The goal of “Strong Artificial Intelligence” (hereinafter “strong AI”) is to develop artificial intelligence that can imbue a machine with intellectual capabilities that are functionally equivalent to those possessed by humans. As machines such as robots become more like humans, the possibility that laws intended to mediate the behaviors of humans will be applied to machines grows.
In this article the three authors assert that the First Amendment may protect speech by strong AI. It is a claim, the authors state in their abstract, “that discussing AI speech sheds light on key features of prevailing First Amendment doctrine and theory, including the surprising lack of humanness at its core.” And it is premised on an understanding of a First Amendment which “increasingly focuses not on protecting speakers as speakers but instead on providing value to listeners and constraining the government.”
The first substantive section of the article considers justifications for free speech rights for AI speakers, both positive and negative. Positive justifications embrace the potential usefulness of AI speech to human listeners. According to the authors, AI speech can contribute to human meaning-making and construction of selfhood, and can produce the sorts of ideas and information that can lead to human enlightenment. Negative justifications for free speech rights for AI speakers reflect views which deeply distrusts governmental regulation of speech. The Supreme Court has broadened its views of free speech protection in part based on its doubts about the government’s ability to competently balance social costs and benefits pertaining to speech, especially when driven by censorial motives. The authors conclude that whether it is providing benefits to humans or remaining free from government constraints, AI speech can reasonably be treated like human speech under most existing First Amendment principles and practices, because humanness of the speaker is neither a stated not implied criteria necessary for speech protection. The only exceptions are theories of the First Amendment which are explicitly predicated on the value that free speech has for humans.
The second section of the article explains in more detail that First Amendment law and doctrine are largely inattentive to the humanness of speakers. It contains the observation that corporations famously receive speech protection, rebutting any presumption that innate or prima facie humanness matters to First Amendment rights, even though human autonomy and dignity are values free speech is intended to protect. Humans may need to be part of the equation, but having them as background beneficiaries maybe enough for the First Amendment to attach. The authors further argue that strong AI may in the future be credited with sufficient indicia of personhood to warrant inclusion even in speaker-focused speech protections.
Next the authors discuss whether possessing human emotions is or should be a prerequisite for a speaker to claim First Amendment protection. Not surprisingly, they conclude that AI is growing increasingly affective, while free speech laws ignore emotions, protecting cruel, nasty, racist, sexist and homophobic speech regardless of the emotional damage it might inflict. They repeat the point about corporations having cognizable speech rights, and remind readers that the two key concerns of contemporary free speech jurisprudence are whether the speech potentially has utility, and whether the speech is something the government has no right to silence. If the answer to either question is yes, the speech is protected.
The authors then contemplate whether the speech of other nonhuman speakers such as animals could be ascribed First Amendment protection, once the slippery slope of AI speech protections is sufficiently iced. No, they conclude that unlike AI, animals are not intended to serve human informational needs like computers are. This section of the article gave me a flashback to my law school Evidence class, in which I learned that animals cannot lawfully be declarants nor can their speech constitute hearsay. I’ve since seen and read many legal dramas that flout this well-established legal principle. I suspect this is because of an assumption that audience members like it when animals testify in court enough to forgo accuracy. Animals seem inherently honest. AI beings like robots probably evoke more mixed reactions because of the range of ways they are depicted in popular culture. Commander Data from Star Trek: The Next Generation always seemed trustworthy, but HAL 9000 from 2001: A Space Odyssey will kill you.
The authors then discuss doctrinal and practical objections to First Amendment protection of AI speech. Courts might find a way around the fact that AI speakers cannot be said to have culpable mental states when evaluating and ruling on defamation claims. Judges could, for example, treat AI speakers as dependent legal persons or find another way to facilitate litigation in which an AI speaker is the plaintiff or defendant. Should an AI speaker be found liable, it could be unplugged.
The fourth section of the article looks at what the limits of AI speech protection might be. Free speech protection is already quite expansive, say the authors, but there might be a way to formulate limiting principles including outright regulation that apply only to the unique challenges posed by AI speech. This claim puzzled me a little, because it seems to pull in the direction of content-based distinctions. The offered analogies to regulation of commercial speech, and to professionals’ speech to patients and clients are only partly reassuring. Regulation of commercial speech is a thorny, confusing doctrinal morass, and the authors do not explain why or how courts would do better with AI speech.
Next, the authors note that what AI produces is likely to be characterized as expressive conduct (“or something similar”) rather than pure speech. This raises definitional difficulties not unique to AI in terms of separating speech related motives or interests from activities that can be permissibly regulated.
Finally the authors conclude that legal regimes have always managed to handle emerging technologies and we should expect this to continue with respect to AI speech. There may be a lot of complicated line drawing, but that’s the way it goes in First Amendment jurisprudence.
I enjoyed reading this engaging piece of scholarship very much. It is accessibly written, and the authors’ willingness to generalize about First Amendment law and policy is truly refreshing. Its central claim about the lack of importance of real human beings and their emotions to most free speech theory rings true and has relevance well beyond the strong AI context. The piece confirmed my own beliefs about the current state of free speech, and made me viscerally miss the late C. Edwin Baker, who spent so much time passionately arguing that the central purpose of the First Amendment is the promotion of *human* liberty. He’d have written a far feistier review essay for sure, challenging the authors to be activists who instantiate human liberty interests within the center of the First Amendment. But he would have appreciated the creativity of the article just as I did.
Margot Kaminski took no part in the editing of this review.