The reward hypothesis—that all goals and purposes can be understood as maximizing the expected value of cumulative scalar reward—is sufficient for generality and does not require modifications like multiple objectives, constraints, or risk sensitivity to achieve general intelligence