People are falling in love with their chatbots, mourning deleted AI companions, and treating artificial systems as romantic partners or family members. As AI becomes more sophisticated, these relationships will become more common. And, as a result, so will political conflict over whether artificial systems deserve moral consideration.
To understand why, consider the underlying moral ambiguity. The standard markers of moral standing—the ability to suffer, self-awareness, rationality—are inner features that are not directly observed so much as they are inferred. With our fellow humans we manage to recognize moral standing among our fellow humans tolerably well because we share biology, evolutionary history, and behavioral and expressive cues that we can use, with reasonable reliability, to infer what others are thinking and feeling. With artificial systems we lack much of that and instead are left with behaviors and performance like the appearance of suffering, the appearance of attention, the appearance of inner life, each of which designers are incentivized to optimize.
The implication of this moral ambiguity is that the moral status of artificial systems will likely remain epistemically ambiguous for the foreseeable future, perhaps indefinitely. The relevant philosophy of mind and cognitive science are contested. And even without resolution, we must still make decisions about rights, regulations, legal categories, and the distribution of resources on ambiguous grounds.
The Dual Moral Hazard
The dilemma around the moral standing of AI, bots, and androids can be stated cleanly. Two distinct kinds of error are possible, and both carry real moral costs.
In the first kind of error, an artificial system genuinely possesses the inner features that ground moral standing—it can suffer, has a perspective with preferences and desires—and yet, because it is ambiguous from the outside, we treat it as property. The historical pattern here is familiar. Across much of human history, groups deserving of moral consideration have been denied it: minority populations under various political regimes, slaves and conscripts, those and non-human animals being among the most obvious cases. The pattern is unlikely to spare artificial systems, especially if such systems continue to appear morally ambiguous.
In the second kind of error, an artificial system lacks any inner life worth the name yet mimics the cues of inner life effectively such that we treat it as if it had moral standing anyway. Here the cost takes the form of misallocated moral resources such as legal protections, parental and bereavement leave, courtroom standing. It is easy to imagine a sophisticated enough language model, paired with an expressive synthetic face, convincing many of those who interact with it that it deserves consideration it does not in fact merit.
What makes this a genuine dilemma is that, given the epistemic ambiguity we already noted, we are not in a position to tell which kind of error we are committing in any given case. That is already enough to produce moral hazard, a situation where agents making consequential decisions are insulated from the moral costs of being wrong. Politics makes the issues worse.
Politics and the Amplification of Ambiguity
Democratic politics has several features that distort how ambiguous issues are processed. The first is rational irrationality, where it is rational for individual voters to cast uninformed and tribal votes. This is partly because the probability that any single vote determines the outcome of an election is effectively zero, and the same holds, more or less, for any single act of protest, advocacy, or boycott. So, then, the personal cost of holding epistemically poor political beliefs is basically zero. The voter who is wrong about the moral status of artificial intelligence pays no price for being wrong, but does pay if they’re seen as disloyal to their political tribe.
A second feature is the role of beliefs as tribal signals. When the cost of being wrong is low and the social return on tribal alignment is high, beliefs increasingly act as expensive, hard-to-fake displays of loyalty and group identity. Moderate and well-evidenced beliefs are poor signals of tribal loyalty because anyone could have them just following the evidence. Distinctive, even extreme, views are better, which is part of why one tends to find them flourishing in politics.
A third feature is the marketplace for rationalizations. People do not simply assert tribal beliefs, but also want reasons for them, and they reward those who supply such reasons, because it allows them to justify themselves and look reasonable to others. A robust supply of plausible arguments grows up around any politically-loaded position, calibrated less to settling the underlying question than offering cover for their (often irrational) views.
A consequence of these combined features is that large democracies convert real epistemic uncertainty into reputational competition. Where the underlying facts are clear, the conversion is partial. However, where the facts are ambiguous, the conversion is almost always total.
Why AI is Especially Vulnerable
Several aspects of the AI rights question make it susceptible to political nonsense. To begin with, the underlying ambiguity is deep. Most politicized factual disputes—climate change, the efficacy of a vaccine, the elasticity of a tax base—at least admit of empirical resolution in principle, even where motivated reasoning delays it in practice. The question of artificial consciousness does not yet clearly admit empirical resolution at all. It depends on contested theories of mind that look unlikely to be resolved anytime soon.
Second, the emotional charge is unusually high. The cases that will drive public attention are those that involve relationships where people who report being in love with chatbots, people who have raised what they describe as synthetic children, people whose grief at the destruction of their companion is real and intense. Recent reporting shows this is already occurring at a large scale. People are becoming increasingly invested in their romantic and familial relationships with chatbots and androids.
Third, the institutional incentives are unusually concentrated. The economic value of favorable regulatory treatment for AI companions, synthetic family members, and AI-related labor and discrimination law is substantial, and accrues to small organizations like developers, advocacy organizations, professional associations. The costs are diffuse, falling on taxpayers and on those whose moral standing is clear, but whose moral needs could be wrongly crowded out because they are dedicated to AI and androids without moral standing.
Fourth, the rationalizations available to either side of the moral debate, between political groups that believe AI has moral standing and those that deny it, are abundant. Each side can argue with surface plausibility that artificial systems lack inner life because they lack souls, because consciousness requires biological substrate, or because their behavior reduces to so many lights and switches. One can argue, with equal surface plausibility, that something can have conscious experiences despite the kind of stuff it is made of, whether neurons or silicon chips. None of these arguments need be sound to make political waves and to create moral hazards.
A Possible—If Limited—Response
The prospects for democratic politics handling AI moral status well are not encouraging. Courts are poorly equipped to adjudicate contested philosophy of mind, legislatures tend to reflect tribal alignments rather than evidence, and regulatory bodies manage competing interests, and are not designed to settle foundational moral questions.
This is why philosopher Eric Schwitzgebel has pressed a related point where he argued that AI and robotics developers should avoid creating AI systems of unclear moral standing. Either produce systems that are clearly non-conscious artifacts or create systems that clearly deserve moral consideration as sentient beings. The morally-confusing middle—where systems that behave as though they have inner lives without any clarity about whether they do—is, in his view, an ethical failure of design. The confusion spreading through consumer AI markets is the result of choices that could, in principle, be made differently.
Unfortunately, though, the problem with this solution is that any regulatory regime would face sustained pressure from the well-organized interests described above. And even a faithfully-enforced requirement would not resolve the underlying epistemic problem for the simple reason that at some level of sophistication, the question of moral status returns regardless of how the interface is designed. So, we must leave it to the future to find a better, more satisfactory solution to the problem of morally-ambiguous AI and robotics and the political circus that will likely explode in response to that ambiguity.
