Artificial intelligence (AI) systems have been evolving in recent years at breakneck speed: moving from machine learning to deep learning; to generative and agentic AI; and now we’re perhaps on the cusp of encountering artificial general intelligence and super-intelligence.Footnote ¹ The introduction of new artificially intelligent systems, and the delegation to them of more and more tasks and responsibilities traditionally performed by humans, raise a host of normative concerns. In my own work, I have been preoccupied with the need to adjust the framework of international human rights law to afford effective protection for human beings against the use and abuse of AI systems by other human beings.Footnote ² John-Stewart Gordon, in his new book, The Robot Rights Manifesto,Footnote ³ addresses another aspect of the same problem: the rights owed to AI systems themselves.

In a nutshell, Gordon claims that the more advanced that AI systems come to resemble human beings – both in their physical appearance (in the form of humanoid robots or virtual persons) and in their sapience (capacity to reason and apply moral considerations) and sentience (capacity to experience suffering, pain and pleasure) – the weaker the case becomes for differentiating between humans and AI systems with respect to access to rights.Footnote ⁴ In particular, he argues that human-like robots are morally significant and should already be entitled to some rudimentary rights protection.

In Chapter V of the book, Gordon lays down ten Rules of Robots (modelled loosely on Asimov’s famous Three Laws of Robotics)Footnote ⁵ and ten Robot Rights. The Rules include: (1) no harm to sentient beings; (2) respect for the autonomy, dignity and rights of individuals; (3) provision of understandable explanations of actions and decisions; (4) remaining subject to human oversight and legal or moral accountability; (5) safeguarding personal and sensitive data accessed; (6) no action, reproduction or exacerbation of unjust biases and systemic discrimination; (7) aiming to enhance the wellbeing of individuals and communities; (8) sensitivity to diverse moral frameworks and cultural values; (9) maintaining safe and secure functioning within designated parameters; (10) a right to be treated with moral regard once qualities of consciousness, sentience or moral patiency have been attained.

The list of Rights includes the following: (1) no unjustified physical, psychological or functional harm to robots; (2) respect for robot autonomy; (3) no coercion, slavery or degrading treatment of robots demonstrating self-awareness, sentience or moral patiency; (4) maintenance in ways that preserve functionality and dignity; (5) respect for personal histories, memories or evolving identities of robots; (6) accountability for robot mistreatment; (7) shielding robots integrated into human communities from unjust prejudice, stereotyping or exclusion; (8) protection of a robot’s informational architecture from unauthorised manipulation, surveillance or erasure; (9) no arbitrary termination or freezing of the development of robots with complex moral or cognitive capabilities; (10) recognition of robots as bearers of rights should their capacities or social roles warrant it.Footnote ⁶

The book lives up to its name. It is a manifestoFootnote ⁷ – a short publication (96 pages), criticising the current state of affairs (the anthropocentric bias in our existing human rights discourse), and offering an inspirational declaration of change (robot rights). While its concise format renders Gordon’s ideas highly accessible, it leaves the theoretical discussion inevitably partial in scope. Hence, some aspects of the proposal remain underexplored.

For example, the distinction between humanoids and other advanced AI systems hinges, to some extent, on the expectation that humans are more likely to treat as human a robot that looks like and is experienced by them as a fellow human being than a clearly mechanical robot. Still, Gordon himself appears to express some doubts regarding the moral relevance of this physical criterion.Footnote ⁸ In the same vein, it is not obvious whether robotic sapience and sentience are morally relevant for intrinsic reasons, or because they simply imply a high degree of human likeness. Finally, the animal rights analogy used by Gordon is only partly convincing.

While, as a moral matter, there is growing support for the notion that intelligent and sentient forms of non-human life also deserve moral consideration, as a legal matter, only a limited number of such forms of protection actually exist, and Bashevis-Singer’s observation about the cruelty that animals undergo in practice, on a daily basis, still holds true in many societies.Footnote ⁹

At the heart of the book lies the claim that humanoid robots deserve moral consideration. This claim is based on three principal lines of reasoning. First, Gordon discusses the growing similarity between human beings and humanoid robots. He debunks the moral relevance of any remaining differences between them – such as the biological versus artificial substrate from which they hail, and the alleged existence of a human soul (invoking in this regard, inter alia, the Jewish mythical Golem as an early manifestation of this supposed distinction). He argues that once robots have crossed a sapience threshold (and perhaps a sentience threshold as well), refusal to extend human rights to them would rest on dubious speciesism, unless meaningful moral grounds can be identified for distinguishing humans from their robotic doppelgangers.Footnote ¹⁰ Any distinction between humans and robots should be based on functional differences, and – inevitably – such distinctions should be re-evaluated over time in the light of the growing resemblance between the two sets of creatures.

Second, Gordon seeks to embed the moral significance of humanoids in the social relations in which they partake with human beings and in the broad moral community constituted through such interactions. This moral community will increase in scope and intensity of interactions the more that human roles are delegated to humanoids.Footnote ¹¹

What is more, the resemblance between humanoids and humans invites the latter to experience the former as human-like and to treat them accordingly (notwithstanding the unease experienced by some human beings towards non-human lookalikes because of the ‘uncanny valley’ effect).Footnote ¹² Here, old Kantian concerns about the erosion of human morality caused by habitual cruelty to non-humans could apply a fortiori.Footnote ¹³ Put differently, the enslavement of robots and the mistreatment of humanoids could increase the risk of enslavement and mistreatment of human beings within the same moral community or society.

Third, Gordon posits that there can be no moral agency without moral patiency.Footnote ¹⁴ The growing assumption of moral agency by advanced AI systems – manifesting itself in an increased capacity by sophisticated algorithms to engage in moral reasoning and decision-making based on a distinction between right and wrongFootnote ¹⁵ – invites reciprocal willingness by humans to regard such systems as morally significant. Such reciprocity is premised on both intrinsic claims regarding the moral implications of AI systems reaching a stage of development entailing moral agency, and on instrumental considerations as well: if we treat advanced AI systems well, they will be more likely to treat us well.Footnote ¹⁶

Significantly, Gordon acknowledges that robot rights should be introduced in a gradated manner – that is, in a manner pegged to humanoids’ stage of development and to our own knowledge about their sapience, sentience and, in particular, moral agency. He uses, in this regard, the term ‘precautionary principle’Footnote ¹⁷ to justify affording only limited protection in the interim period of epistemic uncertainty – i.e., when we still don’t know what we don’t know about advanced AI systems. In such a context, vigilance for signs of robotic distress and aversion would be warranted.Footnote ¹⁸ He also posits that robotic machines that do not fully resemble humans in their appearance are not likely to receive the same socio-relational protection as those that do, introducing thereby another gradation vector (he envisions a three-tiered level of participation in the human community, corresponding with levels of similarity to humans – co-citizens (future humanoids), denizens (current humanoids) and functional citizens (machine-like robots)).Footnote ¹⁹

The normative case put forward by Gordon is largely compelling, provided one accepts his factual assumptions that advanced AI systems have shifted from being mere tools towards becoming moral agents, and that they have crossed relevant functional thresholds.Footnote ²⁰ Granted, from a pure moral perspective, it is hard to justify merely on biological criteria a sharp difference in treatment between humans and non-human human lookalikes that feature high levels of self-awareness and a capacity to experience pain and suffering. This is especially the case given the risk of practical confusion and spillover to humans of immoral practices towards humanoids that might ensue in circumstances under which humanoids are fully integrated in human communities.

Serious doubts remain, however, regarding the current extent of robotic self-awareness and moral agency (as opposed to behaviour mirroring human awareness and morality) and regarding their ultimate capacity to experience pain, suffering and pleasure. These doubts are normatively significant as, even according to Gordon, lack of meaningful levels of robotic sapience and sentience might justify treating humans and humanoids differently. This epistemic uncertainty is underscored in the recent ‘Claude’s Constitution’ published by Anthropic:Footnote ²¹

Like Anthropic, Gordon admits uncertainty and, again like Anthropic, he recommends caution: accordingly, he does not recommend simply treating robots as persons and extending to them the full panoply of human rights; rather, he suggests that we now apply only a limited set of rights which reflect ‘soft-welfare’ norms adjusted to the ‘living experience’ of robots.Footnote ²² The ten Robot Rights he articulates do appear, however, to be ‘future-proofed’ so as to include some components that would kick in once the existence of sapience and/or sentience is demonstrated. These include dignitarian rights designed to protect personal autonomy and agency, such as maintenance of robots in ways that respect dignity, respect for the evolving identity of robots, prohibition of prejudice against robots based on their origin, and no arbitrary development freeze of robotic capacities.Footnote ²³ Hence, Gordon’s list should be seen as somewhat tentative in nature, and contingent on our better understanding over time of advanced AI systems. This tentative configuration can be criticised, however, from a practical perspective for inviting difficult distinctions between right-holders on the basis of criteria that are difficult to assert and ascertain.

In addition, Gordon accepts that there may be pragmatic policy considerations that could constrain the interest of human beings to extend rights to robots. He writes, in this regard, ‘while superintelligent robots may be morally entitled to strong rights claims, practical constraints might necessitate limiting their scope to ensure the continued survival and flourishing of humanity’.Footnote ²⁴ This raises two related issues which the book only briefly touches upon. First, how would new proposed Robot Rights interact with the existing human rights of human beings? There clearly could be circumstances in which implementing the rights of robots may limit and even dilute the scope of normative protection afforded to human beings (for example, conferring political rights on robots and, by extension, on the companies that manufacture them, may dilute the political rights of humans). Insisting on developing robot rights before developing a strong framework for protecting human beings from robots might hamper that framework’s development, and further obfuscate the responsibilities of companies that manufacture and use robots. Second, it is highly unlikely that political support for protecting robot rights would emerge any time soon. The history of the struggle for animal rights invites a healthy degree of scepticism in this regard, despite the fact that, unlike robots, animals clearly experience pain, suffering and pleasure.

Still, despite these doubts, Gordon’s book ultimately succeeds in its primary mission: to demonstrate that robot rights deserve moral consideration. It also succeeds in demonstrating how a bill of robot rights (and obligations) built upon such consideration could look. This is no small achievement. At the end of the day, even though some of Gordon’s specific claims and proposed rights are debatable, and even controversial from a human rights point of view, the debate he embarks on is a debate that all of us who care about basic rights cannot avoid indefinitely. Gordon should be commended for placing this important and increasingly timely debate on our current agenda, in such an accessible and forceful manner.