I'm not sure if you believe you're disagreeing with me, but you're not. Goals are necessary for an optimization process, but errors in specifying those goals (or in the goal-specification procedure itself) are potentially existential risks, for a powerful enough optimization process.