Differences

This shows you the differences between two versions of the page.

--- public_testing_of_relevant_skills_or_abilities [2022/12/17 23:20] – katjagrace
+++ public_testing_of_relevant_skills_or_abilities [2023/02/05 20:21] (current) – removed katjagrace
@@ Line 1: / Line 1: @@
-====== Public testing of relevant skills or abilities ======
-Main problem: people are scared of sounding stupid in discussing AI risk, and so fail to raise questions that they should raise.
-Idea: everyone in the AI safety community does tests estimating e.g. math skill, calibration.
-Merits of this idea:
-  * Good evidence on competence allows better allocation of trust to different people's abilities
-  * Good evidence on competence reduces reliance on bad evidence on competence (e.g. does the person themselves confidently assert their own intellectual dominance?)
-  * Good evidence on competence plausibly reduces reliance on biases that rest on weak or no evidence (e.g. it is harder to implicitly discount the views of women, or of quiet people, if you have strong evidence about their actual abilities.)
-  * Good evidence on competence available to all reduces the potential stakes from saying things that may appear stupid. To the extent that the incentive to not appear stupid shapes conversations, this might be a huge improvement.
-Considerations against:
-  * Seems brutal, unpleasant to be judged negatively in front of everyone
-  * Antithetical to general sense of peerhood and equality, which probably has various benefits e.g. better discussions, better relationships.
-Examples:
-  * Forecasting track records and prediction market profits are somewhat this, though a small fraction of people participate, and it can be hard to compare them, even within the same platform.