Probably not obvious for folks not into machine-learning (or other compsci areas with similar terminology), I know... not the best cross-disciplinary metaphor :)
What you refer to I would call "current/local reward" vs "total (lifetime) reward", or "punishment" if you take the negative.
And, by "local optima" I mean that "any small tweak/modification to the current set of laws" would result in worse systems, even if there probably are combinations of laws/rules that would make a better society further away in the space of possibilities. To get from current state to a higher local optima (or to an imaginary global one), would involve going through much worse intermediary states - things like near destruction of civilization or other stuff nobody would want to live through...
Maximizing human well-being seems axiomatic to our project here; not a 'playpen'.