An overlooked part of being a good Bayesian is revisiting your past failures of imagination, so that past stuggles with Knightian Uncertainty can be transformed into simple failures to accurately forecast probailities.
I posted earlier today about things getting weirder, but it’s worth considering the exercise proposed by Jonathan V. Last over at The Bulwark, where he goes through his own worst case scenario from November of 2024 and then compares it to the current observed reality.
JVL provides the following list of current events that he never even considered as possible. Going through the list below, which of these would you have considered as genuine possibilities? Not whether you predicted they would happen, but whether you would have even considered in your forecast that they could happen.
I think there are only two items, maybe three on that list I would have thought of as >1% chance of happening. That’s a failure of my imagination and I don’t think I am alone. If we’re good Bayesians, I think that means not just updating our priors, but updating how we update, and opening the door to the darker parts of our imagination when forecasting going forward. No, I’m not enjoying it either.
Economist Writing Every Day 