I think "build a recursively self-improving Artificial Intelligence WITHOUT a stable, specifiable goal system so that it can improve into a superintelligence and do world optimization" is more frightening, even if less ambitious :)