I do want to note that all of the things you list took place around 2017-2018, and our work and plans have changed since then.
My observations about 80k, GPI, and CFAR are all ongoing (though they originated earlier). I also think there are plenty of post-2018 examples related to CEA’s work, such as the Introductory Fellowship content Michael noted (not to mention the unexplained downvoting he got for doing so), Domassoglia’s observations about the most recent EAG and EAGx (James hits on similar themes), and the late 2019 event that was framed as a “Leader’s Forum” but was actually “some people who CEA staff think it would be useful to get together for a few days” (your words) with those people skewing heavily longtermist. I think all of these things “contribute to the view that EA essentially is longtermism/AI Safety?”(though of course longtermism could be “right” in which case these would all be positive developments.)
Where we have to decide a content split (e.g. for EA Global or the Handbook), I want CEA to represent the range of expert views on cause prioritization. I still don’t think we have amazing data on this, but my best guess is that this skews towards longtermist-motivated or X-risk work (like maybe 70-80%).
I would love someone to do a proper survey of everyone (trying to avoid one’s own personal networks) who has spent >1 year thinking about cause prioritization with a scope-sensitive and open-minded lens. I’ve tried to commission someone to do this a couple of times but it hasn’t worked out. If someone did this, it would help to shape our content, so I’d be happy to offer some advice and could likely find funding. If anyone is interested, let me know!
I agree with Linch’s concern about the selection bias this might entail: “a core problem with the “neutrality”/expert-views framing of this comment is selection bias. We would naively expect people who spend a lot of time on cause prioritization to systematically overrate (relative to the broader community) both the non-obviousness of the most important causes, and their esotericism.”
Also related to selection bias: most of the opportunities and incentives to work on cause prioritization have been at places like GPI or Forethought Foundation that use a longtermist lens, making it difficult to find an unbiased set of experts. I’m not sure how to get around this issue. Even trying to capture the views of the EA community as a whole (at the expense of deferring to experts) would be problematic to the extent “mistakes” have shaped the composition of the community by making EA more attractive to longtermists and less attractive to neartermists.
I appreciate that CEA is looking to “outsource” cause prioritization in some way. I just have concerns about how this will work in practice, as it strikes me as a very difficult thing to do well.
I also strongly share this worry about selection effects. There are additional challenges to those mentioned already: the more EA looks like an answer, rather than a question, the more inclined anyone who doesn’t share that answer is simply to ‘exit’, rather than ‘voice’, leading to an increasing skew over time of what putative experts believe. A related issue is that, if you want to work on animal welfare or global development you can do that without participating in EA, which is much harder if you want to work on longtermism.
Further, it’s a sort of double counting if you consider people as experts because they work in a particular organisation when they would only realistically be hired if they had a certain worldview. If FHI hired 100 more staff, and they were polled, I’m not sure we should update our view on what the expert consensus is any more than I should become more certain of the day’s events by reading different copies of the same newspaper. (I mean no offence to FHI or its staff, by the way, it’s just a salient example).
My observations about 80k, GPI, and CFAR are all ongoing (though they originated earlier). I also think there are plenty of post-2018 examples related to CEA’s work, such as the Introductory Fellowship content Michael noted (not to mention the unexplained downvoting he got for doing so), Domassoglia’s observations about the most recent EAG and EAGx (James hits on similar themes), and the late 2019 event that was framed as a “Leader’s Forum” but was actually “some people who CEA staff think it would be useful to get together for a few days” (your words) with those people skewing heavily longtermist. I think all of these things “contribute to the view that EA essentially is longtermism/AI Safety?”(though of course longtermism could be “right” in which case these would all be positive developments.)
I agree with Linch’s concern about the selection bias this might entail: “a core problem with the “neutrality”/expert-views framing of this comment is selection bias. We would naively expect people who spend a lot of time on cause prioritization to systematically overrate (relative to the broader community) both the non-obviousness of the most important causes, and their esotericism.”
Also related to selection bias: most of the opportunities and incentives to work on cause prioritization have been at places like GPI or Forethought Foundation that use a longtermist lens, making it difficult to find an unbiased set of experts. I’m not sure how to get around this issue. Even trying to capture the views of the EA community as a whole (at the expense of deferring to experts) would be problematic to the extent “mistakes” have shaped the composition of the community by making EA more attractive to longtermists and less attractive to neartermists.
I appreciate that CEA is looking to “outsource” cause prioritization in some way. I just have concerns about how this will work in practice, as it strikes me as a very difficult thing to do well.
I also strongly share this worry about selection effects. There are additional challenges to those mentioned already: the more EA looks like an answer, rather than a question, the more inclined anyone who doesn’t share that answer is simply to ‘exit’, rather than ‘voice’, leading to an increasing skew over time of what putative experts believe. A related issue is that, if you want to work on animal welfare or global development you can do that without participating in EA, which is much harder if you want to work on longtermism.
Further, it’s a sort of double counting if you consider people as experts because they work in a particular organisation when they would only realistically be hired if they had a certain worldview. If FHI hired 100 more staff, and they were polled, I’m not sure we should update our view on what the expert consensus is any more than I should become more certain of the day’s events by reading different copies of the same newspaper. (I mean no offence to FHI or its staff, by the way, it’s just a salient example).