I thought you were referring only to unbounded utility functions. If utility is bounded, your statement EU(F) → 0 as p(F) → 0 is true (though the meaning of the notation might be confusing; expectations are supposed to be taken over the entire probability space), but I don’t think it gives reason to believe that everything will cancel out nicely, though it is possible that my intuition is distorted by thinking about busy-beaver size bounds, which are unlikely to be implemented in real agents anyways.
I thought you were referring only to unbounded utility functions. If utility is bounded, your statement EU(F) → 0 as p(F) → 0 is true (though the meaning of the notation might be confusing; expectations are supposed to be taken over the entire probability space), but I don’t think it gives reason to believe that everything will cancel out nicely, though it is possible that my intuition is distorted by thinking about busy-beaver size bounds, which are unlikely to be implemented in real agents anyways.