I disagree with your draconian approach (but I understand where it’s coming from) because I do not think that issues like these should be just swept under the rug.
Ironically, machines autogenerating Seinfeld-like scripts literally cannot have intent.
We can go a step back to the intent of the creators of the system. I assume the creators did not intend to auto-generate distasteful and/or bigoted jokes. But it's Seinfeld-like. And it seems if it's supposed to generate bad jokes eventually it'll make some mean jokes.
It's something of an interesting thought experiment about intent vs. effect with publication of speech. All effect, no intent.
Prima facie, I can see the logic in treating the operator of the AI legally/socially responsible in the same way as if they had published the material themselves. If you unleash a bot that e.g. goes around writing defamatory book reviews, you should be liable for defamation. But I am not sure.
We treat negligence differently from malice in both criminal and civil contexts, and even within negligence there are gradations about whether or not a reasonable person would have foreseen the outcome.
I think LLMs slot into that model just fine.
Really curious how this applies to infinite, AI generated, content without oversight.
Like, what is the “intent” in this Seinfeld example. Can an LLM have “intent”?
It’s the intent of the LLM instructor/creator that matters for things like mens rea. And in the case of “no oversight”, that just sounds like typical negligence, like leaving a car in neutral when parking on a hill.