[extropy-chat] Inside Vs. Outside Forecasts

[Eliezer S. Yudkowsky]

Robin Hanson wrote:
> 
> This nice thing about using formal statistical analysis is that if you 
> do it right it should tell you when you are screwed.  If your posterior 
> isn't much different from you prior, why then the data didn't tell you 
> much.  So there's not much harm in trying the statistics.  We can't 
> pretend we don't have beliefs on hard problems - so we have to try what 
> we can to get the best estimates we can.

Ideally, yeah.  If you don't have reference class fights.  Let's put it 
this way:  If you set up a reference class and your formal statistical 
analysis claims we're not screwed and makes a definite prediction with a 
confidence bound, and I set up a different reference class and my formal 
statistical analysis claims we're not screwed and makes a definite 
prediction with a confidence bound, and the two confidence bounds don't 
remotely overlap, then I stand by my statement that we're screwed.

-- 
Eliezer S. Yudkowsky                          http://singinst.org/
Research Fellow, Singularity Institute for Artificial Intelligence