I feel like if it took thirty or forty more years of research before a more advanced, more goal-oriented AI became capable of destroying humanity for paperclips, that would still be a bad outcome. The road ahead is long and we're moving too fast to only consider what's in the rear view mirror.