How to actually identify an author using a computer?

As with any statistical method, this requires lengthy texts. These are not always available. The text about Bitcoin itself is 3,756 words, less than the minimum threshold of one and a half thousand. But what can we do? Nothing prevents us from trying.

It turns out like this:

The Bitcoin white paper clusters with Wright. This result is of course not to be trusted, as is any result obtained on texts of less than 10,000 words. But it was interesting. Thanks, delta!

It happens that the delta works poorly even on sufficient volumes. It does not cope with texts of different genres (above I simply deleted the file with “The Inspector General”, there was no point in including it in the sample). And among many successful experiments, it suddenly turned out that the delta regularly breaks down on theological texts. I don't know what discursive stretches are prepared there, but the delta gets confused in the testimony, clusters the wrong thing, and generally demonstrates militant atheism. Although it would seem so.

And for the especially inquisitive, I can recommend a fascinating book by Lennon, B. (2018). Passwords: Philology, Security, Authentication. Harvard University Press. There the author, in particular, points out the historical and ideological kinship of cryptography and philology: “But the revolutions of 1848 swept away many of Europe’s ‘black cabinets’, and when the imperial United States created its first military intelligence cryptanalytic service in 1917, it recruited personnel from two seemingly unlikely places: Riverbank Laboratories, a private research foundation in Illinois whose staff included amateur scholars working to decipher Francis Bacon’s supposed secret authorship of Shakespeare’s plays; and university English departments, especially at Chicago and Yale, where the faculty included Stratfordian opponents of the Baconians in Shakespeare studies” (p. XVII).