Case study: generative AI and copyright
Last updated
Was this helpful?
Last updated
Was this helpful?
every person who contributed to the billions of creative artifacts that generative AI systems are trained upon. — Louis Rosenberg
+ everyone who shared, commented on or upvoted it
It’s reasonable to consider a “humanity tax” on generative systems trained on massive datasets of human artifacts. It could be a modest fee on transactions, maybe paid into a central “humanity fund” or distributed to decentralized accounts using blockchain. I know this sounds like a strange idea but think of it this way: If a spaceship full of entrepreneurial aliens showed up on Earth and asked humanity to contribute our collective works to a database so they could generate derivative artifacts for profit, we would likely ask for compensation. — Louis Rosenberg
...then I started googling individual sentences. It turns out most of them are near word-for-word reproductions of Wikipedia sentences. If the AI were a student, it would be flunked for plagiarism. — Janelle Shane
is constructed and determined by a (human) judge comparing their expectations to the outcome (output)
if the result meets expectations — ‘whale facts’ are correct, or the opinion piece is satisfactory in style and content — the judging editor may state that the text is ‘about external reality’, and publish it
If the result is incorrect, insufficient or inconsistent with the respective possible world (fictional or non-fictional) — e.g., the requested whale facts contain statements such as ‘dolphins also live in the desert’ (Shane 2020) or other statements that are false or nonsensical in relation to the external reality, the judgement may evaluate that the text is ‘about nothing’.
— Viidalepp, Auli. 2022. “The Semiotic Functioning of Synthetic Media.” Információs Társadalom 22 (4): 109–18. .
