The dangers are matters of governance, and governance is something humanity does

The argument

A second argument concedes the risks and disputes what follows from them. Proponents observe that every powerful technology has arrived with a plausible case that it would prove catastrophic, and that the historical response has not been abandonment but the construction of institutions: rules about who may build, standards for what may be sold, liability for harm, inspection, licensing and the slow accumulation of professional norms. Proponents argue that the harms attributed to artificial intelligence are of exactly this governable type. The distinction they draw is between a harm inherent in a technology and a harm arising from how it is deployed. Surveillance, manipulation, concentrated ownership, biased decisions and labour displacement are all consequences of choices about use, and proponents argue that each has an existing regulatory analogue: data protection, advertising rules, competition law, anti discrimination requirements and labour policy. The technology does not determine which of these choices are made, so the case that it is bad for humanity has to assume the choices will go badly everywhere and permanently. Proponents argue that this assumption is unwarranted and self defeating. It is unwarranted because governance of earlier technologies improved, usually after failures that made the need plain. It is self defeating because treating the technology as inherently harmful discourages exactly the public investment, research and regulatory capacity that would make it safe, leaving development to those least interested in restraint. On this account the honest conclusion is that artificial intelligence is good for humanity if it is governed, that governing it is within human competence, and that the work of governing is what the question should be about.

Premises

Counter-arguments

Critics reply that the argument assumes governance arrives in time, and that the characteristic feature of this technology is that it does not. Capability advances in months while rules take years, the systems are built privately and released before anyone outside can assess them, and the firms that hold them also fund the expertise regulators would need. They add that the appeal to history selects the technologies humanity survived and passes over the harms that were governed only after they had become permanent, which is precisely the outcome in question here.

Rejecting the premises

[Rejecting P1] The comparison selects the technologies that were in fact governed successfully and omits those whose harms became permanent before any institution reached them. [Rejecting P2] A regulatory analogue existing is not the same as it being effective, and the pace at which capability advances is a feature of the technology rather than of its deployment. [Rejecting P3] Naming a danger accurately is what has historically produced regulatory capacity, so the premise mistakes warning for discouragement. [Rejecting C] At most the premises show that the technology could be good if governed well, which leaves open whether it will be.