I agree that the risk of an accidental strike was a huge problem with the theory of nuclear deterrence, but the question is: compared to what? In expectation or even in a 1st percentile scenario, was MAD worse than a world where the USSR is a unilateral nuclear power? For that matter, what would it have taken to get a stronger SALT treaty sooner?

I think you need to have people thinking through this stuff at a nuts-and-bolts level if you want to avoid getting dominated by a slightly less nice adversary, and so too with AI. Does a unilateral guarantee not to build autonomous killbots actually make anyone safer if China makes no such promise, or does that perversely put us at more risk?

I’d love to know that the “no killbots, come what may” strategy is sound, but it’s not clear that that’s a stable equilibrium.