Thanks Aaron. Glad to hear the invitee list included a broader list of organizations, and that you’ll consider a more explicit discussion of potential selection bias effects going forward.
(I was the interim director of CEA during Leaders Forum, and I’m now the executive director.)
I think that CEA has a history of pushing longtermism in somewhat underhand ways (e.g. I think that I made a mistake when I published an “EA handbook” without sufficiently consulting non-longtermist researchers, and in a way that probably over-represented AI safety and under-represented material outside of traditional EA cause areas, resulting in a product that appeared to represent EA, without accurately doing so). Given this background, I think it’s reasonable to be suspicious of CEA’s cause prioritisation.
(I’ll be writing more about this in the future, and it feels a bit odd to get into this in a comment when it’s a major-ish update to CEA’s strategy, but I think it’s better to share more rather than less.) In the future, I’d like CEA to take a more agnostic approach to cause prioritisation, trying to construct non-gameable mechanisms for making decisions about how much we talk about different causes. An example of how this might work is that we might pay for an independent contractor to try to figure out who has spent more than two years full time thinking about cause prioritization, and then surveying those people. Obviously that project would be complicated—it’s hard to figure out exactly what “cause prio” means, it would be important to reach out through diverse networks to make sure there aren’t network biases etc.
Anyway, given this background of pushing longtermism, I think it’s reasonable to be skeptical of CEA’s approach on this sort of thing.
When I look at the list of organizations that were surveyed, it doesn’t look like the list of organizations most involved in movement building and coordination. It looks much more like a specific subset of that type of org: those focused on longtermism or x-risk (especially AI) and based in one of the main hubs (London accounts for ~50% of respondents, and the Bay accounts for ~30%).* Those that prioritize global poverty, and to a lesser extent animal welfare, seem notably missing. It’s possible the list of organizations that didn’t respond or weren’t named looks a lot different, but if that’s the case it seems worth calling attention to and possibly trying to rectify (e.g. did you email the survey to anyone or was it all done in person at the Leaders Forum?)
I think you’re probably right that there are some biases here. How the invite process worked this year was that Amy Labenz, who runs the event, draws up a longlist of potential attendees (asking some external advisors for suggestions about who should be invited). Then Amy, Julia Wise, and I voted yes/no/maybe on all of the individuals on the longlist (often adding comments). Amy made a final call about who to invite, based on those votes. I expect that all of this means that the final invite list is somewhat biased by our networks, and some background assumptions we have about individuals and orgs.
Given this, I think that it would be fair to view the attendees of the event as “some people who CEA staff think it would be useful to get together for a few days” rather than “the definitive list of EA leaders”. I think that we were also somewhat loose about what the criteria for inviting people should be, and I’d like us to be a bit clearer on that in the future (see a couple of paragraphs below). Given this, I think that calling the event “EA Leaders Forum” is probably a mistake, but others on the team think that changing the name could be confusing and have transition costs—we’re still talking about this, and haven’t reached resolution about whether we’ll keep the name for next year.
I also think CEA made some mistakes in the way we framed this post (not just the author, since it went through other readers before publication.) I think the post kind of frames this as “EA leaders think X”, which I expect would be the sort of thing that lots of EAs should update on. (Even though I think it does try to explicitly disavow this interpretation (see the section on “What this data does and does not represent”, I think the title suggests something that’s more like “EA leaders think these are the priorities—probably you should update towards these being the priorities”). I think that the reality is more like “some people that CEA staff think it’s useful to get together for an event think X”, which is something that people should update on less.
We’re currently at a team retreat where we’re talking more about what the goals of the event should be in the future. I think that it’s possible that the event looks pretty different in future years, and we’re not yet sure how. But I think that whatever we decide, we should think more carefully about the criteria for attendees, and that will include thinking carefully about the approach to cause prioritization.
Thank you for taking the time to respond, Max. I appreciate your engagement, your explanation of how the invitation process worked this year, and your willingness to acknowledge that CEA may have historically been too aggressive in how it has pushed longtermism and how it has framed the results of past surveys.
In the future, I’d like CEA to take a more agnostic approach to cause prioritisation, trying to construct non-gameable mechanisms for making decisions about how much we talk about different causes.
Very glad to hear this. As you note, implementing this sort of thing in practice can be tricky. As CEA starts designing new mechanisms, I’d love to see you gather input (as early possible) from people who have expressed concern about CEA’s representativeness in the past (I’d be happy to offer opinions if you’d like). These also might be good people to serve as “external advisors” who generate suggestions for the invite list.
Good luck with the retreat! I look forward to seeing your strategy update once that’s written up.
Thanks Aaron. Glad to hear the invitee list included a broader list of organizations, and that you’ll consider a more explicit discussion of potential selection bias effects going forward.
(I was the interim director of CEA during Leaders Forum, and I’m now the executive director.)
I think that CEA has a history of pushing longtermism in somewhat underhand ways (e.g. I think that I made a mistake when I published an “EA handbook” without sufficiently consulting non-longtermist researchers, and in a way that probably over-represented AI safety and under-represented material outside of traditional EA cause areas, resulting in a product that appeared to represent EA, without accurately doing so). Given this background, I think it’s reasonable to be suspicious of CEA’s cause prioritisation.
(I’ll be writing more about this in the future, and it feels a bit odd to get into this in a comment when it’s a major-ish update to CEA’s strategy, but I think it’s better to share more rather than less.) In the future, I’d like CEA to take a more agnostic approach to cause prioritisation, trying to construct non-gameable mechanisms for making decisions about how much we talk about different causes. An example of how this might work is that we might pay for an independent contractor to try to figure out who has spent more than two years full time thinking about cause prioritization, and then surveying those people. Obviously that project would be complicated—it’s hard to figure out exactly what “cause prio” means, it would be important to reach out through diverse networks to make sure there aren’t network biases etc.
Anyway, given this background of pushing longtermism, I think it’s reasonable to be skeptical of CEA’s approach on this sort of thing.
I think you’re probably right that there are some biases here. How the invite process worked this year was that Amy Labenz, who runs the event, draws up a longlist of potential attendees (asking some external advisors for suggestions about who should be invited). Then Amy, Julia Wise, and I voted yes/no/maybe on all of the individuals on the longlist (often adding comments). Amy made a final call about who to invite, based on those votes. I expect that all of this means that the final invite list is somewhat biased by our networks, and some background assumptions we have about individuals and orgs.
Given this, I think that it would be fair to view the attendees of the event as “some people who CEA staff think it would be useful to get together for a few days” rather than “the definitive list of EA leaders”. I think that we were also somewhat loose about what the criteria for inviting people should be, and I’d like us to be a bit clearer on that in the future (see a couple of paragraphs below). Given this, I think that calling the event “EA Leaders Forum” is probably a mistake, but others on the team think that changing the name could be confusing and have transition costs—we’re still talking about this, and haven’t reached resolution about whether we’ll keep the name for next year.
I also think CEA made some mistakes in the way we framed this post (not just the author, since it went through other readers before publication.) I think the post kind of frames this as “EA leaders think X”, which I expect would be the sort of thing that lots of EAs should update on. (Even though I think it does try to explicitly disavow this interpretation (see the section on “What this data does and does not represent”, I think the title suggests something that’s more like “EA leaders think these are the priorities—probably you should update towards these being the priorities”). I think that the reality is more like “some people that CEA staff think it’s useful to get together for an event think X”, which is something that people should update on less.
We’re currently at a team retreat where we’re talking more about what the goals of the event should be in the future. I think that it’s possible that the event looks pretty different in future years, and we’re not yet sure how. But I think that whatever we decide, we should think more carefully about the criteria for attendees, and that will include thinking carefully about the approach to cause prioritization.
Thank you for taking the time to respond, Max. I appreciate your engagement, your explanation of how the invitation process worked this year, and your willingness to acknowledge that CEA may have historically been too aggressive in how it has pushed longtermism and how it has framed the results of past surveys.
Very glad to hear this. As you note, implementing this sort of thing in practice can be tricky. As CEA starts designing new mechanisms, I’d love to see you gather input (as early possible) from people who have expressed concern about CEA’s representativeness in the past (I’d be happy to offer opinions if you’d like). These also might be good people to serve as “external advisors” who generate suggestions for the invite list.
Good luck with the retreat! I look forward to seeing your strategy update once that’s written up.