[kmeisthax]

This would be an extremely novel mechanism of copyright litigation and I doubt it would fly in an American court with its' emphasis on highly individualized legal rights and obligations. And, if it did get accepted by the courts, that's halfway to an even crazier argument: that the MIT license only allows individual distribution to known parties; i.e. no hosting the code on a website or seeding it on BitTorrent, because that's not "small scale" and doesn't "involve a person".

[jongjong]

You can only seed it on BitTorrent if it comes with the license which identifies the original author and acknowledges their copyrights over the code. Also there is definitely an assumption that a human will read the license or at least implicitly consent to the terms before using or modifying the software. When ingested by AI, the author gets zero credit and no consent has taken place between any sentient being on either side of the contract... Or at least none that are legally acknowledged as sentient or having legal rights.

[bigbuppo]

And the thing is, you point out the easy out on this for similarly licensed code... a giant list of authors and contributors that may have code included in the generated output. It's a win/win for everyone. The original authors get their acknlowdgement, and the AI company gets to bill the users of AI for all the tokens for that multi-gigabyte copyright disclosure file.