The short version of a long argument is that there are no known AI control (motivation) algorithms that are stable in the limit as the supergoal statements become abstract enough to drive a general intelligence.

At least, not yet?