The podcast discussion boils down to what I posted above:
Yudkowsky attempts to answer that at the end of the podcast by saying:
Build the off switch.
But the switch he describes is impractical as it requires a global coordination effort that doesn’t and can’t exist. AI is not a single thing that resides in one country or is controlled by one government or one laboratory or relies on the rules/laws of any single culture. AI is this amorphous thing spilling out all over the globe that cannot be contained in any one neatly controlled box. So his “solution” is magical thinking:
Track all the G.P.U.s, or all the A.I.-related G.P.U.s, or all the systems of more than one G.P.U. … put them all in a limited number of data centers under international supervision and try to have the A.I.s being only trained on the tracked G.P.U.s, have them only being run on the tracked G.P.U.s. And then, if you are lucky enough to get a warning shot, there is then the mechanism already in place for humanity to back the heck off.
I don’t think he really believes the off switch he describes can be implemented after the fact. He’s just wistfully laying out what should have been done as a coordinated effort in the early stages of AI development, but that genie is long out of the bottle. What he’s saying is that without this (impossible) off switch, eventually we’re doomed.