Re:Predictions as a Substitute for Reviews
by bo1024 on Wednesday August 12, 2020 @10:11AM
(#24132217)
Great insight, this is a general issue with making decisions based on predictions. It comes up in Hanson's "Futarchy" / decision markets too.
For example, if you have a huge bias against recruiters and never take their calls because your predictions are < 3%, but the true probability is 50%, you'll never find out those predictions are wrong.
One way to mitigate this is to add random exploration. Whenever you decide not to do something, with probability say 1/20, just do it anyway.
For example, if you have a huge bias against recruiters and never take their calls because your predictions are < 3%, but the true probability is 50%, you'll never find out those predictions are wrong.
One way to mitigate this is to add random exploration. Whenever you decide not to do something, with probability say 1/20, just do it anyway.