Re:OpenAI's justification for why training data is fair use,...
by ang_cire on Thursday October 05, 2023 @02:03PM
(#37781856)
See, for me this section also jumped out, but for the opposite reason.
They are intentionally being deceptive about the aim of AI training. The aim is not to create a model, for the sake of having a model, the aim is to create a model that produces content for "direct human consumption".
I could use their argument to say it's fine for me to illegally download movies because I intend to simply create a corpus of works, and whether that corpus will be consumed by humans afterwords is (by their logic) irrelevant. Or that I should be able to download them because I only intend to transform them into AMVs.
You still have to legally obtain the original work for Fair Use to apply, and in many cases they are not legally obtaining them. "Publicly-accessible" has been ruled repeatedly in copyright law not to inherently imply "legal to copy", but that is exactly what they're claiming should be applied here.
They are intentionally being deceptive about the aim of AI training. The aim is not to create a model, for the sake of having a model, the aim is to create a model that produces content for "direct human consumption".
I could use their argument to say it's fine for me to illegally download movies because I intend to simply create a corpus of works, and whether that corpus will be consumed by humans afterwords is (by their logic) irrelevant. Or that I should be able to download them because I only intend to transform them into AMVs.
You still have to legally obtain the original work for Fair Use to apply, and in many cases they are not legally obtaining them. "Publicly-accessible" has been ruled repeatedly in copyright law not to inherently imply "legal to copy", but that is exactly what they're claiming should be applied here.