One question—why expect much of an intermediate “big robots” phase? If you really do have a superintelligence, nanotech might turn out to be in many ways easier than clunky humanoid robots operating in the environment, and faster to iterate and develop on. We’re already assuming at this point intelligence has skyrocketed way past what it would be needed to do that. Only reasons not to go there would be if the AI itself judges it’s too dangerous, or downright impossible, and in that case you never go there at all.
One question—why expect much of an intermediate “big robots” phase? If you really do have a superintelligence, nanotech might turn out to be in many ways easier than clunky humanoid robots operating in the environment, and faster to iterate and develop on. We’re already assuming at this point intelligence has skyrocketed way past what it would be needed to do that. Only reasons not to go there would be if the AI itself judges it’s too dangerous, or downright impossible, and in that case you never go there at all.