I like that you pointed out a difference between an ideal regret minimizing agent and actual humans (having a regret threshold that makesyou shutdown). It seems like another important difference between the “proper” regret minimizer and people is that people are bad at scaling regret of possible better outcomes based on the likelyhood of those better outcomes.
It seems like we put possible better outcomes into only a handful of categories, “Well, there was a chance...” “I might have been able to...” “I definitely could have if I tried” and asign regret based on which of those few ranges fits. This would lead to feeling a lot of regret that was disporportional to the chances of a better outcome, which seems to fit how peple act. Unless I misundertand regret minimization as-a-decion-theory, it seems like a “proper” regret minimizer would not do that.
I like that you pointed out a difference between an ideal regret minimizing agent and actual humans (having a regret threshold that makesyou shutdown). It seems like another important difference between the “proper” regret minimizer and people is that people are bad at scaling regret of possible better outcomes based on the likelyhood of those better outcomes.
It seems like we put possible better outcomes into only a handful of categories, “Well, there was a chance...” “I might have been able to...” “I definitely could have if I tried” and asign regret based on which of those few ranges fits. This would lead to feeling a lot of regret that was disporportional to the chances of a better outcome, which seems to fit how peple act. Unless I misundertand regret minimization as-a-decion-theory, it seems like a “proper” regret minimizer would not do that.