Stuart Russell warns that we "had better be quite sure that the purpose put into the machine is the purpose which we really desire" to avoid superintelligent AI conflicting with human objectives.