I know this question has probably been posted and answered more than a few times on here, but I can’t seem to locate a satisfactory cache of information. Long story short : I play correspondence chess on two phone apps. How can I determine whether an opponent is using software to beat me?
It doesn’t amount to anything considering the games don’t count for anything but the curiosity is killing me.
The short answer is “you can’t”.
Various individuals and organizations have claimed they can detect computer assistance by comparing the moves of the game to the moves suggested by a strong computer engine (like Fritz), but there has been no formal test of this approach.
“You can’t” is a partial answer. It all really comes down to statistics. If you go over the game with a strong program and all of your opponents’ moves match the engine, then that may be reason to become suspicious. Suspicion, however, doesn’t mean he or she actually did it.
You may be more sure by analyzing many games, or a really long endgame. To figure out the pattern, you need a decent sample size of competitive moves, i.e. from the end of theory to the point where one player has a decisive advantage. A single 20 move miniature is rarely adequate proof of anything, except that the loser likely overlooked a decisive tactic.
Strong humans can match a leading engine around 70% of moves, on average, in a serious tournament. Yet anyone might play a game or two honestly and still match 100%. Just because a computer finds a deep tactic or suggests a bizarre trick doesn’t mean that some humans might also come up with the same move, even if only by accident. We might become much more certain by analyzing a significant sample of games with the same behavior. Nonetheless, it amounts to nothing more than circumstantial evidence.
Michael Aigner
And it should catch only those cheaters dumb enough to always (or nearly always) pick the engine’s top move. But, say a clever cheat proceeded something like this:
1 Pick the most distant alternative that’s not worse than “x” (20?) centipawns worse than the engine’s top pick
2 If you get an advantage greater than 300 centipawns, start raising the distance threshold, so you never go above 250 centipawn edge.
Following a procedure similar to this, your move would rarely coincide with the engine’s top pick, and your play would reflect a lot of “blunders” where you get sloppy when ahead and “overlook” various crushing moves.
Are the game(s) in question in a USCF tournament?
Alex Relyea
Using computers is part of the territory. I personally do not (what’s the point?), but it’s obvious many people “cheat” on real-time servers and even more consult with computers on slow chess websites. My chatting privs on ICC were recently suspended for calling someone a nasty name. This guy was rated 200 points lower than me (and was never rated above 1700 until that night) and beat me at 3-0 11 games in a row.
If you play long enough it will happen to you.
Alex, no these are just games on an Android app called Chess by Post. I also play on another called Chesspresso. (I like both of them and recommend them, despite a few minor flaws) I’m certain I’ve played and list to a few engines on both apps. I was just curious if there was a way to tell when. I don’t mind. I’ll play an engine and try to learn from it like I would any stronger opponent.
Thanks all for the great responses. The members here never fail to deliver.
The reason I ask is that most places allow computer assistance, like the ICCF. USCF doesn’t, and Alex Dunne says that if you suspect your opponent is using a computer, then let him know. See this month’s “Check is in the Mail” column, for example. I’m quite sure that none of my USCF opponents are using engines.
Alex Relyea
IMO, correspondence based chess is virtually dead in the sense of how it existed pre-computer. I just resigned a non-rated game on ICCF where my opponent was clearly using a computer. At around move 14 I decided to run the moves that far through both Houdini and Rybka. My opponents moves, out of the book which I took him very early on, matched the computer every move on one or the other program. So, I started consulting Houdini, Rybka 4 and Fritz 12 for every move. I ran the programs minimum of one hour for each program/position but was still behind by 1.5 to 2.5 paws on all programs until around move 40 when my position steadily deteriorated as my opponent made the best computer moves. That’s the reality in today’s world. If you play unrated games mostly as I do in chess.com, the chances are better of having an opponent who is not computer assisted. Once you have a few friends who you come to know as non-computer users, it becomes much more fun for all. I just won a game from a friend in Anand’s home town when he blundered and allowed a mate in one. About two months ago I made a howler that lost me a rook and the game. We can both laugh about our misfortunes.
While its possible that pure computer determination of moves is rampant in single “correspondence” games, it doesn’t make sense in high levels of ICCF play where there are a number of strong players and a significant number of games.
Frankly, there are too many games requiring significant analysis by a computer to have them all determined by a computer. One would have to own several (probably 4 or more) strong computers to accomplish this routinely, and maybe 6 or more to do it regularly.
While that’s possible, it certainly means that pure computer determination is not as prevalent as most people would think.
What is much more likely is symbiotic move determination where there human is assessing candidate moves, moving alone in some positions, testing odd candidate ideas (computers, even strong ones, can miss the point of a move because of the horizon effect and under value or over value the move until a significant distance down a variation. In essence, the person performs heueristics and the computer performs grunt analysis. I suspect that is what is prevalent in ICCF at this point. Pure computer (or even predominantly computer) move determination is likely just too expensive.
I know that on ICC I’ve had people call me a computer - and I see myself as at best average in 3 minute for my OTB strength. There are some truly great players out there who are not using computers.
And given the number of games in a short period, probability flux happens and someone can get on a roll. I’ve had a “weaker” player get on a roll and win a half doze in a row, only to storm back and take the next 10. When playing a large number of games in a short time, streaks happen.
Indeed. Between evenly matched players you would probably expect a streak of 11 wins roughly once in a thousand sets of eleven games! Perhaps more often with psychological factors. That’s not all that many…
This is what the top correspondence players do, and win games when their opponent constantly plays the “best” computer move. A lot of those guys have the resources to run multiple computers, but know the limitations you already pointed out on move evaluations far down the tree.
Computers are allowed for international correspondence rated play. I would bet the majority of non-rated players (chess.com, redhotpawn, etc.) don’t use computers either out of following the site rules or trying to take the moral high ground – some have loud aversions to using books and databases.
Good night all.
@Lafferty
I know your original post and some later pulled posts argued that anyone can use a computer to win. I noted that at the top levels a symbiotic relationship between player and computer was necessary to assist with computer heueristics. You scoffed at this (as opposed to Skoff-ed which is what a former USCF President would do.)
At any rate - so how about a two game match, 1 with each color, books, databases, computer assistance allowed, but no human assistance allowed. We’ll see if you are right - if computers are the great equalizer and if the player doesn’t make much difference. Should be fun. We can play on the beta server.
What would happen if you input the starting position into a computer, but tried to trick it to make it seem like just a random problem position? Of course a human would instantly recognize the starting position long before you could even finish setting it up. But would a computer sense that something looks familiar too? Or would it be “out of book” at move 1, so to speak? If it doesn’t recognize the starting position, how would it find the best move to play?
If you took away the computer’s opening book, it would indeed be out of book at move 1 and just treat the opening position like any other. When I have tried this with engines they tend to make pretty safe standard developing moves.
Note that nobody programs an engine expecting it to be out of book at move 1, so nobody bothers to try to make them play particularly well from the opening position. (Of course you have to be prepared to deal if your opponent starts 1.a4 or something, but it’s not like the out-of-book moves that engines come up with in responses to moves like that are particularly terrible.)
I heard report, several years ago, about a U.S. player in a domestic tournament being bagged for using computeer assistance. I suppose this could be done with spyware.
What will help you in correspondence chess is data bases not engines.
It’s difficult, but not quite impossible to tell. Someone rated 2200 on the server, with a USCF or FIDE rating of 1730, is almost definitely using a computer. Similarly, someone rated 1400 on the server who finds a crushing tactic after playing your pet variation perfectly for 18 moves is on something.
A friend of mine, who had his PhD, was rated in the 1700’s OTB and was well over 2200 correspondence - in the 1980’s. So he clearly was not using a computer. He simply had bad nerves OTB and was also good at research.