Free versions available via standard ebooks and Gutenberg are often based on the copyright of the translation and so can be dated or just considered lower quality than other, more recent, translations.
Can you run the older translations through an AI to jazz them up a bit and maybe secretly steal the IP from other translations?
Or, since we're fudging the IP issue anyway, are the underground book pirate rings issuing AI translated versions of Harry Potter (or a more recent equivalent) into niche markets yet?
So while certainly not the case for most books, if you have a pirate Harry Potter then there's a fair chance that an actual human did the translation.
Unrelated, sometimes I'd get a fake unofficial chapter and then I'd have to decide on-the-fly whether Draco undressing Harry felt in line with J.K. Rowling's universe so far.
While I find AI impressive, I think the demonstration proof that it isn't yet at that level is that ChaosGPT etc. have not already destroyed civilisation.
(OTOH, that someone made ChaosGPT and set it running, is reason to try to stop anyone publishing any better models until they can be proven safe: we don't want to find out something has passed this threshold, whatever that means, via it ending civilisation).