/AI-timelines-deference

A very brief analysis of the AI timelines deference survey conducted by Sam Clarke (GovAI)

Primary LanguageJupyter NotebookMIT LicenseMIT

AI timelines deference

A very brief analysis of the AI timlines deference survey conducted by Sam Clarke. Please see the forum post (EAForum and LessWrong) for the motivation and outline of this survey, and the notebook for some (very basic) exploratory data analysis of the responses.

Image description

Some people and organisations have highly correlated views (as evidenced by co-authoring the same report or repeatedly expressing similar views in public). This figure shows the number of deference responses for two influential, but ideologically different, clusters.

Image description

If we bin the overall responses into six main categories ("Open Philanthropy cluster", "MIRI cluster", "Inside view" responses, deferences to "Daniel Kokotajlo", and simply "Everyone else" who received a deference), we get something that looks like this. Note that "Open Philanthropy cluster" = [Ajeya Cotra, Holden Karnofsky, Paul Christiano, and the Bioanchors report] and "MIRI cluster" = [MIRI, Eliezer Yudkowsky].