Computer Cheating in Chess Tournaments

[samsloan]

According to a report by Kenneth W. Regan cited above, the moves by a
2200 player will match the computer 48% of the time, by a 2300 player
50% of the time and moves by a 2700 player matches the computer 56% of
the time.

Thus, if a player previously known to be a 1900 player suddenly starts
making moves that match the computer 90% of the time we can be sure
that he is cheating.

Sam Sloan

[None]

Not sure, but confident. Maybe a 99.6% level of confidence

[MikeMurray]

And if the cash box is empty every day, and the book-keeper drives a new Caddy and takes periodic trips to Vegas, we can be sure he is embezzling.

It seems to me quite simple to construct an algorithm to evade Regan's screen. Call it the "Engine Manager".

Relative Blunders: When running a blunder check, the software often finds "blunders" where the winner passes up, e.g., a +7.50 move and makes one netting "only" a +4.25 -- a relative blunder that makes no difference (assuming competent play subsequently).

It's a relatively rare position that has only one or two plausible moves. Quite often, there will be eight or nine moves differing by only 15 or 20 centipawns.

First cut at Rules for the Engine Manager:

(1) Never make the top or runner-up move in the list of the engine's candidates, unless necessary to avoid loss. Scan down the list until encountering a candidate 15 centipawns worse than the leading candidate.
(2) At some random interval, select a candidate 40 centipawns worse.
(3) At some random interval, offer a draw when you have a 50 centipawn edge.
(4) Always make Relative Blunders where possible.
(5) If a sacrifice requires over "x" ply (20?) to demonstrate its correctness, choose an alternative.
(6) Allow manual override. Assume the operator is high Expert or better with at least some notion of "computer moves" and how to avoid them. A clever, patient operator would throw a game every now and then to avoid meteoric rating jumps.

Now, a real GM could probably examine a game played according to such an algorithm and suspect chicanery, but the game should pass automatic cheat screening.

[EZoto]

On Wed, 5 Jun 2013 06:47:27 -0700 (PDT), MikeMurray
<Smoke...@gmail.com> wrote:

yes but that is one game. Anand has a term that he made up and now
is catching on in chess. He would call a move Fritzy if he examined a
game used by a computer.

EZoto

[abc]

samsloan wrote:
> According to a report by Kenneth W. Regan cited above, the moves by a
> 2200 player will match the computer 48% of the time, by a 2300 player
> 50% of the time and moves by a 2700 player matches the computer 56% of
> the time.

That's only an 8% difference, over a range of 500 rating points.
Rather surprising imho, and it would seem to make the cheat-detecting
task more difficult.

> Thus, if a player previously known to be a 1900 player suddenly starts
> making moves that match the computer 90% of the time we can be sure
> that he is cheating.

Wouldn't a cunning cheater be able to avoid detection by making
moves ranging from the second to fifth-best computer move or so,
except for clear blunders, and still end up at mostly GM strength?

abc