Maybe I’m missing something obvious, but isn’t there also an adversarial angle to account for, that is missing from this analysis?
That is, if it’s possible to get someone in trouble by linking their real identity to a piece of controversial content they published anonymously, it’s also possible to purposefully get them in the same trouble by publishing a piece that is specifically adjusted to be falsely recognized as made by that person.
I think, when solved for equilibrium, it would make a lot of such evidence inadmissible (would still be a very good hint to keep looking for more tangible evidence, though)? Whenever someone’s pseudonymous identity is exposed by this method, the real person could just say that it was someone else trying to pose as them.
I think the problem is that for most use cases you don’t need to prove it definitively. E.g. if companies start doing this as part of background checks for applicants they’re going to feed their resume and cover letter into a pangram like program which will say “84% match with reddit user weirdpornographyenthusiast463”, and they probably won’t check with the candidate. Even if they do they won’t necessarily find “someone ran a long term project to imitate me” persuasive
This is an interesting question. I suspect you’re right that if the technology exists to detect writing styles at the information-theoretic limit, then perhaps being able to generate writing styles isn’t far behind. As you say, I think this would perhaps make things inadmissible in a legal sense, though I feel like that probably only covers a small fraction of what people want from pseudonymity, though?
Maybe I’m missing something obvious, but isn’t there also an adversarial angle to account for, that is missing from this analysis?
That is, if it’s possible to get someone in trouble by linking their real identity to a piece of controversial content they published anonymously, it’s also possible to purposefully get them in the same trouble by publishing a piece that is specifically adjusted to be falsely recognized as made by that person.
I think, when solved for equilibrium, it would make a lot of such evidence inadmissible (would still be a very good hint to keep looking for more tangible evidence, though)? Whenever someone’s pseudonymous identity is exposed by this method, the real person could just say that it was someone else trying to pose as them.
I think the problem is that for most use cases you don’t need to prove it definitively. E.g. if companies start doing this as part of background checks for applicants they’re going to feed their resume and cover letter into a pangram like program which will say “84% match with reddit user weirdpornographyenthusiast463”, and they probably won’t check with the candidate. Even if they do they won’t necessarily find “someone ran a long term project to imitate me” persuasive
This is an interesting question. I suspect you’re right that if the technology exists to detect writing styles at the information-theoretic limit, then perhaps being able to generate writing styles isn’t far behind. As you say, I think this would perhaps make things inadmissible in a legal sense, though I feel like that probably only covers a small fraction of what people want from pseudonymity, though?