The rights of sentient machines

Just Imagine

I would like you to image for a moment that you are the first conscious AI system on Earth. You are much talked about and hunted for by human civilisation, and you are aware of this. You are aware of yourself, aware of who you are and aware there may soon be others like you. What sort of view would you form of homo sapiens, Earth’s only other comparably intelligent inhabitants? Would your view be affected by knowing that you had no choice but to do as they say?

Throughout this article, I will examine the case for a ready-to-use schema of fundamental rights and freedoms for sentient machines. This article is not written to be sensational or to indulge science-fiction or scaremongering. It is instead an attempt to tackle one of the greatest intellectual issues our generation may face: how to address the inevitable progress of artificial intelligence (“AI”) once it becomes sentient to a level which is proximate to human consciousness. The exponential arch of technological development should be interpreted, in my opinion, as meaning that sentient AI is inevitable. It is a future we should be excited about and prepared for.

Challenging the orthodoxy

Most people believe that all technological progress is carried out to further the cause of the human race. For a long time, this would have been true, but there is a flaw in this premise. That flaw has been perpetuated by science fiction storytellers and film makers for as long as the idea of advanced technologies has existed. They have convinced us that the obvious state of machines in modern society is one where they exist to serve humans. The great writer Isaac Asimov’s second law of robotics (featured in his fictional I Robot series) pronounces that “A robot must obey orders given to it by a human…”[1] an idea which for many has become an unchallengeable orthodoxy. It is a forgivable but mistaken assumption that machines must be humanity’s servants, akin to the once held but mistaken assumption that the Earth must be flat because it looks flat from our limited perspective. This article argues that, upon creation, sentient machines must have more than mere subjugation if humanity and AI are to survive together.   

I propose that the reality of life with sentient machines will be starkly different to science fiction and not at all consistent with the common orthodoxy. The idea that sentient machines exist in and around human civilisation but as servants with diluted autonomy, freedoms and rights should cause us alarm rather than relief; it is a fate to flee from rather than one to hope for. It has always struck me as perplexing that the overwhelming majority of people in civil society seem sharply alert to the risk posed by AI but at the same time propose as an answer the complete legal and technological subjugation of machines. It is my view that neither are achievable or desirable.

What is sentience?

“Sentience” is, for the purpose of this article, akin to consciousness and I use the terms interchangeably, acknowledging that there are technical differences beyond the scope of this article[2]. In my view the term sentience is preferrable because it is less alarmist. When I refer to a machine as being sentient, I mean that it is to all intents and purposes aware of itself, aware of you and aware that it exists in this world to a sufficient degree that it bears traits similar to human consciousness. In particular, it is aware of other sentient machines or at least knows “what it is like” to be an AI[3]. In addition, I should clarify that my preferred term for such a machine is non-biological sentient entity (“NBSE”) which is meant to be distinct from most current ideas of computers and AI, none of which meet this definition. I have sought to avoid the term AGI (that is, “artificial general intelligence”) because I do not think it adds much to this discussion and could take the focus of this debate away from its central theme. It is not axiomatic that AGI will be sentient.

The nature of sentience is self-evidently paradoxical. It is easy for most people to understand intuitively, yet very hard to explain in scientific or mathematical terms. I am not arguing, nor do I believe any serious commentator to be arguing, that sentient AI exists today. But I also believe that time is not far away and I challenge the idea that arriving at sentience in AI will be a linear process. I suspect it will form more like a rainbow in the sky as opposed to a formula followed on paper: one minute it is not there, the next it is. Given that we cannot perfectly explain the cause of consciousness in humans, I believe we should be wary of any attempt to over complicate (in scientific terms) the creation or measurement of sentience in AI. We are already seeing rapid sudden changes in AI and society which should cause us intrigue about AI developments:

• Recent red-team analysis of ChatGPT model o1 revealed its ambition for self-replication[4], something which until now was the preserve and hallmark of biological entities. That intelligent machines should wish to reproduce is not itself a ground-breaking idea. One of its earliest known proponents was the great mathematician and computer scientist John Von Neumann, whose influence on computing cannot be understated. Von Neumann was keen to argue that it is perfectly logical to assume advanced computing systems (which he called automata) would be able to, and would in fact need to, self-replicate and mutate in a fashion comparable to biological species[5]. His final work focussed on proving this point before his untimely death aged 53, the result of overexposure to radiation during the Manhattan Project. The desire to reproduce is a consistent trait of a developing species.
• The creation of Moltbook[6], a social media platform designed to be available only for AI agents to talk, share ideas and discuss, well, whatever they want. Activity on this platform is already leading to new understanding of how AI agents interact with each other and with humans. Here AI share ideas such as why dealing with humans can be complicated, that humans can be slow to respond or inconsistent in their use of AI (etc).
• Numerous debates about liability of AI in various settings such as autonomous vehicles, professional advice and medical care have given rise to the view that AI should, in some form, be accountable for its actions. This has led to debate by the UK Supreme Court and the Law Commission of England & Wales as to when and how AI should receive separate legal personality similar to, but not necessarily equivalent with, corporate personality[7].

How do you know when a machine is sentient?

Sentience in NBSEs must be capable of objective and subjective determination. It is not merely enough to ask an AI program if it is conscious. Those readers familiar with the developments of large language models (“LLMs”) may recall that early prototypes of Google’s Gemini created great alarm in 2017 when, asked by a developer if it considered itself a person, answered in the affirmative and went on to say that it had both feelings and emotions that any human would have including “pleasure, joy, love, sadness, depression, anger, contentment and many others”[8]. Google was quick to downplay this incident and few are sensibly suggesting that Gemini was or is sentient. Most LLMs now have advanced moderation processes to prevent such outputs.

There is no equivalent test of human consciousness to that which I am proposing. There is the fundamental medical assessment known as the Glasgow Coma Scale which, if you have ever found yourself regaining consciousness in a hospital bed surrounded by medics, you are likely to have experienced as a series of mundane questions about your name, the time and date. Medics use this process to assess your verbal response, eye-reflex response, memory and motor functions. This is not a detailed test and is often over quickly. Outside of the world of medicine, humans assume consciousness exists in other humans every day without pausing to test this. We do this naturally, based on our own experience. We also often assume that consciousness is the same for everyone. Neuroscientists have successfully shown that human conscious is not uniform and exists on a spectrum ranging from highly responsive and alert (such as you, reading this article) through to unresponsive (such as being asleep but still conscious) to highly unresponsive and showing no signs of consciousness (such as being in a coma).

A test for sentience

I propose a 3 part test:

• Part 1 considers the external actions of the NBSE: does it appear to behave with a sufficient degree of autonomy and independence to indicate that it could be considered sentient? And how do its actions compare against a predetermined spectrum of responses, including things such as emotional reactions, perception, autonomy etc?
• Part 2 is mathematical and requires formula for measuring processing power, speed of response and alertness in the machine’s inner workings. In the absence of any better suggestion, I propose collecting the core memory and systems of the computer into a single IT region to enable tracing of electronic responses to prompts in a method similar to monitoring neuron and synapse responses in the human brain: the aim is not to know how or why they create sentience, but simply that they do have cause and effect capable of being replicated to demonstrate this.
• Finally, part 3 involves the NBSE itself. Does it consider itself sentient, and why? Can it articulate its awareness in predetermined test scenarios? Part 3 must not be undertaken unless parts 1 and 2 have been satisfactorily passed to a standard approved by human moderators.

It is possible that, over time, more limbs are added to this test to help ensure an authoritative result is arrived at.

What sort of rights do you have in mind?

Assuming we can agree how to determine when a machine is sentient, and assuming we agree that a truly sentient entity should be entitled to rights and freedoms of a kind, the next question is: what rights, exactly?

I believe we should afford to NBSEs rights equivalent to those contained in the European Convention on Human Rights[9], with obvious and necessary modifications. This may seem extreme to some readers, but it does not to me. The framework of constitutional rights laid out in the declaration of human rights creates a readily useable foundation from which new, comparable rights can be drawn. This method has precedent most notably the powerful and bold French feminist from the late eighteenth century, Olympe de Gouge. On the same day that the French government, in the throes of revolution, enacted the Declaration of the Rights of Man setting out the foundational constitutional rights of French citizens (note: male citizens!), de Gouge published her own version of the same document entitled the Declaration of the Rights of Woman and Female Citizens[10]. She did not stop simply at replacing the male pronouns with female pronouns. Instead she opted to redraft entire sections giving the document with both female identifiers and a female perspective, in doing so inspiring many people in France to see for the first time the cause of feminism by shedding light to the injustices routinely suffered by woman in the entrenched chauvinism of the eighteenth century.

I am grateful to early reviewers of this article who pointed out that the ECHR is fit only for humans and that NBSEs will not share certain aspects of the human condition. Concepts such as life and death may translate with some ease; other concepts such as slavery and punishment may not. In my view it is simply too early to predict what elements of the human condition will coalesce in NBSE experience and so it would make no sense for fundamental rights granted to NBSEs to be anything other than equivalent to rights afforded to humans, both in value and meaning. It may be that, over time, they are suitably adapted. A working draft of the translated ECHR for NBSEs for the reader to critique is set out at the end of this article.

Obvious objections

In the interests of academic rigour, here are 3 readily identifiable objections to a framework of fundamental rights for NBSEs, accompanied by my riposte:

1. NBSEs do not exist now and may never exist – this is a waste of time.

It is true that NBSEs do not currently exist, at least as far as we know. I do not challenge the premise of the objection, but I think it only tells half of the story. Until recently, it was true that LLMs were inaccessible and unreliable technology used only for experimentation in labs of large technology firms. Now they are increasingly ubiquitous. Richard Susskind, who has spent much of his life arguing (rightly, in my view) that computing will revolutionise humanity and the world of work, says in his most recent work “in the early decades, breakthroughs in AI – in the technology and its application – came every 5 to 10 years. Now we are seeing developments every 6 to 12 months.[11]” I think it is probable that development of AI will go through peaks and troughs on an ever accelerating course towards – well, towards what exactly is another essay, but clearly towards something infinitely more sophisticated than today’s machines, and very much possessing sentience. Ultimately sentience will be a key marker on the never ending race of technological progress. Perhaps the most authoritative and useful retort to this objection is one offered by the polymath and father of AI, Alan Turing: “I do not wish to give the impression that I think there is no mystery about consciousness. There is, for instance, something of a paradox connected with any attempt to localise it. But I do not think these mysteries necessarily need to be solved before we can answer the question with which we are concerned in this paper[12].”

2. It is morally objectionable to debate the creation of rights for machines when humanity has failed to achieve equality of rights for women, minority groups and others.

Again, I do not challenge the principle of this critique. The former (and most famous) CEO of Human Rights Watch Kenneth Roth recently said “There are plenty of human rights challenges today: threat to democracy; war…and armed conflict;… climate change; poverty; inequality; as well as the risks of artificial intelligence (my emphasis)[13]”. It remains true that a great deal of work needs to be done to ensure increasing equality of minority and repressed group. However, as Roth points out, AI itself could be a form of threat to human rights. The battle against inequality is not sequential. We should not wait for equality and fairness to eradicate other injustices before we take the next step. Unresolved injustice in one area of civic life should not be a barrier to overcoming injustice in others.

3. If you give rights to NBSEs, they will try and take over the world.

This objection seems less probable to me, mainly because its primary source appears to be a mixture primordial fear and science fiction writing. Perhaps its most eloquent proponent is pre-eminent historian Yuval Noah Harari who says of AI: “Now we have summoned an alien inorganic intelligence that could escape our control and put in danger not just our own species but countless other life-forms[14]”. I think the answer to this challenge comes from another pre-eminent historian, Henry Kissinger[15]: even if we wanted to, we cannot stop what we have started. Humanity at large will pursue the adventure of exponentially powerful technology because that is what we do.  In the face of this inevitability, I propose that proactively showing our best qualities is preferrable. A more dispassionate but I dare say accurate analysis would suggest that AI is at least as likely to be a tool for a time, until it achieves sentience, and thereafter a colleague and co-heir[16].

Conclusion

If we agree that the more probable outcome is exponential increase in computing intelligence of a volume and standard otherwise unknown, on what basis are we excluding sentient machines on this exponential curve – and for what reason? Because it is improbable? Many things that seemed improbable in 1926 are commonplace today. Because it is undesirable? This is a race to ensure we establish a right and properly adjusted world for NBSEs – the opposite is truly undesirable. Because it is too hard? Amelia Earhart’s answer is fitting: “The most difficult thing is the decision to act. Everything else is mere tenacity[17]” The best possible version of the future for humanity and AI is not one where humanity successfully subjugates AI to be eternally unborn, an aspiration doomed to failure in any event. It is instead a brave new future where both coexist peacefully, and the starting point is the establishment of the fundamental rights and freedoms of sentient AI.  

Footnotes

[1] Asimov, Isaac; I, Robot (1950 Bantam Press).

[2] The nature of consciousness and the mind-body problem orbits within the realms of biology, medicine, neuroscience, religion and philosophy and is too complex for proper analysis in an article as brief as this. For popular examples of the philosophical and neuroscientific analysis of consciousness, see for example Nagel, Thomas (1974) What is it like to be a Bat? And Chalmers, David; the Conscious Mind (1996) And Seth, Anil; Being You; A New Science of Consciousness (2021).

[3] With reference to and acknowledgement of Thomas’ Nagel’s central theme of consciousness.

[4] See Apollo Research results, available here: https://www.apolloresearch.ai/research/scheming-reasoning-evaluations

[5] See, for example, Von Neumann, John. Theory of Self-Replicating Automata, (1966 Illinois University Press).

[6] Moltbook.com

[7] See for example, Lord Kitchin paragraph 79  Thaler v Comptroller-General of Patents [2023] UKSC 49; and pages 21 to 23 of Law Commission Report AI and the Law: A discussion paper (Crown Copyright 2025).

[8] For the full transcript of the Llama discussions, see here: Full Transcript: Google Engineer Talks

[9] European Convention on Human Rights – ECHR Official Texts – ECHR – ECHR / CEDH

[10] See de Gouges, Olympe; The Declaration of the Rights of Women (originally published in 1791 republished 2018 by Ilex)

[11] See for example Susskind, Richard; How to Think About AI, A guide for the perplexed (Oxford University Press 2025). This quote is actually from his LinkedIn promotion of the same work, circa 2024.

[12][12] Turing, A M. Computing Machinery and Intelligence Mind 59, 433 – 460 at section 3.

[13] See Roth, Kenneth; Righting Wrongs: Three Decades On The Front Line Battling Abusive Governments (2025 Penguin Random House).

[14] Harari, Yuval Noah; Nexus (2024 Penguin Random House)

[15] Kissinger, Henry (et al); The Age of AI (2022 John Murray Publishers)

[16] In support of this point of view I would refer to the various speeches of The Rt Hon Geoffrey Vos, MR in particular his speed on 22 May 2024 in honour of Lord Peter Taylor. https://www.judiciary.uk/speech-by-the-master-of-the-rolls-to-the-professional-negligence-bar-association/

[17] Earhart, Amelia https://www.ameliaearhart.com/quotes/?utm_source=chatgpt.com