[Added later: I replied to myself below (:--) after realizing that the question is not whose speech is being protected, but whose speech, if anyone's, is unprotected.]
This is an informative piece, but it doesn't seem to account for the fact that AI-generated speech is not the speech of the corporation that developed the AI system or the developers who implemented it, nor is it the AI user's speech. So whose speech is being protected by the First Amendment when AI speaks? Assuming that the AI system itself is not legally considered a person with free speech rights (as is currently true, though that may change), there is no speaker whose rights need protecting.
That seems to make it reasonable to consider AI-generated "speech" a *product*, unlike the human-generated speech in a book, lyrics in a song, etc. (although that's a product in some sense if it's sold).
The analogy with libraries and internet sites that merely provide users with speech from others also seems questionable, given that AI systems don't do that -- they generate new speech. The fact that their "training" incorporates human speech (writing, etc.) doesn't validate the analogy since AI systems don't reproduce the speech they were trained on (though they may sometimes come close).
This also seems to call into question the importance of the distinction between an AI system "telling" a computer to do something that causes harm (e.g., to hack into another computer system) and an AI system telling a human to do something that causes harm (e.g., to commit suicide). When a human does those things, one clearly involves speech and the other doesn't, but when an AI system does those things, there's little or no difference from the AI's "point of view". We can view both things as "actions" or both as "speech" if we wish (if we consider computer "code" a form of speech).
I'm updating my comment above after realizing that the question is not whose speech is being protected, but whose speech, if anyone's, is unprotected. (If the AI companies claim that AI output is their speech and so deserves First Amendment protection, then both questions are important.)
I suppose I would question whether AI "speech" is speech in the sense of the First Amendment. As I recall, FIRE (?) has said (to paraphrase) that it's speech because it's a grammatical string of words (and sentences, paragraphs, etc.). But if someone constructed a mechanical contraption that generated grammatically correct output (in written form, for example), that seems qualitatively different from the human speech that Founders wanted to protect, and also from the human speech that the Supreme Court has declared unprotected; namely, the expression of ideas by people.
If there's no speaker (in the legal sense), there's no speech, and so there's presumably no question of legal protection or liability. If we *do* consider AI output to be speech, and if the AI system is not a legal person, it seems that the only possible speaker is the AI company (or the individual humans involved in creating and selling the AI system). But the AI output does not express their ideas, so that interpretation seems to not fit the First Amendment notion of speech.
If we do someday give AI systems legal personhood, then the question will be "Who is punished when it causes harm?"
"Well, if developers face broad liability for chatbot advice that later leads to harm, they will restrict what their models can say to minimize the chances that any output lands them in court."
Welcome to "Legal costs and OJ Jury judgements as policy making tool". The liability bar, overwhelmingly Democrat, figured out with cigarettes that they can impose unpopular ideas about what you should be able to buy simply by bringing lawsuits until the market cries uncle. Guns, medicine, AI, data centers, the list is endless.
The first remedy has to be "loser pays". "Loser's lawyer pays" would be even better, because one of the roles of a lawyer is to tell clients frankly "Your case is nuts, and you shouldn't bring it." After that, I'm not sure, but one thing is sure, our current system is broken.
[Added later: I replied to myself below (:--) after realizing that the question is not whose speech is being protected, but whose speech, if anyone's, is unprotected.]
This is an informative piece, but it doesn't seem to account for the fact that AI-generated speech is not the speech of the corporation that developed the AI system or the developers who implemented it, nor is it the AI user's speech. So whose speech is being protected by the First Amendment when AI speaks? Assuming that the AI system itself is not legally considered a person with free speech rights (as is currently true, though that may change), there is no speaker whose rights need protecting.
That seems to make it reasonable to consider AI-generated "speech" a *product*, unlike the human-generated speech in a book, lyrics in a song, etc. (although that's a product in some sense if it's sold).
The analogy with libraries and internet sites that merely provide users with speech from others also seems questionable, given that AI systems don't do that -- they generate new speech. The fact that their "training" incorporates human speech (writing, etc.) doesn't validate the analogy since AI systems don't reproduce the speech they were trained on (though they may sometimes come close).
This also seems to call into question the importance of the distinction between an AI system "telling" a computer to do something that causes harm (e.g., to hack into another computer system) and an AI system telling a human to do something that causes harm (e.g., to commit suicide). When a human does those things, one clearly involves speech and the other doesn't, but when an AI system does those things, there's little or no difference from the AI's "point of view". We can view both things as "actions" or both as "speech" if we wish (if we consider computer "code" a form of speech).
I'm updating my comment above after realizing that the question is not whose speech is being protected, but whose speech, if anyone's, is unprotected. (If the AI companies claim that AI output is their speech and so deserves First Amendment protection, then both questions are important.)
I suppose I would question whether AI "speech" is speech in the sense of the First Amendment. As I recall, FIRE (?) has said (to paraphrase) that it's speech because it's a grammatical string of words (and sentences, paragraphs, etc.). But if someone constructed a mechanical contraption that generated grammatically correct output (in written form, for example), that seems qualitatively different from the human speech that Founders wanted to protect, and also from the human speech that the Supreme Court has declared unprotected; namely, the expression of ideas by people.
If there's no speaker (in the legal sense), there's no speech, and so there's presumably no question of legal protection or liability. If we *do* consider AI output to be speech, and if the AI system is not a legal person, it seems that the only possible speaker is the AI company (or the individual humans involved in creating and selling the AI system). But the AI output does not express their ideas, so that interpretation seems to not fit the First Amendment notion of speech.
If we do someday give AI systems legal personhood, then the question will be "Who is punished when it causes harm?"
"Well, if developers face broad liability for chatbot advice that later leads to harm, they will restrict what their models can say to minimize the chances that any output lands them in court."
Welcome to "Legal costs and OJ Jury judgements as policy making tool". The liability bar, overwhelmingly Democrat, figured out with cigarettes that they can impose unpopular ideas about what you should be able to buy simply by bringing lawsuits until the market cries uncle. Guns, medicine, AI, data centers, the list is endless.
The first remedy has to be "loser pays". "Loser's lawyer pays" would be even better, because one of the roles of a lawyer is to tell clients frankly "Your case is nuts, and you shouldn't bring it." After that, I'm not sure, but one thing is sure, our current system is broken.