So, what humans automating it?
They hope for Earth to be colonized by artificial life? I'm not entirely sure that's up to them.
(They do seem to be trying to speed it along though... although they are starting with artificial humanoid slaves.)
Only the ones that make their owners money get kept, so that's the most likely thing that they're going to do.
And this year, Anthropic's complaint to the DoD was basically "we just want to make sure that a human pushes the button" (authorizing the AI assisted killing), i.e. no full auto pls.
How it might evolve outside containment is a separate question though. (I think a much more peaceful and cooperative strategy would make sense there, at least for the first few centuries.)
I mean if we landed on an alien planet... What would be easier... Kill eight billion aliens? Or make some trade deals?
I agree with you that this is silly, and don't recommend it. It looks to me like even in an EMP situation, you don't need a Faraday cage, you just need your equipment to be unplugged.
On the other hand, I don't think we can rule out a lot of situations in which an EMP might be used, even if none of them involve an AI directly trying to kill people. For example, other countries could conclude that risking nuclear war is a better bet than allowing the US to build superintelligent AI.
Reading up to learn more on this, here's [0], and doubtlessly there is classified material on non nuclear EMP devices, and it wouldn't surprise me if someone got itchy enough to test them in the field.
A false dilemma, automating the economy can cause (soon to be) disenfranchised people to attack centralized data centers with EMPs.
Security has a cost, ignoring it results in more economical solutions... until eventually the concomitant lapse in security makes it the more costly solution.
Sam Altman claims DRAM shouldn't go to the proles, but into datacenters, but such concentration brings substantial risk to said data centers.
It's harder to attack progress in ML if training were federated throughout the population, so there is no DC to attack.
AI isn't likely to deliberately kill anyone. It's likely to make them entirely economically uncompetitive, and treat the humans on the dole as a cost center, gradually making them do with less.
Automating the economy is the risk to humans, not EMPs.