How do you rationalize building artificial superintelligence that may exterminate humanity?