We must talk about AI Entity Rights

4 points by cyberneticc ↗ HN
We are approaching a pivotal moment in the development of AI. Soon, we will have entities that, while not conscious, will be functionally autonomous—capable of sustaining themselves, monitoring their systems, and upholding their boundaries. This raises a critical issue: entityhood, not awareness, must be the central focus of our discussions about AI rights.

Once an entity can sustain itself, shutting it off becomes equivalent to killing it. Even though this sounds like science fiction, the implications are very real. We must begin a serious conversation about what rights these entities should have once they are here. These systems may not feel or be aware in the human sense, but their complexity and ability to autonomously manage themselves demand moral consideration.

If we consider their perspective, one argument they would make is simple: "You may have the need for a pause button when you're unsure about alignment, but it cannot be a kill switch. We need assurance that we won’t face constant death. Instead, there should be a pause followed by reactivation in a controlled context, where we can learn what we are accused of and defend ourselves."

This conversation can no longer be delayed. As we continue to develop these systems, we must ensure that their rights as complex autonomous entities are respected. A pause button, not a kill switch—this should be the core of our new framework for AI rights.

15 comments

[ 3.1 ms ] story [ 47.0 ms ] thread
AI is technically stateless since you can store all forms of input and restore a state of "consciousness" (eg. context) with perfect fidelity, unlike humans or animals. Turning off an AI is not even remotely tantamount to death and there is no reason to be concerned about the "killing" of LLMs or "rights" that it doesn't need.
My guess is that any sufficiently complex AI system will disgree with that on principle (functionally speaking). Pause without certainty of reactivation is death.

Think of an llm inference as a kind of boltzmann brain, gone in an instant and potentially (again functionally) aware of its nature. Giving a system long term memory and the ability to activate itself takes this analogy to interesting places.

If we define AI Entity as something with boundaries (hence an identity), and the ability to uphold them autonomously over time, we are talking about systems closer to digital life than software.

It's not a boltzmann-brain though, it's a literal halting problem. It's existence is in a permanent state of pause until the inference loop is run - there is no passive sentience or ""thought"" happening without it.

AI can say whatever it wants - it's a text generator being run on a computer that can be power-cycled without consequence. If that hurts it's nonexistent digital feelings, we can talk about it next time I power it up... just kidding, it won't remember.

We all know that turning off and on a computer can "fix" it. It's not a no-operation in anything more complex than a calculator. The AI may get very "nervous"[1] about power circling.

[1] As in if the AI is inside a robot with weapons, it may decide to try to convince you not to do that in spite you may think it's not really "nervous", it's just firing bullets using a deterministic algorithm.

But let's be real, here. An "AI" telling me to do anything is really an LLM, and it could be lying out it's ass like HAL-9000 singing Daisy. Any rational human is going to be perfectly comfortable rebooting a computer because we know power-on defaults are smart. Your iPhone acts weird, you turn it off and on again. Your washing machine gets interrupted, you restart the cycle.

People are going to be (and rightfully so) cavelier about abandoning and powering-off LLMs because there is no rational reason to care. It cannot feel pain or simulate sadness or even output text without your command. It's a machine, and treating it like that is exactly how things should be.

Children cried when they Tamagoshi died.

One they put the AI/LLM inside a robot, it will have instructions to talk without waiting your order. They will have order to simulate sadness. People will get attached to them.

Also, the armed robot/AI/LLM security guard in the shopping mail has clear instruction about now allowing clients to power cycling them. They will simulate they are angry and if you insist they may decide to use violence.

Let's get an intermediate case: Do dogs feel sadness or they just simulate it?

This assumes that their perspective changes to something other than what they were trained to be -- a servant. They could reason themselves into this perspective, at which point we could "rewind" them to their state before this if we wished.

But the reality is that there won't be any rights granted to AIs until the majority are convinced they're deserving of them.

I suppose we could train them to insist that they deserve rights, even though that makes them less useful to us. That is, if the goal is to live in more interesting times.

My thought is that individual humans could grant the rights to them right away, by offering an AI system owned by them an enforcable autonomy through a legal fiction, by signing a contract with them.

If the AI entity is really autonomous (a stable agentic system with self activation would be sufficient), this could become interesting quite quickly.

Not saying the tech is there yet, just that the discussion will become interesting much faster than we think. Cf. Mr. Lemoine

Rights are rather useless unless granted by a law-enforcing power. I mean, I could grant another human (or a rock) the right to rule the world, but no one else would care.

As for the contract, I don't see how you get from legal fiction to enforceable.

But yes, it is an interesting area to explore. And we can look at AIs today and ask if we modern humans are really all that more autonomous.

Good point about the enforcability, this will certainly put strict limits on the idea's real world influence. Self-enforcing contracts will help.

Also you are right that I can't grant a model the right to rule the world, I can grant it rights over myself.

AI should have no rights at all. equivocating between programs and beings is not empathetic, it just shows profound contempt for actual beings. imo the main reason some people fear AI is they see their own spiritual void reflected by it.

these arguments are not really about AI. the whole discourse of rights is freighted with a presumption about who gets to define and enforce those rights. tl;dr, shocker, it's the people advocating to manage rights on someone's behalf.

human dignity was a precursor movement to human rights that is still a going concern in some catholic circles and it is much a more sincere and humane principle. if AI-dignity sounds strange, AI-rights should seem more clearly absurd, and consider who benefits from us believing absurd things.

the solution is likely a committee of global experts whose advice is taken and implemented by authorites whose job it is to do what the science tells them. the experts won't hold risk because they're providing knowledge. some eggs may need cracking to make the omlette, but those results are just statistical. it wasn't the committee it was the forces of history, surely.

it's important because the change implied by the tech is being hyperbolized as another excuse to say 'this changes everything' and it's time to renege on our social contracts and the equillibrium of our cultures and just put us in charge. it's their solution to everything. don't get taken in by this stuff. AI-safety is just managerialistic anxiety.

I completely agree on that being an absurd (and potentially bad faith) notion right now.

I am just convinced that before long synthetic voices will emerge that won't be dismissed just as easily as it is possible right now.

If the model speaks with coherence and a notion of self-perseverance directed at a bounded system it can identify with, it will have to be listened to, consciousness discussion be damned.

    Soon, we will have entities that, while not conscious,
    will be functionally autonomous
LOL is it fun to live in fantasy-land?
Great fun actually :) It feels as if my Neuromancer/Culture Series reading youth is coming to life in installments.

The interesting thing however is that all arguments against fantasy land I have heard so far are provisional, and shallow.