forcing moves available to them? Thinking about all of this stuff reminds me very much of Hertan’s “Forcing Chess Moves”.

    I’d be willing to bet that if these positions occurred in most readers games, they’d miss 19.Bb6!! in Game 1 and resign Game 2, but by showing these problems under the header “Computer Moves”, you bias their thinking process and make it much easier to find the solutions – I found both in under a minute. Encouraging news for those who might want to improve their calculation abilities.

  2. Michael Bartlett
    June 17th, 2014 at 15:26 | #2

    Interesting enough I suffered chess blindness on Game 2. The only move I could see was the queen capture. I saw it as the only legal move on the board. The first position I saw the tactical motif but went for a different move with the bishop. I learned from that game to be alert to ways to wreck your opponent’s pawn structure through force.

  3. Michael Bartlett
    June 17th, 2014 at 15:27 | #3

    (my legal move I mean the only move that did not result in checkmate as a reply)

  4. Ray
    June 17th, 2014 at 15:39 | #4

    Great article again, thanks! I did find 19.Bb6, but not because I knew there was a ‘computer move’. If you solve chess puzzles regularly (like 5-6 times per week), you quite often encounter similar types of moves as this one, and to be honest I have never been annoyed because they were too ‘computer-like’. On the contrary, I generally can only marvel at the beauty of chess :-).

  5. June 17th, 2014 at 16:30 | #5

    I agree that the heading before the exercise biases the thinking… I guess that the ideal exercise book would be a combination of Calculation / Attack+Defense / Positonal Play /Strategic Play… with no headings. Of course we would still have ‘tactical looking’ positions hinting that we should look for tactics. But we would also have positions looknig tactical where nothing works and we just have to go on with a plan. And the converse where some unexpected tactic pops up in anotherwise quiet position.
    For instance, once the first position is given as an exercise, finding Bb6 is “ok-ish”. But in a real game? Not so easy. It also reminds me of a combination from Fressinet in Calculations, where he plays a stunning Bg6 move in a Meran. Again a straightforward one to find when looking at a tactical exercise, but VERY easy to miss in the flow of a game.

  6. Jacob Aagaard
    June 17th, 2014 at 22:40 | #6

    @TonyRo
    Thanks for the feedback Tony. Obviously when I make claims in an article on the blog, they have not necessarily been mulled over as much as things I would write in a book. Sometimes I am just making it up on the cuff 🙂 to see the reactions…

  7. TonyRo
    June 17th, 2014 at 23:03 | #7

    Wasn’t meant to be negative if you took it that way, I really enjoyed it! Provoked a lot of interesting thoughts about my own abilities and game that I mulled over during lunch.

    Have you ever thought of doing a book based on these random articles you post on the blog? There’s a ton of amazing content (and reactions that might be educational) on here already. I know a popular poker book company that has done a few of those, though I’m not sure about their reception or sales. Yusupov’s bright green “Chess Lessons” was like that, and it’s still one of my favorite books!

  8. Gary
    July 30th, 2014 at 17:12 | #8

    Dear Jacob,
    I like your article, in the sense that it is true what you say. Computer moves are nothing more than difficult moves for humans to find, most of the time. I would even say, the tactics are sometimes so advanced (in many lines it ‘just fits’, it ‘ just works’) that it’s not possible to write books on it. The tactics can be unique in that sense.

    However, my complaint is the following. Why work on these difficult tactics? If you have problems that even 2650 can’t work out, than why even bother? The supergrandmaster didn’t get so far by solving problems finding ‘computer moves’, they got good because they solved solvable puzzles with increasing difficulties. If you train your pupils to find ‘computer moves’ or very difficult moves to find, aren’t you training them to look for things in a normal game that simply costs too much time?

    Please respond.

    Greetings, Gary

  1. No trackbacks yet.

 Limit your comments to