2023-08-02 Wed 18:45 In [[Ref. Philip Tetlock 2015 - Superforecasting- The Art and Science of Prediction]], the best way to make a forecast is to start with the outside view and then adjust based on [[Inside view]]. Outside view is when you say "what kind of thing is this" and "how do things like this normally behave". So eg if you're trying to forecast the probabilty that Donny Dictator is gonna get overthrown this month, you would say okay what kind of thing is this? Dictator coup. Or maybe government coup. And then you look at, okay, how common is that event? For any random dictator (or government) what's the probability of their getting overthrown in a random month? Ie, at this stage, you ignore all your special inside-view knowledge about Donny Dictator and why it's not looking good for him in particular. Of course, you have a lot of wiggle room in how you identify the comparison class for Donny. Setting that aside, this is the idea. So maybe you find that on an outside view the probability is 1%. Then to complete your forecast, you go to the inside view. There are riots in the streets. Okay maybe that increases it by 50%, to 1.5%. Maybe Donny's brother was killed this morning. Maybe push it up to 2%. Etc. This method will often get you a very different result from if you had gone the other way around.