Can a machine own anything?

Every few years the idea of legal personhood for autonomous systems comes back around, usually with an analogy to companies attached: we already grant personhood to a fiction, so why not to a machine? The analogy is doing a lot of work it cannot support, and the reasons it fails are more useful than the proposals it is used to justify.

What the company analogy misses

A company is not an autonomous legal person. It is a structure through which humans act, and every capacity it has traces back to people: directors who decide, shareholders who own, officers who sign, and a registry that records which humans are answerable for it. Take the people out and a company cannot do anything at all — not sign, not decide, not be sanctioned in any way that changes behaviour.

Personhood for a company is a convenience for organising human responsibility. It is not an example of responsibility existing without humans, which is what the analogy is usually invoked to establish.

The four capacities, and where each one breaks

Owning. Property law needs somebody who can transfer, defend and be deprived of a thing. A system can hold a private key; it cannot hold title, and a court can reassign what it controls without the system having any standing to object.

Contracting. Agreement requires capacity to agree. Automated contract formation works today precisely because the agreement is attributed to the human or company operating the system, not to the system itself.

Being liable. Liability that cannot be enforced is not liability. Turning a machine off is not a sanction; it is asset disposal, and it falls on the owner. Compensation has to come from somebody with assets, and that is a person or a company.

Standing. Being able to sue and be sued in your own name. Nothing in EU or Polish law provides this for a system, and the idea of an “electronic personality”, once floated in the European Parliament, met substantial expert objection and did not become law.

What is actually developing

Not personhood — obligations that attach to whoever deploys. EU rules on artificial intelligence spread duties between those who build systems and those who put them to use, broadly in line with the risk involved, and more autonomy tends to bring closer scrutiny rather than diffused responsibility. The direction of travel is that more capable systems come with more documentation, more traceability and more identifiable humans attached, not fewer.

That is the opposite of the personhood story, and on reflection it is the more sensible outcome. Granting a system personhood would mainly create a way for the humans who profit from it to stop being answerable for it.

What this means for the programme

It sets a ceiling, and the ceiling is not moving soon. An autonomous operation runs inside a human legal entity, with a named owner, and any economic behaviour that would require capacity to contract stays in a sandbox with no real counterparties.

What remains genuinely open — and worth research — is narrower and more useful: what must an autonomous operation record, bound and expose in order for the humans responsible for it to actually be able to answer for it? That question matters whether or not personhood ever arrives, and unlike personhood, it can be worked on now.