Predictably Predictable Futures Talk: Using Expected Loss & Prediction Innovation for Long Term Benefits

I re­cently gave a talk at the 3rd Oxford Work­shop on Global Pri­ori­ties Research

This cov­ered a few top­ics:

  1. How we can think about and use Ex­pected Loss to un­der­stand and op­ti­mize judge­men­tal fore­cast­ing se­tups.

  2. Some thoughts around how (1) ap­plies to our de­ci­sions about long-term fore­cast­ing.

  3. A few slides on Fore­told and how similar sys­tems could be used to help with (1) and (2).

I re­cently recorded a quick ver­sion of this talk on YouTube. As always, I’d be cu­ri­ous to get feed­back on these ideas.

Origi­nal Talk Abstract

The last sev­eral decades have brought a large num­ber of con­tri­bu­tions to statis­ti­cal and judge­men­tal fore­cast­ing en­deav­ors. The next sev­eral decades may bring sig­nifi­cantly more.

As things get more ad­vanced, not only will our un­der­stand­ing of how to pre­dict dis­tant events im­prove, but our un­der­stand­ing of the ac­cu­racy of these fore­casts will im­prove as well. This should help us to re­solve ques­tions re­gard­ing when and how much to trust these fore­casts.

This talk will pre­sent an overview of what we can ex­pect if things go well and how think­ing about this can in­form our cur­rent ac­tions when op­ti­miz­ing for the long term fu­ture.


Ex­pected Loss Graphs

Blog posts on “Am­plify­ing gen­er­al­ist re­search via fore­cast­ing”:



No comments.