Over a decade ago, I realized that AI someday would be able to trace anonymous writing samples back to their original author. As I write this, I have zero doubt that the US government possesses software that can identify individuals from their distinct writing style. In other words, given sufficient corpus of known works (with representative word choices, grammar, punctuation, sentence structure, and other details) software can generate a model useful to determine the probability that another sample came from that same person. For example, I worry that my own comment history might be used to unmask my real identity someday, given that I have made vaguely seditious remarks here.
At this point in the game, I would not bet against someone here on HN coming up with a prototype to do this using COTS software libraries over the course of a weekend. If you do such a thing, please use your new power responsibly.
