It’s ultimately just statistics, and AI is far better than humans at finding patterns in massive data sets. They are perfectly suited to stylometry. The more text they have of a given user, the higher the probability of cross-referencing against identified users (Facebook, Twitter, etc).
Even if you never post publicly, the problem is western governments have been engaging in dragnet mass surveillance for decades, which means the US gov — probably many govs, especially your own — almost certainly have every text message, email, etc you’ve ever written; all are transmitted unencrypted. This means the US, and likely your own gov, have the highest probability of deanon-ing you. They also have the highest budgets ace access to perform such a task.
Again, it’s only a matter of time until 99% of “anonymous” accounts, above a certain word count, are cross-referenced; likely only a few thousand words. I suspect they could already do 99% of accounts above 10 or 20k words.
I’d be interested in this, do you have sources (also ideally to test this myself)?
Here’s a general primer https://programminghistorian.org/en/lessons/introduction-to-stylometry-with-python
It’s ultimately just statistics, and AI is far better than humans at finding patterns in massive data sets. They are perfectly suited to stylometry. The more text they have of a given user, the higher the probability of cross-referencing against identified users (Facebook, Twitter, etc).
Even if you never post publicly, the problem is western governments have been engaging in dragnet mass surveillance for decades, which means the US gov — probably many govs, especially your own — almost certainly have every text message, email, etc you’ve ever written; all are transmitted unencrypted. This means the US, and likely your own gov, have the highest probability of deanon-ing you. They also have the highest budgets ace access to perform such a task.
Again, it’s only a matter of time until 99% of “anonymous” accounts, above a certain word count, are cross-referenced; likely only a few thousand words. I suspect they could already do 99% of accounts above 10 or 20k words.