But I don't think copyright on visual images actually works like that, that it needs to be an exact copy.
If I draw my own pictures of Mickey Mouse and Goofy having a tea party, it's still a copyright infringement if it is substantially similar to copyright depictions of mickey mouse and goofy. (subject to fair use defenses; I'm allowed to do what would otherwise have been a copyright infringement if it meets a fair use defense, which is also not cut and dry, but if it's, say, a parody it's likely to be fair use. There is probably a legal argument that Copilot is fair use.... the more money Github makes on it, the harder it is though, but making money off something is not integral to whether it's a copyright violation in the first place, just to fair use defense).
(yes, it might also be a trademark infringement; but there's a reason Disney is so concerned with copyright on mickey expiring, and it's not that they think there's lots of money to be spent on selling copies of the specific Steamboat Willy movie...)
> There is actually no percentage by which you must change an image to avoid copyright infringement. While some say that you have to change 10-30% of a copyrighted work to avoid infringement, that has been proven to be a myth. The standard is whether the artworks are “substantially similar,” or a “substantial part” has been changed, which of course is subjective.
https://www.epgdlaw.com/how-can-my-artwork-steer-clear-of-co...
https://ogc.harvard.edu/pages/copyright-and-fair-use
This AI re-mixing stuff is so new, I think few legal observers would say they could definitely predict what the courts will do with it. Nobody really knew how the Google Books case, for instance, was going to go until it went.
- https://i.imgur.com/VikPFDT.png
I also don't know if I would anthropomorphize ML to that degree. It's a poor metaphor and isn't really analogous to a human brain, especially considering our current understanding, or lack thereof, of the brain, and even the limited insight we have into how some of these models work from the people who work on them.
P.S. I am not a lawyer.
Want to say that again?
The crux of the lawsuit's argument is that the AI unlawfully outputs copyrighted material. This is evident in many tests with many people here and on Twitter even getting verbatim comments out of it.
AI art, in the other hand, is not capable of outputting the images from its training set, as it's not a collage-maker, but an artificial brain with a paintbrush and virtual hand.