Cheating in chess

[Wadih Khoury]

I'll PM you. But in old money, I suppose I'm not a massive believer in being able to improve from 170-ish level to cruising at 230-240+ type level 10 months later. Just seems a bit unlikely.

Thanks for letting me know.
I think this debate here shows why we need cameras, screen shares, etc... as we end up in a world of constant suspicion, and the possible false positives. The measures are not only a good way to protect honest players from cheats, but also to protect them from suspicion.

Juniors do win against strong players and lose against weaker players, but such is chess. You normally have a 10% chance of beating someone 400 elo higher than you, and some of these juniors have been playing multiple thousands rapid games (it's insane, I don't know how they do it), and even more blitz (and don't get me started on the tens of thousands played on bullet..)

As an illustration, I'll talk about an OtB example of junior variability: one of the 2 current French big hopes, Marco Materia (already has a norm at 12), was rated 2163 FIDE (!!!) in the 2019 U10 championship. He finished 4th, because he lost and drew against 1500s and even a 1477 (!!!). See crosstable.
http://www.echecs.asso.fr/Resultats.asp ... &Action=Ga

At the same time, he was already beating masters. That was over the board chess. Beating masters and losing to 1400s.
A few coaches told me that strong juniors have very variable results, because they have not yet experienced enough games and positions, so can take radically wrong decisions in unfamiliar positions. They also have different behaviours and play styles when playing strong players than when playing their peers. Against weaker players, they have to push for the win and sometimes mess up their attack, while against stronger players they often just need to wait for the right tactic to show itself.

Compound:
- variability of results
- the fact that the ECF and Fide grades are not 10 months out of date, but more like 20+ months.
- the intense training they may have had during lockdown (not friends, not football, no holidays, just home with a computer.
- the sheer volume of games which increases the statistical chance they will beat titled players (also don't forget that even titled players make big blunders)
And it gives a plausible hypothesis that some juniors may have explainable performance.

[Matthew Turner]

As always Wadih makes some interesting points, but it should be remembered that platforms do not know ECF or FIDE grades. They are also looking at performance in terms of move matching rather than results.
Fluke results happen, a junior might defeat an IM (perhaps because the titled player blundered) and then lose to a 1400, but in platform terms it is perfectly possible than the 'performance' was the same in both games.
'Breakout performances' happen, but they don't just emerge out of the blue, they are part of a trend improvement.
There are loads of rapidly improving juniors competing on chess.com and Lichess perfectly innocently without getting flagged. False positives do happen (very rarely), but not because one group of innocent players is particularly susceptible to being flagged.

[JustinHorton]

You normally have a 10% chance of beating someone 400 elo higher than you

Are you sure about that

[Pete Heaven]

There is an expectation that you would score 8 points from 100 games against someone 400 points higher rated, which is close, were it not for draws.

[JustinHorton]

That last clause is passably important though

[Wadih Khoury]

Correct, I did a bit of rounding and the correct words would be about points not wins. But in short 8 wins or 16 draws against a 400+ every 100 games.

Out of curiosity I went and checked an online calculator which gives 3% win and 10% draw against a +400. Which means roughly one game out of 7-8 you can get a result against a much stronger opponent.

Hence if you play 100 games, you have 95% chance of beating at least once the opponent. With 10 games, you still have 26% chance of winning at least once. So with thousands of players playing multiple games, we should be witnessing a significant amount of upsets on a daily basis.

[John McKenna]

Rebrand it as a FIDE Online World Championship and a multitude of sins will be covered.

I'm not sure what you mean. Anything online has Online in the title.

Nick, what you quote me saying above was in response to Paolo C's -

"... maybe it's just not possible to have an online competition branded as a world championship by FIDE."


Paolo left too much distance between the words " world championship", "FIDE" and "online".

I simply put them together in the right order.

"Online" has become so common (and over-the-board so rare) that it is now an implied default and does not even need to explicitly appear any longer - particularly in the UK.

Though there was still some confusion in the thread -

Re: £10,000 Scottish Junior Tournament

Where the following appeared -

"I hadn't realised from the original post that this was an online event via lichess..."

The expectation of a sudden return to over-the-board junior chess was no doubt partly to blame.

By the way, elsewhere on the forum, there has been some mention and criticism of FIDE's rival the ACO (Amateur Chess Organization) and it's use of the World Amateur & Senior Championship titles.

The same accusation of misrepresentation - due to the spectre of "external assistamce" in this case - could be levelled at FIDE with respect to its recent fabrication of onlne "World Championships".

(NB: I trust your standing with FIDE is secure enough for it to survive your intended participation in the ACO event in Oct.)

[Geoff Chandler]

Hi John,

Yes I thought the £10,000 event was OTB. As I said I could not read all the Times article and the
top thread when I posted was advertising a junior OTB event. So typical Chandler thinking 2+2 =5.

A lad on Facebook posted today in the British Chess News:

"There're programs which are designed to help people cheat in a way that evades detection.
The program makes some human mistakes but still plays at GM strength."

How can a computer possibly know what a human mistake is.
Getting a computer to suddenly decide now is the time to slip in a human blunder is beyond me.

[Matt Bridgeman]

Hi John,

Yes I thought the £10,000 event was OTB. As I said I could not read all the Times article and the
top thread when I posted was advertising a junior OTB event. So typical Chandler thinking 2+2 =5.

A lad on Facebook posted today in the British Chess News:

"There're programs which are designed to help people cheat in a way that evades detection.
The program makes some human mistakes but still plays at GM strength."

How can a computer possibly know what a human mistake is.
Getting a computer to suddenly decide now is the time to slip in a human blunder is beyond me.

There's one computer project I know has created a chess engine that makes human-like mistakes on Lichess. Not to cheat, but to give people a useful tool to create positions and play against - Maia;
https://maiachess.com

...for Maia fans, the developers told me the stronger versions are coming soon.

[Joseph Conlon]

My two cents on playing variability:

I think some of this comes down to playing style. To beat players much stronger than yourself, you need to play 'critical' moves - there's little chance (short of them actively losing the game) to defeat much stronger players just playing 'safe' moves. On the other hand, the same 'critical' moves will increase the chance of losing against lower-rated players, as if (to take a random example) you calculate a forced win through a decisive sequence six moves deep, there is a chance of you missing the decisive refutation on move 7 of the sequence.

I'm not sure how much this is a feature of juniors specifically - I think its also true with adults that some players are more likely to pull off upsets, and also more likely to lose against much weaker players. Given sufficient games though, ratings don't lie.

As OtB I think it will be interesting to see how much juniors have actually improved and how eg lichess ratings map onto actual ones. I guess my very rough estimate is that (say) 2200 lichess rapid is mapping onto about 170 ECF, but I can imagine this flicking 20 ECF grading points either way.

Aside: setting up the pairings for the OTB tournament I am running next week, I find myself wishing that the ECF had expressed the numerical values of the new rating systems in terms of the old 3-figure numbering. I will always have a sense of what a 130 or a 150 or a 170 player means, I am not sure I will ever manage that with 1300 or 1500 or 1800....

[Ian Thompson]

"There're programs which are designed to help people cheat in a way that evades detection.
The program makes some human mistakes but still plays at GM strength."

How can a computer possibly know what a human mistake is.
Getting a computer to suddenly decide now is the time to slip in a human blunder is beyond me.

It might depend on how you define a mistake. It wouldn't be difficult, for example, to program the computer to avoid complicated continuations when simple ones are available. Suppose the computer is clearly winning, let's say a piece or several pawns up. The computer calculates that's its mate in 15 if it sacrifices its queen. The computer could easily be programmed not to sacrifice material if it is clearly winning and it is going to be material down for more than, say, 5 moves. More generally, it could be programmed to never sacrifice material if it was going to be more than X moves before it regained that material.

[Roger de Coverly]

How can a computer possibly know what a human mistake is.
Getting a computer to suddenly decide now is the time to slip in a human blunder is beyond me.

The "cheat finder" programs operate by comparing the human choice to the top three choices of an engine. You could potentially program a chess engine to periodically select a move well down its list of best choices, provided that it was only a slightly worse choice or that the position was so overwhelming that anything sensible would win.

[Matthew Turner]

The "cheat finder" programs operate by comparing the human choice to the top three choices of an engine.

This is simply not true

[Roger de Coverly]

I will always have a sense of what a 130 or a 150 or a 170 player means, I am not sure I will ever manage that with 1300 or 1500 or 1800....

The "traditional" exchange rate was easy to remember.

225 = 2400
200 = 2200
175 = 2000
150 = 1800
125 = 1600
100 = 1400

[Matthew Turner]

With various different lines of best fit, the relationship has been updated over time, but it has been a constant that 200 equates to 2200. My experience is that the Lichess rating is pretty accurate, possibly 100 points inflated, but not really more than that. Hence, I would be surprised if many 170 players could maintain a 2200 rating.