Both Bostrom and Omohundro argue these values should be used in trying to predict a superintelligence's behavior, since they are likely to be the only set of values shared by most superintelligences. They also note that these values are consistent with safe and beneficial AIs as well as unsafe ones. Omohundro says: