Why are the computer Ratings So Inflated?

Sort:
MammalBrain

Not only are the engine ratings inflated  (by 600 points in my estimation) but the play of  the  computer  at a  fixed  level  seems to  vary greatly. It seems like it  factors in a  certain amount of inaccuracies, mistakes and blunders but  the errors  don't seem to be much like the common types of errors of a human player my Rapid  rating is 900-1050. but  I  have  beaten the  engine  at 1500  many times, and  just  beat  the  engine  at  1600 - but it  wasn't  my  best  game and I was even definitely losing before  the   engine made a crakky  move.

tygxc

Artificially weakened bots err, but in a non-human way.

StockOfHey

Is it hard to make an AI that thinks like a human?

jofchess

Or it is the other way round: chess.com ratings are deflated, and the bot rating is realistic? Rapid mean is only 632 now, with most players in the 400 range in the distribution. So if you can play 50:50 against the intermediate bots, then you are an intermediate player, independent of the chess.com ratings that have gotten lower and lower over time. (The bots play in mean their rating, but not exactly in every single game)

chrislamuk

I've beaten the 1800 bots a few times and they do blunder horribly. But the Li bot at 2000 seems to be a different beast... she/it plays a pretty mean game and won't give you much for free. I only play Li now and I've beaten her just once!

chrislamuk

I used to play the bots in Fritz (remember that?) and again I got to around the 2000 level so there seems to be some science in these bot ratings? wink

magipi
tygxc wrote:

Artificially weakened bots err, but in a non-human way.

It is notoriously difficult to tell from a game if it was played by a human or a machine.

Also, "erring in a non-human way" is something that doesn't exist. Humans make every possible mistake in chess. And then some.

tygxc

@9
Top engines are much stronger. Weakened bots are artificially weakened: programmed to err so as to be weaker than they would be.

@8

"It is notoriously difficult to tell from a game if it was played by a human or a machine."
++ I think it is easy to tell.

"Humans make every possible mistake in chess."
++ For humans there is some flawed logic in every mistake.
For a weakened bot it is just random.

magipi
tygxc wrote:

"It is notoriously difficult to tell from a game if it was played by a human or a machine."
++ I think it is easy to tell.

Okay, so let's try. Here's a game from chess.com. One or more of the participants might be human, and one or more of the participants might be a bot. Can you tell which side is what, and how do you do it?

chrislamuk

The temptation is to say both are bots but I guess they're both humans!

tygxc

@11

Two weak players for sure. That makes it harder.
When both are stronger, then it is easier to discern the bot.
Maybe this is a bullet game?
Probably two weakened bots.
5 Bh3 defies logic
12...Ba6 instead of recapture defies logic
13 Rb1 defies logic

magipi
tygxc wrote:Two weak players for sure.

Probably two weakened bots.

So are they players or are they bots? Come on, it is easy to tell, isn't it?

tygxc

@14

My guess two bots, except if it is a bullet game.

jofchess

white could be a weak bot, because making too bad moves with both rooks, black could be human, winning easily in the end

magipi

It was a 3-minute blitz game played by 2 super weak humans. At least white is almost certainly a human (he's very active on the forums), but black might be anything. Who knows?

It is extremely difficult to tell if a chess player is human or not.

 
PromisingPawns

@magipi now show a game which is above 1500 strength, then let us guess whether it's a human or bot.

StockOfHey

"Who is sussy baka?" happy

StockOfHey

Is someone suspicious if they make a lot of sound moves then suddenly they made a huge blunder?

tygxc

2 humans rated 130-159 playing 3|0? So play any legal move, but play it fast...
Even less weak humans in 3|0 a viable strategy is: play any legal move that does not lose material.
The point is a reasonable human usually recaptures, even if it is bad.
A reasonable human sometimes forgets some hanging piece, but rarely puts it en prise.
I seriously doubt weakened bots pass your Turing test

Bots and humans of reasonable strength in a reasonable time control.

darlihysa

free engines are boring and all the same!! you cant find an engine with extreme attacking or defending tendencies or an engine that makes blunders testing your calculating skills or a drawish barbell engine to kill your muscles while training!!