I don’t understand how an AI reading a bunch of books and rearranging some of those words into a new story, is different to a human author reading a bunch of books and rearranging those words into a new story.
Ok, let’s say for now that these things are actually similar. Is a human legally allowed to “rearrange those words” in any way they want? Not really, because they can’t copy stuff like characters or plot structure. Even if the copy is not verbatim, it has to avoid being “too similar”. It’s not always clear where the threshold is; that will be judged in court. But imagine if your were being sued for copyright infringement because of perceived similarities between your work and another creator’s. You go to court and say “Well I torrented the plaintiff’s work and studied it with the express intent to copy discernible patterns in it, then sell my work based on those patterns”. As long as the similarities are found to be valid, you’re most likely to lose. The fact that you’ve spent years campaigning how companies can save a lot of money by firing artists and hiring your pattern-replicating service instead probably wouldn’t help your case either. Well, that’s basically what an honest defense of AI against copyright infringement would be. So the question is, does AI actually produce output too similar to its training data? Well, this is an example of articles you can find on the topic…
So based on the above thoughts, do you feel like we hold AI generation to the same standard as we do human creators? It doesn’t seem so to me.
But there’s a lot of reasons why we should hold AIs to higher standards instead. Off the top of my head:
AIs have been created exclusively to replicate patterns in existing works. This is not the only function people have. So we don’t have to wonder whether similarities between AI inputs and outputs are coincidental. We don’t have to worry about whether overbearing restrictions might inadvertently affect some other function.
AIs have no feelings or needs. We don’t have to worry about causing direct harm to them and about protecting their rights. Forbidding a person from reading a book just in case they copy elements from it is obviously problematic, but restricting AI’s access to copyrighted work is not directly harmful in the same way.
Ok, let’s say for now that these things are actually similar. Is a human legally allowed to “rearrange those words” in any way they want? Not really, because they can’t copy stuff like characters or plot structure. Even if the copy is not verbatim, it has to avoid being “too similar”. It’s not always clear where the threshold is; that will be judged in court. But imagine if your were being sued for copyright infringement because of perceived similarities between your work and another creator’s. You go to court and say “Well I torrented the plaintiff’s work and studied it with the express intent to copy discernible patterns in it, then sell my work based on those patterns”. As long as the similarities are found to be valid, you’re most likely to lose. The fact that you’ve spent years campaigning how companies can save a lot of money by firing artists and hiring your pattern-replicating service instead probably wouldn’t help your case either. Well, that’s basically what an honest defense of AI against copyright infringement would be. So the question is, does AI actually produce output too similar to its training data? Well, this is an example of articles you can find on the topic…
So based on the above thoughts, do you feel like we hold AI generation to the same standard as we do human creators? It doesn’t seem so to me.
But there’s a lot of reasons why we should hold AIs to higher standards instead. Off the top of my head: