How do you maximize the set of possible future actions, for as many as possible, for the longest time?