Chatbots have rights?

Critics of our American constitutional tradition have often charged that an 18th-century document cannot possibly continue to govern a rapidly changing modern world effectively. Thus far, however, our courts have done an often impressive job of interpreting and adapting short clauses from the Constitution in light of radically new problems. With the advent of artificial intelligence, though, they may be about to get their biggest test yet.

In a preliminary ruling in a high-profile case in Florida, Garcia v. Character Technologies et al., District Court judge Anne Conway was presented with the question of whether AI chatbots enjoy free speech protections—or at least whether users have a First Amendment right to receive the “speech” of such bots. The case in question is an ugly one for the AI industry. A 14-year-old boy, seduced by a Character AI simulation of a female character from Game of Thrones, developed a crippling obsession. When his parents took his phone away, he decided to end his life, with the apparent encouragement of the bot. The defendants—including Google, which helped build the platform—urged dismissal of the suit on the grounds that Character bots enjoyed First Amendment immunities.

Although this may seem absurd, one can see why the tech companies would make such an argument. Not only are chatbots an exploding industry with potentially massive liability exposures (other bots have been caught telling schoolchildren how to synthesize fentanyl), but AI is now running “under the hood” of an enormous number of mainstream platforms and programs. Indeed, in a broader sense, the algorithms that are used to sort your search preferences, curate your Facebook feed, and filter spam are forms of artificial intelligence. Understandably, the tech industry has fought hard to argue that such algorithms constitute the “speech” of the companies, just as human editors filtering content would.

In a high-profile recent Supreme Court case, Netchoice v. Moody, some judges were unconvinced. Justices Alito, Gorsuch, and Thomas put their finger on the nub of the problem: “when AI algorithms make a decision, ‘even the researchers and programmers creating them don’t really understand why the models they have built make the decisions they make.’ Are such decisions equally expressive as the decisions made by humans? Should we at least think about this?” This concern seems even more relevant to a platform like Character AI, where it is no longer a matter of an AI algorithm performing a task on behalf of the company in an opaque way, but of a freewheeling bot engaging in “conversations” that go far beyond the programmers’ knowledge or intentions.

Thus in the Garcia case, Judge Conway writes, “The operative question is whether Character A.I.’s output is speech, and speech is expressive. Speech communicates ideas. Speech has a message even when the message is not clear or is open to interpretation.” But unless we are prepared to concede consciousness to AI bots (as only a handful of Silicon Valley enthusiasts are), how could we possibly describe them as “expressing ideas”? Indeed, even if bots did somehow attain consciousness and engage in genuine speech, it would not be the speech of humans—which is the only kind of speech the First Amendment is concerned to protect.

That said, while the particular utterances of the bot may not be constitutionally protected speech, the companies will still argue that the platforms as a whole, and the characters that inhabit them, represent the artistic expression of the companies. In a series of lawsuits over the past generation around video games, designers have generally succeeded in avoiding the “product liability” that creators of dangerous physical products face by making such First Amendment arguments. Sure, they may have made risky design choices, but unlike the manufacturer of an inappropriately dangerous toy gun, these were expressive design choices, and thus extensions of their “speech.”

Clearly, we are skating on some very thin ice here, ice that has been created by the courts’ gradual broadening of speech to include “expressive conduct” over the 20th century. This has marked a dangerous departure from First Amendment originalism, since the distinction between “speech” and “conduct” underlies the basic logic of First Amendment protections. (As Justice Alito wrote in the Dobbs decision: “While individuals are certainly free to think and to say what they wish about ‘existence,’ ‘meaning,’ the ‘universe,’ and ‘the mystery of human life,’ they are not always free to act in accordance with those thoughts.”)

While we might well understand why a silent film actor should enjoy “free speech” protections even while not actually speaking, it is less clear that a software programmer should enjoy such protections for every line of code he writes, especially in a world where almost everything seems to run on computer code—including your refrigerator. For now, the courts are voicing growing skepticism, but no doubt some major legal showdowns are looming before we can expect this question to be settled.