A note from Re:Create: AI and Copyright: From Excused to Justified

One of the most frequently invoked and most compelling policy arguments for permitting unlicensed AI training is that obtaining specific permission to train on billions or trillions of copyrighted works is impossible and throttling AI development with impossible demands would have starkly negative economic and national security consequences. As one prominent commentator recently put it, “[Y]ou just can’t do that because it’s not do-able.” But that argument can still fall flat if it isn’t clear that there is also a strong moral argument for training without copyright holder permission. Luckily, there is a strong moral argument in favor of unlicensed AI training, based on the freedom to learn and to create.

We are all free to learn from copyrighted works and to create new things based on what we’ve learned. What if The New York Times could sue you for sending a Slack message to coworkers repeating a fact you learned about in the morning paper? Or if Wikipedia had to pay every media company that owns an article in the footnotes before its editors could draft and publish a new entry? That would be a radical expansion of copyright, turning big media companies into roving censors and tax collectors. And that’s what licensing AI training amounts to: a demand for payment or permission to make and share something new just because a copyrighted work was involved somewhere in the process that led to its creation. That’s not just impossible or impractical; it’s immoral. It threatens freedom of speech, freedom of inquiry, and free enterprise. Fair use protects these uses not just because they’re expedient; it protects them because they are justified.

Obviously, Re:Create members have explained in many contexts and at great length that a whole panoply of important social and political questions are raised by AI technology. Fair use isn’t the end of the discussion about AI’s benefits and challenges, but it does provide a normative as well as a practical answer to the question, “Should training on copyrighted material require permission from copyright holders?”  

Archives