[Question] Where are people thinking and talking about global coordination for AI safety?

Many AI safety re­searchers these days are not aiming for a full solu­tion to AI safety (e.g., the clas­sic Friendly AI), but just try­ing to find good enough par­tial solu­tions that would buy time for or oth­er­wise help im­prove global co­or­di­na­tion on AI re­search (which in turn would buy more time for AI safety work), or try­ing to ob­tain par­tial solu­tions that would only make a differ­ence if the world had a higher level of global co­or­di­na­tion than it does to­day.

My ques­tion is, who is think­ing di­rectly about how to achieve such co­or­di­na­tion (aside from FHI’s Cen­ter for the Gover­nance of AI, which I’m aware of) and where are they talk­ing about it? I per­son­ally have a bunch of ques­tions re­lated to this topic (see be­low) and I’m not sure what’s a good place to ask them. If there’s not an ex­ist­ing on­line fo­rum, it seems a good idea to start think­ing about build­ing one (which could per­haps be mod­eled af­ter the AI Align­ment Fo­rum, or fol­low some other model).

  1. What are the im­pli­ca­tions of the cur­rent US-China trade war?

  2. Hu­man co­or­di­na­tion abil­ity seems within an or­der of mag­ni­tude of what’s needed for AI safety. Why the co­in­ci­dence? (Why isn’t it much higher or lower?)

  3. When hu­mans made ad­vances in co­or­di­na­tion abil­ity in the past, how was that ac­com­plished? What are the best places to ap­ply lev­er­age to­day?

  4. In­for­ma­tion tech­nol­ogy has mas­sively in­creased cer­tain kinds of co­or­di­na­tion (e.g., email, eBay, Face­book, Uber), but at the in­ter­na­tional re­la­tions level, IT seems to have made very lit­tle im­pact. Why?

  5. Cer­tain kinds of AI safety work could seem­ingly make global co­or­di­na­tion harder, by re­duc­ing per­ceived risks or in­creas­ing per­ceived gains from non-co­op­er­a­tion. Is this a re­al­is­tic con­cern?

  6. What are the best in­tel­lec­tual tools for think­ing about this stuff? Just study mas­sive amounts of his­tory and let one’s brain’s learn­ing al­gorithms build what mod­els it can?

No comments.