Irrational computer decision I cant understand.

[MJMcCready]

Hi there, so after abandoning online chess for good because it allows too many bad habits to creep in, I just play against Lucus Chess now against the engine Cinimmon 1.2c rated 1930 or so it claims. I like playing sacrificial stuff against it as it makes it much more fun. So once again along came the kitchen sink attack played by me, but because I ran so low on material and couldnt force mate, I opted for threefold repetition. It only had to repeat the same move once more but instead resigned. I've ran it through the engines which estimated the final position as 0.00 and the next line being -4.7 ( I was white). So why would it resign? I cant understand it at all. Here's the game anyway. I don't think I missed a win at any point and the engines confirm I most certainly did not. Why did it resign when I was going for threefold repetion?[Event "Play against an engine"] [Site "Lucas Chess R 2.18c"] [Date "2025.02.11"] [White "Marcus"] [Black "Cinnamon 1.2c"] [Result "1-0"] [ECO "B23"] [TimeStart "2025.02.11 18:16:05"] [Opening "Sicilian Defence: Grand Prix Attack"] [Termination "Resignation"] [TimeEnd "2025.02.11 19:00:47"] [TimeControl "600+5"] [PlyCount "51"] 1.e4 {[%emt 00:00:3.61]}{[%clk 00:10:1.39]} c5 {[%clk 00:10:4.99]} 2.Nc3 {[%emt 00:00:1.88]}{[%clk 00:10:4.51]} Nc6 {[%clk 00:10:9.99]} 3.f4 {[%emt 00:00:1.84]}{[%clk 00:10:7.67]} g6 {[%clk 00:10:14.99]} 4.Nf3 {[%emt 00:00:1.88]}{[%clk 00:10:10.79]} Bg7 {[%clk 00:10:19.98]} 5.Bc4 {[%emt 00:00:8.56]}{[%clk 00:10:7.23]} e6 {[%clk 00:10:24.98]} 6.O-O {[%emt 00:00:9.55]}{[%clk 00:10:2.68]} Nge7 {[%clk 00:10:29.97]} 7.d3 {[%emt 00:00:2.28]}{[%clk 00:10:5.40]} d5 {[%clk 00:10:34.97]} 8.exd5 {[%emt 00:00:2.29]}{[%clk 00:10:8.11]} exd5 {[%emt 00:00:20.42]}{[%clk 00:10:19.55]} 9.Bb5 {[%emt 00:00:4.32]}{[%clk 00:10:8.79]} Be6 {[%emt 00:00:20.02]}{[%clk 00:10:4.53]} 10.Ng5 {[%emt 00:00:33.30]}{[%clk 00:09:40.49]} Bd7 {[%emt 00:00:19.64]}{[%clk 00:09:49.88]} 11.Qe1 {[%emt 00:01:53.06]}{[%clk 00:07:52.44]} Nb4 {[%emt 00:00:19.28]}{[%clk 00:09:35.60]} 12.f5 {[%emt 00:03:8.12]}{[%clk 00:04:49.32]} Nxc2 {[%emt 00:00:18.94]}{[%clk 00:09:21.67]} 13.Qe2 {[%emt 00:00:8.10]}{[%clk 00:04:46.22]} Nxa1 {[%emt 00:00:18.58]}{[%clk 00:09:8.09]} 14.Ne6 {[%emt 00:04:16.79]}{[%clk 00:00:34.43]} fxe6 {[%emt 00:00:18.23]}{[%clk 00:08:54.86]} 15.Qxe6 {[%emt 00:00:20.10]}{[%clk 00:00:19.33]} Bd4+ {[%emt 00:00:17.90]}{[%clk 00:08:41.95]} 16.Kh1 {[%emt 00:00:11.32]}{[%clk 00:00:13.01]} a6 {[%emt 00:00:17.59]}{[%clk 00:08:29.36]} 17.f6 {[%emt 00:02:45.70]}{[%clk 00:02:39.77]} Bxb5 {[%emt 00:00:17.27]}{[%clk 00:08:17.10]} 18.f7+ {[%emt 00:00:34.66]}{[%clk 00:02:10.11]} Kf8 {[%emt 00:00:16.97]}{[%clk 00:08:5.13]} 19.Bh6+ {[%emt 00:00:8.20]}{[%clk 00:02:6.91]} Bg7 {[%emt 00:00:16.66]}{[%clk 00:07:53.47]} 20.Bxg7+ {[%emt 00:00:3.47]}{[%clk 00:02:8.44]} Kxg7 {[%emt 00:00:16.37]}{[%clk 00:07:42.10]} 21.Qf6+ {[%emt 00:00:2.17]}{[%clk 00:02:11.27]} Kh6 {[%emt 00:00:16.10]}{[%clk 00:07:31.00]} 22.Rf3 {[%emt 00:07:33.32]}{[%clk 00:04:54.09]} Bd7 {[%emt 00:00:15.80]}{[%clk 00:07:20.19]} 23.Qh4+ {[%emt 00:04:7.83]}{[%clk 00:00:51.26]} Kg7 {[%emt 00:00:15.55]}{[%clk 00:07:9.64]} 24.Qf6+ {[%emt 00:00:24.64]}{[%clk 00:00:31.62]} Kh6 {[%emt 00:00:15.27]}{[%clk 00:06:59.37]} 25.Qh4+ {[%emt 00:11:36.45]}{[%clk 00:04:16.79]} Kg7 {[%emt 00:00:15.02]}{[%clk 00:06:49.36]} 26.Qf6+ {[%emt 00:00:6.76]}{[%clk 00:04:15.03]} 1-0

[Roger de Coverly]

So why would it resign? I cant understand it at all. Here's the game anyway. I don't think I missed a win at any point and the engines confirm I most certainly did not. Why did it resign when I was going for threefold repetion?

Bad logic in the programming is an explanation. It may have been working under an instruction to avoid draws by repetition without having the caveat that the draw should be played when a loss is the alternative.

[MJMcCready]

So why would it resign? I cant understand it at all. Here's the game anyway. I don't think I missed a win at any point and the engines confirm I most certainly did not. Why did it resign when I was going for threefold repetion?

Bad logic in the programming is an explanation. It may have been working under an instruction to avoid draws by repetition without having the caveat that the draw should be played when a loss is the alternative.

Yes, there may be truth in that, it does give you the option to choose when the computer should resign, like early or late. But still, resign whilst playing along with threefold repetition?

[Joey Stewart]

Reminds me of those rubbish old interactive chess boards in the 80s and 90s which would rather die then ever accept a draw and would force you to play hundreds of pointless moves at times

[MartinCarpenter]

The weaker engines tend to be weakened - as far as I can see - by being programmed to 'blunder' roughly every X moves.

That's probably normally seen in their taking a suboptimal but not insane move, but could easily produce something like this too. It's frankly fairly annoying to play against, not at all human like.

[Christopher Kreuzer]

The weaker engines tend to be weakened - as far as I can see - by being programmed to 'blunder' roughly every X moves.

That's probably normally seen in their taking a suboptimal but not insane move, but could easily produce something like this too. It's frankly fairly annoying to play against, not at all human like.

So how are the stronger and strongest engines 'weakened'? Do they have some way of ranking their options so they can select a suboptimal move? I mean, you could go by centipawn assessments, but that can still produce moves that look strange. And how do humans select a suboptimal move when trying to weaken their own play? (I have tried this, in some positions the options are just not there.)

[MJMcCready]

The weaker engines tend to be weakened - as far as I can see - by being programmed to 'blunder' roughly every X moves.

That's probably normally seen in their taking a suboptimal but not insane move, but could easily produce something like this too. It's frankly fairly annoying to play against, not at all human like.

So how are the stronger and strongest engines 'weakened'? Do they have some way of ranking their options so they can select a suboptimal move? I mean, you could go by centipawn assessments, but that can still produce moves that look strange. And how do humans select a suboptimal move when trying to weaken their own play? (I have tried this, in some positions the options are just not there.)

Okay, the position had been repeated twice already, the same move played again would have resulted in threefold repetition and draw agree. The suboptimal move suggested was g4 but I looked into that in detail and saw mate was not possible and being two pieces and a rook down I went for the draw. I thought about Re3, picking up the Knight but Nc6 in reply holds the fought and runs me out of checks. It left me no option to accept I didn't have enough material left to deliver mate with the defensive reosources it had. There were other lines where black could have sacced his queen for my rook left me with no attacking chances, and it still ahead on material, so I had to go for a draw, but since the position had already been repeated twice, why did it resign why repeating its last move would have got the draw? All othe options left me at best -4.7? The last two moves it played stopped my attack, play it again and draw. Why resign instead when the draw is already in the bag?

[MJMcCready]

Two more questions and do be please be honest, I won't take offence.

What strength would you say the computer was based on that game? It claims 1930. Sound about right? It doesn't to me.

Also what strength did I play at about (all honest answers unoffensive). I've played very little competitive or chess indeed for over.a decade now so it's hard assess roughly where I am abouts.Around 1900?

[Christopher Kreuzer]

I am not strong enough myself to assess your play there.

Maybe the computer program is not programmed for 3-fold repetition?

Interesting point found here:

https://www.cs.vu.nl/~wanf/theses/vrzina-bscthesis.pdf

Finally, there were some left-over steps to make the engine playable, such as a communication protocol, time management, and draw detection. The UCI standard was adopted to communicate with the engine, and time management was solved in a simple yet effective manner by taking 1/20th of the total time left. Draw detection was partially solved by preventing the engine from voluntarily doing three-fold repetition in a winning position. However, it remains unable to detect potential forced draws from the opponent. This leads to weaker engines capitalising upon this by forcing a draw while Tesseract is in a winning position. By testing Tesseract against other engines that were rated by CCRL, an estimated ELO range of 2400-2450 was deduced.

It seems that some computer engines are programmed to avoid 3-fold repetitions. Not quite sure why.

Section 5.4 'Draw detection' is relevant.

Turns out it is far harder than you might suppose to detect 3-fold repetition. Requiring the program to keep track of positions in the game and in its search tree and 'detect' the three-fold repetitions seems to impact on performance.

[Christopher Kreuzer]

I would have paid more attention in computer science classes 35 years ago (oh dear, tempus fugit) if I had known how relevant it would be to chess! I am not sure whether it is encouraging or worrying that people are writing (undergraduate) theses in 2023 titled 'Piece By Piece: Building a Strong Chess Engine'...

Pithy summary of computer chess:

The first chess engine dates back to 1957 when Alex Bernstein created an engine that runs on the IBM 704 [1]. Since then, major engines have shown up or fizzled out again, like the famous Deep Blue, Fruit, AlphaZero, LeelaZero, and the current best engine Stockfish.

[MJMcCready]

Maybe the computer program is not programmed for 3-fold repetition?


It seems that some computer engines are programmed to avoid 3-fold repetitions. Not quite sure why.


Turns out it is far harder than you might suppose to detect 3-fold repetition. Requiring the program to keep track of positions in the game and in its search tree and 'detect' the three-fold repetitions seems to impact on performance.

I suppose that may be possible yes especially since computers can't care whether they win, lose, or draw. I'd like to know who was aided and abetted with coming up with that one, since it's a normative aspect of the game, and chess is generally drawish anyway.

[MartinCarpenter]

The weaker engines tend to be weakened - as far as I can see - by being programmed to 'blunder' roughly every X moves.

That's probably normally seen in their taking a suboptimal but not insane move, but could easily produce something like this too. It's frankly fairly annoying to play against, not at all human like.

So how are the stronger and strongest engines 'weakened'? Do they have some way of ranking their options so they can select a suboptimal move? I mean, you could go by centipawn assessments, but that can still produce moves that look strange. And how do humans select a suboptimal move when trying to weaken their own play? (I have tried this, in some positions the options are just not there.)

I think they do what they can by limiting search and so on, but it's still too strong.

So you just have a random number generator trigger off every X moves on average. When it does it plays the second, third, whatever best move. Sometimes that's just slightly suboptimal, sometimes it's an absurdly crude blunder because there was only one halfway decent option.

At least that seems to be what the lichess SF levels do, I might be wrong and there are obviously many options!

[MJMcCready]

Well it certain does play some very odd moves sometimes. I put that down to gamesmanship because it lulls me into a false sense of security and makes think I can punish it and win but do more than once in a blue moon, that I cannot.

[IM Jack Rudd]

The way Chessmaster 2100 used to do it, back in the day, was to add noise to every position evaluation.

[MJMcCready]

The principle difference between computers and humans regarding my foray into pitting my wits against my electronic friend is that computers don't give up and throw the towel in if they start losing. Instead, they are resourceful and focused solely on the position at hand rather than the result they anticipate the most. But then they don't have emotions; therefore, it is to be expected somewhat if not then absolutely.