One reason why EA takes AI risks seriously is exactly because of how hard it is to define what is "good" or what "utility" makes humans happy.