Challenges for humans and computers

by Frederic Friedel
5/10/2020 – Today we want to bring all our locked-in friends something unusual. This article is not going to make you stronger, it will not improve your strategic wisdom, or inform you on what is going on in the (virtual) tournament scene. It is meant to purely entertain – just cause you to smile or chuckle. It includes what we claim is the easiest chess problem ever composed (picture). Enjoy.

ChessBase 15 - Mega package ChessBase 15 - Mega package

Find the right combination! ChessBase 15 program + new Mega Database 2020 with 8 million games and more than 80,000 master analyses. Plus ChessBase Magazine (DVD + magazine) and CB Premium membership for 1 year!


The human challenge

Let us start with two warm-up puzzles for humans. What could be easier than finding a mate in one, or a move that does not mate in one?


The first is quite easy, and fairly well-known, the second (with three promoted black bishops) is tricky. Note that on both diagrams above you can try out moves, and the built-in engine will play a defence if there is one. So I will not be giving you any solutions, but you can enjoy watching Jennifer Shahade explain things in this Sinquefield talk.

Now to trying not to mate your opponent.


I have shown our readers position 3 before. See if you can solve it in ten, twenty or thirty seconds (the human challenge). Position 4 is a little more difficult and requires lateral thinking. I got both from Werner Keym's latest book, Anything but Average, where they are numbers 238 and 239 (of 374 highly entertaining problems). Once again, try entering moves – if you find the one that does not mate the engine will be able to play a counter-move.


Position 5 is, I claim, the easiest problem ever composed! Why?

Position 6 I took from a copy of an old magazine Werner Keym sent me. In it twinkel-eye Dawson tells us that the player with the white pieces, on the move, refused to accept the resignation of his opponent. Instead he wagered that he would not win the game. He then played on and actually lost! Not by overstepping his time, resigning, or playing an illegal move. So how could he lose?

Humans and computers

Now for a genuine challenge for humans – although it is of interest to know if computers find the key move and, more importantly, how long it takes them to understand that it draws.

But first some background: during a Chess960 World Championship, 2009 in Mainz, tournament arbiter Hans Secelle of Belgium was showing a position by D Djaja (below left) to the chess players who wandered by. It was from a study by an unknown (to us) composer, with the initial play left out.

In his book "The King" GM Jan Hein Donner writes: "White makes one more move and it’s a draw! Keres, the two Byrnes, Lothar Schmid, Bisguier and I sat staring at this position for more than half an hour. We couldn’t find it. Can you?" Indeed none of the chess players in Mainz were able to solve this position, and Hans was convinced that computers would also fail miserably. So I bet him a glass of whisky at the bar that Rybka would find the key move. Not necessarily understand why it holds the draw, but it will find the move by discovering that all other moves lose badly and quickly. Hans lost his bet, Rybka found the right strategy in one minute and two seconds.


I will not show you the solution to the Djaja problem at this time – just tell you that when showing the solution to the GMs in Minz Hans Secelle (picture) would give the first white move and then speak for about 15 seconds. After that the GMs would nod and smile and say, "yes, that is right. Very neat!" Computers will find the key move, but will they realize it is a draw (0.00)?

Please try it on your computer with your favourite engines, and post data on the results in the feedback section at the bottom of the page

There are puzzles which can be used to fool certain programs, while others may find the solution with a little help. The study by Matous for you to experiment with. Our silicon pals are often content with 1.Qd6+ Kg8 2.gxh7+ Kh8 3.Bxa5, which hardly yields more than a draw. There is, however, a very clear and forced win with a wonderful second move that is thrilling to see. Can you and your Fritz, Stockfish, or whatever, work it out? Mine cannot find it, but sees it is a win in less than a second after I show it.

Coming shortly to a news page near you: genuine challenges for computers.

Editor-in-Chief emeritus of the ChessBase News page. Studied Philosophy and Linguistics at the University of Hamburg and Oxford, graduating with a thesis on speech act theory and moral language. He started a university career but switched to science journalism, producing documentaries for German TV. In 1986 he co-founded ChessBase.


Rules for reader comments


Not registered yet? Register

cikic cikic 6/9/2020 08:33
Chess engine Crystal solves many of the positions that are difficult for other programs, including Matous etc. Pity it is only 64bit.
BKnight2003 BKnight2003 5/18/2020 05:46
(Spoiler alert!)

1. cxd6 e.p.
kvh kvh 5/17/2020 08:14
@BKnight2003: other than e7# what is the other pawn move ?
Still not getting.
BKnight2003 BKnight2003 5/16/2020 07:00
@kvh: there are two legal pawn moves.
kvh kvh 5/14/2020 04:13
Problem No.4, Wilhelm Kluxen, Die Welt 1947. 8 legal moves are there for white- 7 knight moves and 1 pawn move. All these moves mates black. I am wondering what can be the answer. Eager to see what am I missing. Please provide the answer.
EatMyShorts EatMyShorts 5/13/2020 03:02
To follow-up on what @jeeky1996 stated in his comments to the original article regarding Lc0 never finding the perpetual, the latest Fat Fritz running on my 24 core, 48 thread Threadripper 3960x w/2x GIGABYTE AORUS GeForce RTX 2080 SUPER GV-N208SAORUS W-8GC [Backend — Multiplexing; Backend Options — (backend=cudnn-fp16,gpu=0),(backend=cudnn-fp16,gpu=1); Threads — 4], pushing approximately 85 kN/s, never finds the perpetual either.

Stockfish 11 running on 24 cores/48 threads w/48GB+ hash finds the perpetual in the blink of an eye.

What gives? Has a fatal flaw in NN engines been uncovered? Where are you when we need you, Uncle Albert (Silver)??
PWintershoven PWintershoven 5/13/2020 09:19
Problem n0. 6: Is it a legal move to promote your pawn to a piece of the opposite colour?
Ellery Bann Ellery Bann 5/12/2020 02:49
@albitex "Note: If I increase the hashtables the Djaja problem is solved even faster by Stockfish." What was the orginal hashtable size and the later increased size?
JoshuaVGreen JoshuaVGreen 5/11/2020 06:56
@gurugsk, yes, that's the intended solution. However, in @Frederic's phrasing the "actually lost" part is strongly bound to the "played on" clause, and as no wager is mentioned there it should really apply to the game. Moreover, it's not at all clear that a wager was actually in effect to be lost -- it's possible that White offered it but Black rejected. Wikipedia's phrasing is much better.
gurugsk gurugsk 5/11/2020 06:00
The white played on and Won the Game therefore he LOST the bet.
JoshuaVGreen JoshuaVGreen 5/11/2020 04:15
@Frederic, for Position 6 you stipulate:

"He then played on and actually lost!"

Compare that to Wikipedia's phrasing (in my comment below) and you'll see why I think yours isn't entirely fair.
Frederic Frederic 5/11/2020 03:48
@Thomas Beuman: You are right, Thomas, overstepping the time cannot be how White lost. Should I tell you guys?
Zvi Mendlowitz Zvi Mendlowitz 5/11/2020 07:01
The author of the "Author unknown" problem is V. Röpke (Skakbladet, 1942).
Zvi Mendlowitz Zvi Mendlowitz 5/11/2020 06:46
In Fabel's problem, what is the purpose of the black knights (since the solution is the same without them)?
MJFitch MJFitch 5/11/2020 03:41
YES we understand why it's a "Draw"...The Night, Pawn & King protects key squares allowing the Rook to slide up and down the G-file giving "Perpetual check"
Thomas Beuman Thomas Beuman 5/11/2020 01:43
About position 6: it occurred to me that, if White were to run out of time in this position, he would not lose the game. As it is stated in article 6.9 of the rules of chess: "[The] game is drawn if the position is such that the opponent cannot checkmate the player’s king by any possible series of legal moves." So it would be a draw!
Frederic Frederic 5/10/2020 11:21
On my ChessBase 15 Stockfish 8 gives 0.00, but so do Fritz 17 and Fritz 16, and even Fritz 13. Instantly. Are they getting it from the cache? Also: do you guys understand exactly why it is a draw? I will tell some stories from 2009 about top players wrestling with the study.
NYTed NYTed 5/10/2020 09:12
Update: The Android version of Deep Fritz 15 will also find the solution, just not as quickly as it's Stockfish 11 counter part. ;-)
albitex albitex 5/10/2020 08:44
Instead, the Matous problem n.8 is not solved by any of my engines. However if we play the first two moves (1. Qc8 Kg8 2. Bc7!), immediately any engine find checkmates in 11.

Note: If I increase the hashtables the Djaja problem is solved even faster by Stockfish (about 30 seconds - 1. Nh6 val 0,00). Curiosity, the old Rybka 2.3.2 finds 1. Nh6 in about 1.5 minutes, but with an evaluation of clear advantage to Black.
Kapawner Kapawner 5/10/2020 08:43
The idea in the Djaja study belongs to M.Kliatskin 1924.
albitex albitex 5/10/2020 08:38
Djaja problem n. 7
1. Nh6! After this move the black King
falls into a perpetual check-out network.
My Stockfish 11 on a dual core takes
about 50 seconds to find the move (with draw valutation):
1. Nh6! Rxa7
(1... a2 2. a8=Q Rxa8 3. Rg7+ Kc6 4. Rg6+ Kb5 5. Rg5+ Kc4
(5... Kc6 6. Rg6+ Kd5 7. Rg5+ Ke6 8. Rg6+ Kd5 =) 6. Rg4+ Kb5 =)
2. Rg7+ Kc6 3. Rg6+ Kb5 4. Rg5+ =
NYTed NYTed 5/10/2020 08:29
@Frederic, The Android version of Stockfish 11 spots it quickly. See Post on Talkchess Forum:
JoshuaVGreen JoshuaVGreen 5/10/2020 05:32
@Frederic, I'm not convinced that your phrasing of the stipulation for Position 6 is precise enough to make the intended solution viable. Wikipedia provides a better version:

In the position in the diagram, Black had decided to resign (although it wasn't his turn). White advised his opponent not to give up so quickly. "But I am bound to lose, and there is nothing I -- or you, for that matter -- can do", said Black. But White insisted: "I'll bet you $100 that I can lose this game!" So the two made the bet and White actually lost. He did not resign, lose on time, or anything like that. How did he lose?
Frederic Frederic 5/10/2020 04:01
@jeeky: I would be interested to see if any program shows a near 0.00 evaluation for the Djaja key move. I know that many engines will play it because everything else loses, and it is deemed to lose more slowly (obviously, since in reality it does not lose at all).
jeeky1996 jeeky1996 5/10/2020 02:54
I tried the Djaja study with Stockfish and Lc0, and surprisingly Stockfish found it within 30 seconds while Lc0 couldn't find it at all, even after playing the move. This could very well be an anti neural net position......any Lc0 users got the same results?
Perdurabo Perdurabo 5/10/2020 02:40
The official FIDE term is Fischer Random not Chess960.