Less naive proposals would involve probing the neuroscience of motivation at the algorithmic level.

I think that seems more naive - once you consider the timescales. Machine intelligence seems likely to come before we have the required brain-scanning/brain-understanding technology. Maybe once we have intelligent machines, then we can figure out the brain - but we likely can't use brain scans to let the intelligent machine know what we want in the first place - because the timing will be wrong.