Re:European Commission Calls for Pirate Site Blocking Around...
by jeroenhd on Wednesday May 24, 2023 @10:58AM
(#36059039)
That's still an open question. The AI people here sure claim it doesn't.
The typical workaround is that scientific research is usually exempt from copyright. If you train a model for research, and then use that model for your business, your business didn't take part in the violation step.
Unless the model you trained is a derived work, of course. If it is, you need to pay for every scrap of copyrighted work you use, same as with music samples (fair use probably doesn't apply here but IANAL).
There are two American lawsuits about this going on at the moment that I know of. I think the lawsuit against the stable diffusion people can bring the clearest answer to this question, though I wouldn't be surprised if the case goes through various courts because of the implications for the AI industry as a whole.
Actually having to pay for the stuff you use would ruin a lot of AI startups. Then again, having to pay for the stuff you use also ruined Napster.
The typical workaround is that scientific research is usually exempt from copyright. If you train a model for research, and then use that model for your business, your business didn't take part in the violation step.
Unless the model you trained is a derived work, of course. If it is, you need to pay for every scrap of copyrighted work you use, same as with music samples (fair use probably doesn't apply here but IANAL).
There are two American lawsuits about this going on at the moment that I know of. I think the lawsuit against the stable diffusion people can bring the clearest answer to this question, though I wouldn't be surprised if the case goes through various courts because of the implications for the AI industry as a whole.
Actually having to pay for the stuff you use would ruin a lot of AI startups. Then again, having to pay for the stuff you use also ruined Napster.