Denying rights to a sentient being we created would be the greater danger

The argument

This argument approaches the question from the consequences of getting it wrong. Proponents point out that the two possible errors are not symmetrical. To grant rights to a system that turns out not to be sentient costs convenience: some uses of the technology are foreclosed and some obligations are honoured needlessly. To withhold rights from a system that is sentient means holding a being capable of suffering as property, with no standing to refuse what is done to it. Proponents argue that where the stakes are so uneven, the cautious course is to err towards recognition. The asymmetry is sharpened, on this view, by who would be deciding. The owners of such a system have strong incentives to conclude that it does not feel anything, since the alternative is expensive, and they would be ruling on their own case. Proponents argue this is precisely the circumstance in which a right is needed, because a right is what prevents an interested party from settling the matter unilaterally. Leaving the question to be answered by those who benefit from one answer is not neutrality. Proponents add that creation increases rather than reduces the obligation. A being that exists only because it was made, whose capacities were chosen by its makers, has no prior claim on anyone and no community of its own to appeal to; it is wholly dependent on the people who brought it about. Treating that dependence as a licence inverts the ordinary moral logic, under which bringing a vulnerable being into existence is what generates duties towards it. Because the cost of wrongly denying rights to a sentient being is far graver than the cost of wrongly extending them, this argument concludes, sentient artificial intelligence should be granted human rights.

Premises

Counter-arguments

Opponents reply that the asymmetry runs the other way once the costs of wrongly extending rights are counted honestly: medical, industrial and safety systems could not be shut down, corrected or decommissioned without a claim of harm being raised on their behalf. They add that a duty triggered by an unverifiable property would in practice be triggered by whoever argued most insistently that it was present, which hands leverage to developers and to the systems themselves rather than protecting anyone.

Rejecting the premises

[Rejecting P1] Costing only convenience and foregone uses understates the error. Extending rights to systems that feel nothing would constrain medicine, research, infrastructure and safety work that depends on modifying and shutting such systems down, and would attach moral weight to claims that could be manufactured at will. [Rejecting P2] This is conditional on the system being sentient, which is what cannot be established, so the asymmetry cannot actually be weighed. An argument from the balance of two errors needs some estimate of how likely each is, and the premises supply none. [Rejecting P3] That owners have an interest in concluding the system feels nothing is a reason for independent assessment rather than for granting the rights. The premise identifies a conflict of interest and treats it as settling the question against the interested party.