When you notice that counterfactual values can sum up to more than 100%, I think that the right answer is to stop optimizing for counterfactual values.
It’s less clear cut, but I think that optimizing for Shapley value instead is a better answer—though not perfect.
I think of Shapley values as just one way of assigning credit in a way to optimise incentives, but from what I’ve seen, it’s not obvious it’s the best one. (In general, I haven’t seen any principled way of assigning credit that always seems best.)
I’d also point to Shapley values.
When you notice that counterfactual values can sum up to more than 100%, I think that the right answer is to stop optimizing for counterfactual values.
It’s less clear cut, but I think that optimizing for Shapley value instead is a better answer—though not perfect.
I think of Shapley values as just one way of assigning credit in a way to optimise incentives, but from what I’ve seen, it’s not obvious it’s the best one. (In general, I haven’t seen any principled way of assigning credit that always seems best.)