Superintelligence 23: Coherent extrapolated volition — LessWrong