Cass Sunstein: evaluation

Transcript context

…Can’t you imagine an AI company that generates AI answers, but a second before they’re released to the human interlocutor, they just have some other human who works for them modify the answer a bit. Not materially, and maybe we don’t know how much, but they can truthfully say this is co-authored output. Then they say it’s all protected by the First Amendment because it’s AI and humans, and de facto, you end up with complete AI First Amendment protection. I think the way in is to get concrete on what the speech is and what the reason for regulating it is. If I am working with AI — let’s suppose I’m working for a political candidate, and it’s mostly AI, but a little bit me, and it ends up being the words of a presidential candidate — that can’t be regulated. The idea that you can regulate a politician’s speech because AI played a significant role, or maybe the entire role — that’s very hard to defend. Now, we would have a frontiers issue, I guess, if there’s a law — it seems crazy, but a law — saying that no politician may give a speech which is entirely produced by AI. That would be a new question. It would probably come out unfavorably to the law. If, on the other hand, deepfakes are restricted on the ground that, and when, they give a misleading picture of some person in a way that is either libelous or libel adjacent, then it might well be regulable. If writing code is First Amendment-protected, and again, you can push back on that, but does open-source AI then just get off scot-free?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.