AI Unmasks Anonymous Chess Players, Posing Privacy Risks (science.org) 27
silverjacket shares a report from Science.org: [A]n AI has shown it can tag people based on their chess-playing behavior, an advance in the field of "stylometrics" that could help computers be better chess teachers or more humanlike in their game play. Alarmingly, the system could also be used to help identify and track people who think their online behavior is anonymous. [...] To design and train their AI, the researchers tapped an ample resource: more than 50 million human games played on the Lichess website. They collected games by players who had played at least 1000 times and sampled sequences of up to 32 moves from those games. They coded each move and fed them into a neural network that represented each game as a point in multidimensional space, so that each player's games formed a cluster of points. The network was trained to maximize the density of each player's cluster and the distance between those of different players. That required the system to recognize what was distinctive about each player's style.
The researchers tested the system by seeing how well it distinguished one player from another. They gave the system 100 games from each of about 3000 known players, and 100 fresh games from a mystery player. To make the task harder, they hid the first 15 moves of each game. The system looked for the best match and identified the mystery player 86% of the time, the researchers reported last month at the Conference on Neural Information Processing Systems (NeurIPS). "We didn't quite believe the results," says Reid McIlroy-Young, a student in Anderson's lab and the paper's primary author. A non-AI method was only 28% accurate. [...] The researchers are aware of the privacy risks posed by the system, which could be used to unmask anonymous chess players online. With tweaks, McIlroy-Young says, it could do the same for poker. And in theory, they say, given the right data sets, such systems could identify people based on the quirks of their driving or the timing and location of their cellphone use.
The researchers tested the system by seeing how well it distinguished one player from another. They gave the system 100 games from each of about 3000 known players, and 100 fresh games from a mystery player. To make the task harder, they hid the first 15 moves of each game. The system looked for the best match and identified the mystery player 86% of the time, the researchers reported last month at the Conference on Neural Information Processing Systems (NeurIPS). "We didn't quite believe the results," says Reid McIlroy-Young, a student in Anderson's lab and the paper's primary author. A non-AI method was only 28% accurate. [...] The researchers are aware of the privacy risks posed by the system, which could be used to unmask anonymous chess players online. With tweaks, McIlroy-Young says, it could do the same for poker. And in theory, they say, given the right data sets, such systems could identify people based on the quirks of their driving or the timing and location of their cellphone use.
Is this an Onion article? (Score:2)
Re: (Score:3, Insightful)
Re: (Score:3)
It really amounts to more evidence of the broader issue. Which is the big data people assured us that if they just gathered enough information we'd suddenly be able to crack all these really hard problems.
The computers were going to correlate all the things and someone with some actual intelligence was going to scan down the list and identify the causative relationships. A new era of enlightenment was supposed to suddenly be upon us.
It hasn't really worked out that way. In terms of curing cancer research ef
It's been done for a long time now (Score:3, Interesting)
Identifying people online by their observable decision trees is nothing new. Analysis of the way people construct language they use for messages and chatting, to the number and type of plugins they installed on their browser, to the order of moves and style of play of a number of different games. This has been used for a long time now for advertising and surveillance. Those two things are, if you think about it, exactly the same thing. After all, what is ad tracking but commercially motivated surveillan
Re: (Score:1)
Arguing the logic or lack of for that matter when privacy and freedom are applied to evaluate AI decisions. Does it make sense to DO an evaluation of civil rights being violated in a certain argument WHILE the community will privately use the same argument against a persons for their own personal gain even if untrue OR is the entire argument just some convenient tool to apply the broken logic by the league of nations concerning "thinking machines" who's definition includes all non colonized lands, people a
Re:Is this an Onion article? (Score:4, Interesting)
Does anyone seriously care about whether anonymous chess play can be fingerprinted by AI?
Not really but it can probably be used to identify AnonymousCowards as well. That could be a bit more serious, eg. if governments start sending dissenters to prison based on anonymous online writings.
PS: "We didn't quite believe the results" - oh, really? I'd have been far more surprised if this couldn't identify people.
Re: (Score:1)
Not really but it can probably be used to identify AnonymousCowards as well.
e4, find me!
Re: (Score:2)
Does anyone seriously care about whether anonymous chess play can be fingerprinted by AI?
Yes. Every tiny erosion of privacy hurts us all. And many of those pieces of data can be threaded together. Once this AI knows whose basement an opponent lives in, what's next? AI swatting?
Re: (Score:1)
Competitive chess players often play online under anonymous accounts lest their play become too analysed. It's the same in many sports that can be played that.
More importantly, there is probably going to be a time where every post one made online can be retrieved by an a.i. who's simply fed a corpus of one's writing, — quite chilling indeed, especially in the knowledge that some might be executed for this in some jurisdictions for criticizing their governments who feel safe now behind anonymity techno
Re: (Score:2)
Might be an issue for people who don't want to reveal their identity for fear of discrimination, e.g. because they are from an unpopular country or because of their gender.
The bigger picture here though is that with a large enough set of data pretty much any behaviour can be used to identify people.
I have been doing that for years (Score:5, Funny)
Apparently each time I lost, the other player was some guy named 'Fritz'.
Re: (Score:2)
Wow, a chess-joke got noticed.
Been done a long time ago? (Score:1)
Better to detect cheaters (Score:4, Interesting)
With the COVID lockdown it was always assumed that chess players are not affected, they can play on-line as well as over-the-board.
However, speaking to a few players (2000-2300 ELO range) they all complain about the number of cheaters on platforms like Lichess.
Many of them have moved from 10min + 5sec rapidplay format to 3min + 0s blitz simply because it is harder for cheaters to play in that format.
The cheaters input the opponent's moves into some chess engine and then play the engine's reply against the opponent.
Even a free chess engine you can get on a phone can play at grandmaster level.
On the 3min + 0sec format at least you have a chance to win on time against a cheater.
Catching cheaters would be a much better use of AI than de-anonymising human players.
Re:Better to detect cheaters (Score:5, Interesting)
Lichess does use software to detect cheaters. I know of several kids who had accounts banned for that reason; one of them even admitted to cheating afterwards. They look at what's called average centipawn loss. The metric chess computers use to evaluate a position is measured in units of pawns, i.e. all else being equal, if I'm a pawn up, it's +2 and if I'm two pawns down it's -2. The evaluation functions are fairly subtle, considering positional features too. The traditional (pre-AI) computer algorithms were minimax algorithms: each move by white aimed to maximise this metric, and each move by black aims to minimise it. If you take the computer's evaluation of a position before and after I move, I will have played a non-ideal move frequently. The drop in evaluation is called the centipawn loss (i.e. in units of 0.01 pawns). A top level player might have an average centipawn loss of 10 per move. An amateur like me will be higher. A cheater will be lower. Hence, we can detect cheaters. You can get lucky over a game or two, but if you're consistently playing computer moves, you'll be detected. And yes, there are strategies to avoid detection, but they involve leaning on the computer less with obvious disadvantages.
I don't really get cheating at it. Wow, your computer is good at chess. Who fucking cares? I bet your car can beat me or even a top class runner in a race too.
Re: (Score:2)
They look at what's called average centipawn loss.
I would think the computer chess programs are deterministic. What is the argument against catching cheaters by automating queries to many engines and check if the player moves correlates to the replies of the computer engines?
Re: (Score:2)
They are not deterministic, as they will give different results simply depending on how powerful the hardware they run on is.
They will in fact give different moves based on how the kernel decides to schedule threads in theory on identical hardware, but this is obviously significantly less likely.
Re: (Score:2)
That is done routinely but not to catch cheaters.
When you see the moves of a game supplied so you can replay it, you also often see two figures supplied at the end of the game - how much white diverged from "perfect" play and how much black diverged. Those figures are frequently very low for top grandmasters - most of the time they play the moves the engines think are pretty much optimal. That is one reason there are so many draws at that level.
Re: (Score:1)
And yes, there are strategies to avoid detection, but they involve leaning on the computer less with obvious disadvantages.
So one can only be detected if one use the machine to play at a level higher than Magnus Carlsen and not when one use it simply crank up one's 1600 Elo to 2300?
Still does not seem particularly useful. — We way I read it is simply an automatic ban for anyone who plays at an inhuman level.
Re: (Score:2)
I can't *prove* anything, but there have been many times on Lichess where I am winning a game, then my opponent goes offline for a while, and they come back with a very strong counter in a completely different style than before...
Cheating takes the fun out of it for everyone. Its a *game*. Ratings are cool, but I doubt they are satisfying if you know they are fake...
The next headline: (Score:2)
“Forget about Google StreetView: Police identify more mobsters by their opening chess moves.”
This would be a real concern (Score:2)
Bobby (Score:1)
Bobby Fischer, we can finally find you!
Harder, better, faster, stronger (Score:2)
Pretty sure the AI would have no problem inferring what the first 15 moves were, given the board position on the 16th and assuming the players played (reasonably competently) to win.