Computational irreducibility is the phenomenon wherein simple computational systems cannot be predicted to produce specific long-term outputs without actually running the computation for approximately that many steps—you cannot 'jump ahead' in prediction the way exact science has traditionally allowed.