Why bother with simulations of governance systems and not governance itself at that point?

I do understand “the risk” of putting AI being the steering wheel but if you’re already going to be trusting it this far, the last step probably doesn’t actually matter.