Hacker News new | ask | show | jobs
by CSMastermind 1379 days ago
There are "computer moves" which stand out vs human players. These normally show up in lines where there are many options of roughly equal value and the computer picks a move that is infinitesimally better but out of 'theme' with the position.

They can also show up when for instance there are multiple checkmates in a position. The computer will choose the one requiring the least number of moves even if it requires deep calculation and perfect play. Humans will just trade off material and go for an easy win.

Now that chess engines have started to use neural networks in move selection the amount of "computer moves" has decreased noticeably.

> if you were a top player and looking at the moves of an opponent, could you discern if the style was more similar to a top rated human or a top rated computer?

With a large enough sample size I believe that top players would be able to tell the difference. But that sample size is much larger than a single game or likely even the ~10 games being played in a tournament.

Edit:

Oh I should also mention that in the context of cheating with computers there are more signals to look at than the moves themselves. Time management is normally a huge giveaway for cheating. In online chess this normally manifests itself as players using the exact same amount of time for each move in spite of the positions being very different in terms of complexity.

In the match being talked about above Hans, the challenger, used a suspicious amount of time during the opening sequence. He played the opening moves in around 10 minutes which is weird because if he had memorized the lines he would have played them much faster. If he didn't memorize the lines then it would have taken him much more than 10 minutes to calculate it all.

2 comments

> There are "computer moves" which stand out vs human players. These normally show up in lines where there are many options of roughly equal value and the computer picks a move that is infinitesimally better but out of 'theme' with the position.

To elaborate on this, humans use pattern recognition to identify themes within a position. This is a shortcut that prevents needing to mentally brute-force your way down an enormous tree of possible positions. Elite chess-playing humans are very good at this, but are still very good at spotting potentially non-thematic (perhaps "surprising") moves that offer some quantifiable advantage.

Computers operate very differently from humans. They rigorously evaluate positions to absurd depths. They can examine lines further than 30 moves deep without too much time. Often the moves they come up align with the thematic ideas that humans have (after all, there's a reason why humans have identified these patterns). But at the end of the day, the computer isn't playing thematically. It only cares that the position at the end of best-play by both sides has the score most in its favor.

This leads to computers playing moves that humans would only come up with exceedingly rarely. And if a human came up with that move, there would generally be some clearly-identifiable reward that humans can pick up on several moves later. When a human player plays computer moves, those noticeable rewards are often missing. The cheating human makes a puzzling move, play continues, and many moves later their opponent is worse off. But even after serious analysis it's not entirely clear how that original move brought about this advantageous position.

Of course maybe the human in question really wasn't cheating and stumbled into a brilliancy. Perhaps it's even one they didn't even truly understand the ramifications of when they played it! But when a human makes several of those types of moves in a single game, or even across a single tournament, it brings about extreme suspicion.

If this interests anyone reading this, check out this Daniel Naroditsky video where he analyzes positions with some astounding engine moves which are very difficult for humans to see, but do have logical reasons for why they work: https://youtu.be/GdaU7wpOArs
If there is too little communication between you and behind-the-scenes cheating operator, you run the risk of choosing a move which requires perfect play for a long sequence, otherwise it would give away that you didn't understand the move.
This is a rather poor example. If there are multiple checkmates, humans are usually much more likely to see ones that require the least number of moves. Anyway if there are multiple checkmates, the game is irrecoverably lost by another player and it is easy to see on the board.
It's more complex than this.

The situation would be more like, there is a forced checkmate in 12 but it involves navigating a sharp position that is complex and any miscalculation could equalize the position.

On the other hand, you can simply exchange queens and rooks and have a clear winning endgame, but it will only result in a checkmate in 20 moves.

Almost any human would opt for the latter to avoid the risk. From the computer's perspective, there is no risk.

I tend to analyse my games after playing them, and I must say that NOT ONCE in a few hundred of games played at my level (~2500+ on lichess rapid) I have seen a forced mate in 10+ moves other than in end-game when there are very few pieces on the board left. And when you see such long checkmate, usually it doesn't really matter, because by then all is already clear on the board.

So no, I stand by my point that a specific choice of one of multiple possible checkmates is quite a poor indicator of using a computer to calculate moves.

That's surprising to me especially for rapid but you're better than me (2000 lichess). Under time pressure, if I have a passed a or h pawn and can simply trade down and promote I will opt to do that rather than try to calculate a deep mating combination.

> other than in end-game

This is frequently where such scenarios occur. Many end games are difficult for humans to play with absolute precision, even seemingly simple ones like knight and bishop vs. king. But when there are fewer pieces on the board is exactly when computers are able to perform incredibly deep calculations.

A good example is the notorious 30 move forced mate that Caruana "missed" in game 6 of his world championship match with Magnus, which occurred with only 3 pieces and 3 pawns left on the board.

To be honest, I'm lazy and rarely calculate more than a couple of turns: I just have a mental picture of what I want to achieve and seek ways to get there.

However, I do look up a post-game analysis, and if there are some mid-game deep checkmates, they show up in it, and it is extremely rare. The example you cite about Caruana/Carlsen game actually quite falls into the category of what I referred to as 'endgame mates' - and looking at it with a potentially cheating player in mind, such cases are rather irrelevant: a cheater will win a game or arrive to an overwhelmingly better position in the endgame so such extreme measures with 30+ precise move checkmates will not be necessary.

This is false. If there's a longer mate where every move from your opponent is forced (that is, only one more is legal or doesn't immediately lose) a human player will play that, as opposed to a shorter mate, in say 5 moves, where you have all sorts of different cases to calculate depending on what your opponent plays.
You should be more careful with such statements. Boldly claiming that something is 'false' and backing it up with a rather vague hypotheticals does not make your statement look good, considering that said hypotheticals are not supported by my own chess playing experience. Of course, I don't claim to be a really strong player or something, and my lichess rapid rating is only ~2580, so maybe your expertize in these matters exceeds mine.
That's a serious lichess rapid rating. Are you a titled player?
I don't think so. I only started playing online a couple of years ago to relax, but ~30 years ago I used to go to chess school for a few years and had a First class sportsman degree [0], and even barely qualified for a Candidate for Master of Sport after a couple of local tournaments, but it wasn't awarded to me because I simply stopped appearing: it coincided with me losing interest in chess as time limits on that level grew to ~2+ hours and that was far too much for my jumpy ass. So I don't think that First class title is still active, as even the country that awarded it to me - USSR - doesn't exist any more.

[0]: https://en.wikipedia.org/wiki/Unified_Sports_Classification_...