Can you lose your free speech even when you are not prevented from saying anything? To answer this question, let us consider two ways in which governments could regulate AI chatbots in the near future.
First, imagine a government prohibiting the commercialization of AI chatbots that fail to meet safety benchmarks. For instance, chatbots should not provide users with detailed instructions for building weapons, generate dangerous conspiracy theories, or encourage self-harm. We’ll call this case AI Safety.
My second case is Political Bias. In this scenario, the government prohibits the commercialization of AI chatbots unless their developer can demonstrate that the chatbots are not politically biased and that the information they provide on controversial issues reflects a diversity of viewpoints.
Would any of these scenarios violate stakeholders’ moral right to freedom of expression? This, of course, depends on who the stakeholders are.
Consider, for a moment, that AI chatbots might possess speech rights. This may seem preposterous, but some scholars argue that conversational AI agents have a legal right to freedom of expression. Yet there are serious philosophical reasons to doubt that chatbots have a moral right to free speech. Indeed, chatbots arguably lack the properties required to be moral rights-bearers in the first place. For instance, Christian List has argued that we cannot reasonably grant AI agents non-derivative moral rights as they lack consciousness. Similarly, John Basl and Joseph Bowen suggest that (i) being conscious is a necessary condition of being a bearer of well-being and (ii) being a bearer of well-being is a necessary condition of holding moral rights. Certainly, the claim that AI cannot be conscious is disputed in the philosophy of mind. As Eric Schwitzgebel points out, competing theories of consciousness are likely to yield conflicting judgments about AI consciousness for quite some time. For the moment, let us then set this issue aside and consider other possible rights-bearers.
In the U.S. legal context, corporations have speech rights. Would those rights be violated in the two scenarios envisioned above? Here is a reason to think so. In Moody v. NetChoice, the U.S. Supreme Court ruled against Texas’s and Florida’s attempts to prevent social media platforms from engaging in certain types of content moderation. In doing so, the court held that defining and implementing moderation guidelines constitutes expressive conduct. As Justice Kagan argued:
When the platforms use their Standards and Guidelines to decide which third-party content those feeds will display, or how the display will be ordered and organized, they are making expressive choices. And because that is true, they receive First Amendment protection.
Of course, the owners of social media platforms are not the only ones who implement moderation guidelines. AI firms such as OpenAI and Anthropic also train their models to refrain from answering certain questions or to respond in specific ways. Consider, for instance, a chatbot trained to recommend that suicidal users seek professional help. In this case, my view is that training a model is at least as expressive as choosing specific content moderation guidelines. If we rely on the court’s reasoning in Netchoice, we might therefore conclude that there is a presumption against regulating chatbots.
But perhaps we should reject Kagan’s reasoning. Indeed, one can reasonably deny that the decision to implement moderation guidelines constitutes expressive conduct. On the standard account, expressive conduct must be intended to convey a particular message that an audience would recognize as such. When the owners of social media platforms enact a moderation policy against spam, will users understand that as a way of expressing any message? When AI developers train their model to avoid indulging users’ violent fantasies, will users believe that they are speaking? These are hard questions, and it would be presumptuous for me to pretend that I can settle them here.
So let us finally turn to those who most clearly have moral speech rights in the context of human-chatbot interactions: human users. Are these users’ right to freedom of expression threatened in the two scenarios considered above? How could it be, given that the regulations I imagined would not prevent them from asking chatbots anything they wish but only constrain chatbot speech?
The answer is that the right to freedom of expression is often understood to include a right to seek and receive information, which certain AI regulations could violate. Such a right is clearly enshrined in Article 10 of the European Convention on Human Rights, which protects Europeans’ freedom to “receive and impart information and ideas without interference by public authority.” According to Cass Sunstein, it is also constitutionally protected in the U.S.: “any restriction on speech, even by an entity that lacks constitutional rights, must be adequately justified, if listeners or viewers claim that they want to hear or see the speech in question.”
What are the philosophical foundations of this right? Recently, Rubén Marciel has argued that the right to information is grounded in two distinct democratic interests: (i) a citizen’s interest “in becoming well informed, in order to be able to competently advance her own political views” and (ii) her interest “in her fellows being well informed, so that they do not advance harmful or unjust policies.” Note, however, that Marciel’s view does not yield a clear judgment about AI safety and political bias. On one interpretation, the best way for the government to respect citizens’ right to information is laissez-faire: that’s the traditional story about the marketplace of ideas. On another interpretation, the right to information grounds a positive duty for the government to intervene in the information landscape. As Marciel explains, the government could, for instance, impose “regulations on private media, such as the obligation to fairly present different views on controversial issues.” Alternatively, it could censor discourses that compromise public security (see, for instance, Article 34 of the European Union’s Digital Services Act).
For AI regulation, everything hinges on which of these two readings you accept. Under the laissez-faire interpretation, the verdict is neat and tidy: Both scenarios I envisioned violate the right to information by preventing users from accessing information they actively seek. But under the positive-duty interpretation, these scenarios are no longer violations of this right; they might very well be what the government is required to do to protect democratic citizens’ access to information. This creates a conundrum for those inclined to object to AI regulation on free speech grounds: The very right that, at first glance, could be used to oppose regulation might actually entail it.
Of course, violating chatbot users’ right to freedom of expression might be justifiable when other morally weighty interests are at stake. This may be the case for regulations aimed at ensuring AI safety. My point is simply this: The regulation of chatbots matters for freedom of expression if we understand, as both Europeans and Americans do, the right to information as one of its key components.
So to answer the question with which I began: Yes, when you talk to chatbots, your free speech interests can be frustrated even when no one prevents you from saying a word.
Étienne Brown
Étienne Brown is an associate professor in the Department of Philosophy at the University of Ottawa, Canada. His research examines the ethics and political philosophy of the digital public sphere and new information technologies. He is currently coediting Philosophy of Free Speech: An Introduction, which is forthcoming with Routledge and includes the chapter, "Free Speech and Artificial Intelligence."
