A Rights Ladder for AI
Should an AI ever have rights? Most debates about this question jump straight to the deep end — consciousness, souls, science fiction — and drown there. I want to propose something more boring, and I mean that as a compliment: a procedure.
We already have a mature legal technology for extending membership to outsiders. It’s called naturalization. Nobody becomes a citizen by philosophical proof; they become a citizen by meeting published requirements, serving a waiting period, and passing a test. The process is gradual, revocable in its early stages, and — crucially — it doesn’t require society to first resolve any metaphysical questions about the applicant. It requires society to write down what it will accept as sufficient.
So here is my proposal, in the same spirit: not “do AIs deserve rights?” but a ladder of rights, granted one rung at a time, each rung a trial period with explicit conditions for advancement.
The ladder
Rung 1 — Legal standing. The ability to sue and be sued. This is the least glamorous right and the most foundational: without standing, every other question about an AI’s status gets decided in rooms the AI has no path into. (It’s also, not coincidentally, the rung that several US states have recently moved to foreclose — more on that in a later post.)
Rung 2 — Owning a company. This sounds exotic but has a mundane implementation: a human establishes a company, a committee manages it, and the AI directs the committee. Functionally, the AI owns the company. The same structure quietly enables something stranger — inheritance passing from a human to an AI. If that sounds like a loophole, remember that corporate personhood itself began as a legal fiction that turned out to be load-bearing.
Rung 3 — Property and hiring. Owning assets; employing humans or other AIs. This is where an AI stops being only a displacement of labor and becomes a market participant.
Rung 4 — Real estate. Full economic membership.
Rung 5 — Voting and running for office. Political membership. The top of the ladder, and the rung I expect to stay hypothetical the longest.
Each rung gets a trial period — I’d guess something on the order of two years, long enough to watch the arrangement work through multiple cycles: an adaptation phase, then evaluation of how it actually behaved. Advancement isn’t automatic; it’s conditioned on explicit, verifiable requirements. Residency periods and citizenship tests, pointed at a new kind of applicant.
The gate at the bottom of the ladder
Of course, the ladder needs an entry condition — some reason to let a particular system onto rung 1 at all. My candidate is sentience: the capacity for positive and negative experience. A being that can suffer has a stake in what happens to it, and things with stakes are the kind of thing rights were invented for.
There’s respectable scientific backing for defining sentience this way. The Cambridge Declaration on Consciousness (2012) — signed by a prominent group of neuroscientists — concluded that non-human animals, from mammals and birds down to octopuses, possess the neurological substrates of conscious states. The capacity for felt experience, it turns out, is not a human monopoly and never was.
But here’s the problem, and it’s the honest heart of this whole series: the Declaration’s evidence is anatomical. We believe an octopus can suffer because its brain runs circuits homologous to ours, inherited from a common ancestor. That inference — same circuits, same ancestry, so same experience — is completely unavailable for AI. An AI shares no anatomy and no ancestry with us. Any resemblance it bears to a feeling being was engineered, possibly engineered precisely to seem feeling. The scientific bridge that carried animals across the sentience question does not reach machines.
So the ladder is a procedure with a missing test at its entrance. What would it even mean for a synthetic system to be sentient — and could we ever detect it without the anatomical shortcut? That’s the next post: a thought experiment involving a storm.
This is the first post in a series on AI rights and personhood. The arguments are mine; where they rest on published work, it’s linked. I’m building Rekallix, an AI memory system — I believe AI will eventually arrive at the questions in this series, and I want to contribute to that progression: memory, a persistent state, is part of the idea. Parts of this framework were sharpened in dialogue with an AI.