I've read/heard modern chessengines play more like humans.Especially Rybka seems to have a 'human' style(I believe dyskamil has mentioned it a few times).Now,I don't understand how these machines work and was wondering if it's possible to explain,in layman terms,how they can make a sophisticated calculator play 'humanlike'?
Originally posted by Katastroofthey let it make mistakes now and then
I've read/heard modern chessengines play more like humans.Especially Rybka seems to have a 'human' style(I believe dyskamil has mentioned it a few times).Now,I don't understand how these machines work and was wondering if it's possible to explain,in layman terms,how they can make a sophisticated calculator play 'humanlike'?
Originally posted by Katastroofthey cannot be detected by moderators...
I've read/heard modern chessengines play more like humans.Especially Rybka seems to have a 'human' style(I believe dyskamil has mentioned it a few times).Now,I don't understand how these machines work and was wondering if it's possible to explain,in layman terms,how they can make a sophisticated calculator play 'humanlike'?
Originally posted by KatastroofEngines need a way of assessing how good or bad a position is. They do this by considering various factors such as material count; piece activity; pawn formation; king safety; etc. for both sides. Each of these factors is given a numerical weighting. For example, how should king safety be valued with respect to, say, having an extra pawn? This is a complex problem for programmers.
I've read/heard modern chessengines play more like humans.Especially Rybka seems to have a 'human' style(I believe dyskamil has mentioned it a few times).Now,I don't understand how these machines work and was wondering if it's possible to explain,in layman terms,how they can make a sophisticated calculator play 'humanlike'?
If the weightings are mainly chosen to reflect human reasoning then it may result in the engine playing more “humanlike”. This could be done by the programmer manually setting the weightings using his/her own judgement, or by analysing top human games and tuning the engine to be in agreement with the human assessments.
Of course, other aspects of the engine – such as its calculation ability – will still contribute to making it play like an engine. I think these can be modified to be more human also – e.g. make the search more selective rather than brute force – but they don’t account for the context of an engine playing both strong and humanlike, e.g Hiarcs is more humanlike than Fritz.
There is a “human” version of the Rybka engine. Kaufman, who tunes Rybka’s evaluation function, applied some of his own human reasoning within this version. But during initial testing some of these appeared to be suboptimal in terms of playing strength. Hence, the default Rybka applies some “not so human” weightings. Sometimes test results can disagree with human intuition. [Curiously, since the release of Rybka 3, I’ve seen some test results which favour the “human” version over the default. Time controls, hardware, too small a sample, etc. could all play a part here]
Originally posted by VarenkaThanks for that.I started to fear this would turn into yet another cheating thread 😞
Engines need a way of assessing how good or bad a position is. They do this by considering various factors such as material count; piece activity; pawn formation; king safety; etc. for both sides. Each of these factors is given a numerical weighting. For example, how should king safety be valued with respect to, say, having an extra pawn? This is a compl ...[text shortened]... over the default. Time controls, hardware, too small a sample, etc. could all play a part here]
So if I understand it correct the reason engines have trouble with the positional aspect of a chessgame,especially long term exchange sacs,is because it's hard to put a number on it.
And Rybka and Hiarcs are more humanlike because the programmers did a better job,or gave higher priority to this aspect,than the ones working on Fritz the rest of 'em
Ok,makes sense to me 🙂
Having compiled control stats for various engines over hundreds of games while a game mod here, I've seen no convincing evidence to suggest that Rybka is any more humanlike than other engines.
Although,there may be a qualitative case for it having a more human playing style, this does not translate in to higher expected match-up rates compared to other engines.
It is stronger than other engines, and plays well beyond the capability of any human player at RHP, so the idea that it's use is undetectable is bizarre.
Originally posted by GatecrasherBeing "humanlike" mainly applies in comparison to other engines, and even then the differences may be small. It does not mean that it is close to playing like a human. e.g. all the top engines display tactical ability beyond the capabilites of an OTB GM, regardless of how "humanlike" they are. "Humanlike" is just a subjective term that reflects upon the engine's style.
Although,there may be a qualitative case for it having a more human playing style, this does not translate in to higher expected match-up rates compared to other engines.
From the Hiarcs site: "HIARCS is famous for its human-like playing style". Maybe so in terms of its positional knowledge and weightings. But it will still calculate with the accuracy of a computer; it stills uses endgame tablebases and opening books; it never blunders or gets tired; etc. So maybe a little "humanlike" but very much still a computer! 🙂
My own research into how often correspondence masters match engines indicates that "human like" may well be down to the engine picking the same move as a human when there is no clearly best move. I found that both Rybka and HIARCS were more likely to give a relatively high match up rate in positions where there was no clearly best move. In other words, the engine is more likely to pick the same move as a human GM in positions where there may be several moves that the engine evaluates as approximately equal. In both cases this appears to be down to hardwiring positional knowledge into the engine's evaluation function.
This does not mean that a player who relies on these engine is undetectable, although a good player using one will not be so easily detected. The lack of blunders when playing many games will still be damning evidence.