I’m unfortunately not familiar with Friedman enough to answer, but I really think their idea was “the AI sits down, reasons every possible outcome from first principles, then chooses the plan for its whole future”. The explicitly defined utility function is there