Provides scenarios of "perverse intantiation", i.e. the malicious possible misunderstanding of user-defined goals by AI programs