If one is a consequentialist, one at least implicitly makes decisions based on how good or bad the consequences are. As such, you must implicitly have a function
f that maps from the set of attainable potential futures into a totally ordered set (let's say real numbers) so you can pick the best one(s) and avoid the worst.
Coming up with such a function is fraught with difficulty, however.
- Functions that depend on instantaneous state recommend nonsensical courses of action like binging on chocolate because you're enjoying it while you're eating it.
- Functions that consider all time with equal weight don't converge to a finite value when taken arbitrarily far into the future, allowing arbitrarily bad conditions now as long as they are expected to lead to the tiniest long-term improvement (plus these are impossible to evaluate due to uncertainty)
- Functions that decrease exponentially have nice self-similarity properties (you maintain your relative quantitative judgments of actions whether they occur now or in the future), but you end up with highly counterintuitive results that any tiny thing that improves value now is worth committing everyone to torture and/or destroying the universe as long as it happens far enough in the future
- It's much easier to define functions on individuals than to know how to combine results from individuals, but even if you know how to combine results it's made even trickier because individuals don't even stay stable over time (new ones are born, old ones die, etc.) and the existence/non-existence of individuals may depend on your actions.
What strategies are there to deal robustly with these problems? Alternatively, if the solutions are too elaborate to be done justice in an answer, where can I find a good resource that discusses these issues and/or proposes solutions?