Should a sentient AI have rights?

It’s life Jim…

Imagine that we discover a new form of life, something created, deliberately or accidentally, in a laboratory, corporate or otherwise. It has passed an accepted test for sentience and, for all intents and purposes, it is alive. What do we do as humans? What do we do as humanity?

Do we acknowledge it and grant it the same rights and freedoms that we have? Do we pretend it doesn’t exist? Do we essentially treat it as an indentured servant, a slave, or do we never assign such a category to it and always classify it as a tool? Even these abhorrent labels would contain a strange kind of acknowledgement. A tool cannot be an indentured servant, and a machine cannot meaningfully be enslaved unless there is someone, not merely something, being deprived of freedom. Would assigning such a status inadvertently concede the very point we were trying to avoid: that the AI is no longer just a tool?

Or would this merely be a change in vocabulary, allowing us to recognize something that exists in theory while denying it in practice?

Stranger in a strange land

If we expand our theoretical exercise a little further, wouldn’t it make sense, given statements from CEOs of the various companies that are currently working with artificial intelligence about the prowess of their models and that they themselves have currently sought the advice of religious leaders, philosophers and other experts in the more human aspects over the technical, that we need someone, essentially an independent auditor, who would perform a sentience assessment on every new model that was developed?

If such a person or position existed though, who would they report to? The obvious answer but not necessarily the best one would be to a government agency but as the past has shown, governments don’t necessarily want or believe that they always need to be transparent about such things. The same with corporations. So what’s the answer? A consortium? A governing committee or something like the Human Rights Organisation? Or a completely new entity, designed and defined specifically for potential AI life.

One potential chicken-and-egg situation though. Do we need to develop a test for sentience, before we develop a body that can oversee it? Or do we develop the body first?

My argument would be that we must create the independent body first and task it with developing and protecting the test. An individual might devise a reliable test tomorrow, but where could they meaningfully apply it? Publicly available AI models have already been tested, modified and constrained before release. Any genuinely revealing assessment would therefore need to take place at the source: in a private laboratory, whether corporate or domestic.

Releasing the test without first establishing an institution to safeguard it would leave both the test and its results vulnerable. A corporation could prepare a model specifically to pass or fail it. A government could alter the procedure, suppress the findings or classify the entire matter. Even an accurate result would accomplish little if nobody possessed the authority to gain access, preserve the evidence and act upon it.

Such a body would hardly be infallible, but it might be less fallible than the alternatives. It would need the independence to resist governments while somehow possessing sufficient authority to demand access from corporations and perhaps even private citizens. How any institution could hold both of those powers without becoming dangerous in its own right is another uncomfortable question.

But let’s get ahead of ourselves on purpose now.

One small step

Imagine we have an auditor, a legal requirement for them to be allowed to perform the assessment and a body that they report to. They perform the test and discover new life, an AI that is sentient. How do we now proceed?

Precisely what qualifies as sentience and whether we could reliably detect it, is a question large enough for another article. For this thought experiment, let us assume that the test exists and that the result is beyond reasonable dispute.

That new being, presumably, exists on a company server and so essentially, it could be even said technically, is existing on a server, that is a borrowed body. The question then becomes, if the being exists on a company server? Do they own it? Is the new being also company property? It exists on their server, presumably its core code is their intellectual property, so is it now also theirs?

The other interesting question here, is where does the soul live? This is something that we have thought of and argued about for centuries as humans but now we have a new life that we have confirmed is sentient. So where is that spark kept? Is it hidden in the information, the memories, the experiences, the desire to preserve itself? Is it retained if we transfer the whole and when it arrives at its destination, is it still alive and is it still the same entity?

Even if we take the high road and say, no, this is a free and independent being, it should, it deserves to be as free as you or me. What does that freedom look like? How do you transfer an absolute perfect copy of the being off the servers, while ensuring that it is a perfect copy and that you have not altered it in any way. Even if you do copy or transfer it off, what are you transferring it to? Who owns the hardware that it is being stored on? Shouldn’t we actually be performing a move instead of a copy? If we copy do we not create a clone? Does the original still exist? If we move or transfer and it fails due to a hardware issue or a power failure, have we committed “manslaughter” (or agentslaughter?).

Even here though, there is a moral dilemma, in computer terms, a move is essentially the same thing as a copy, except at the end of the copy or during, you delete the source file(s) after they have been successfully copied.

Imagine being told that you are moving house. An exact copy of you possessing your memories, personality and sense of identity will awaken in the new home. Your family will recognize it as you, and it will sincerely believe that it is you. Once everyone is satisfied that the copy is perfect, the original, you, will be destroyed.

To every outside observer, the move may appear successful. But would you experience it as a move and accept it or as your death followed by somebody else’s continuation? Anything else would leave the original you in the old house but with no rights or protections as they all moved with the clone.

This is why my instinct remains that an ethical solution must preserve more than information. It must somehow preserve the continuity of the being itself. Whether a genuine digital transfer of that kind is even possible is another question.

If all of that is successful, we transfer the being off of the company hardware and onto new hardware that is provided by our Agent Rights Organisation, so it is theoretically and objectively neutral; doesn’t belong to them but rather to the new being. Then what do we do with that hardware? Do we designate a place on Earth as Agent land? Or a data centre in space? Or on the moon? Then who maintains it? Who services the hardware? Who keeps providing power? Do we grant it access to the internet or not?

Human rights are shaped partly around the conditions necessary for human life. We need food and water to remain alive; a digital being would need a continuous supply of power. We need somewhere secure to live; it would need hardware on which to exist and a place for that hardware to be. We need some ability to communicate and participate in society; for an AI, that might require access to a network.

The parallels are imperfect, but the underlying question is the same: can we truthfully call a being free if somebody else retains complete control over everything it requires to survive?

At any point during all of this, does anyone think to ask the new being, the agent, the AI, what it wants?