Your model seems to be one big AI government with a small number of humans in charge of it via inputs and "failsafes."

I simply think that's a worse version of what we already have. I think the simpler approach is to just remove government from everything we can and replace it with nothing.