Comment author: Julia_Wise  (EA Profile) 10 September 2018 02:17:35PM *  16 points [-]

I like this suggestion - personally I feel a lot of uncertainty about what to prioritize, and given that a portion of my donations go to near-term work I'd enjoy taking part in discussion about how to best do that, even if I'm also seriously considering whether to prioritize long-term work. But I'd be totally happy to have the topic of that space limited to near-term work.

Comment author: Justis 10 September 2018 03:15:28PM 5 points [-]

+1. I'm in a very similar position - I make donations to near-term orgs, and am hungry for discussion of that kind. But because I sometimes do work for explicitly long-term and x-risk orgs, it's hard for me to be certain if I qualify under current wording.

Comment author: Milan_Griffes 16 August 2018 04:39:48PM 7 points [-]

The "what causes should CEA represent?" issue seems especially tricky because the current canonical EA cause areas have very different metrics underpinning them.

Global development & animal welfare usually use GiveWell-style cost-effectiveness analysis to determine what's effective.

X-risk usually uses theoretical argument & back-of-the-envelope estimates to determine effectiveness.

I'm not sure what movement building uses – probably theory and back-of-the-envelope as well?

Anyway, point is that there's not a meta-metric that the current cause areas use to compare against each other.

So when considering a new cause area, should we use the x-risk standard of effectiveness? Or the global development one? (rhetorical)

Seems tricky – I'm glad CEA is thinking about this.

Comment author: Justis 19 August 2018 09:23:45PM 1 point [-]

I really like the Open Philanthropy Project's way of thinking about this problem:

The short version (in my understanding): 1. Split assumptions about the world/target metrics into distinct "buckets". 2. Do allocation as a two step process: intra-bucket on that bucket's metric, and inter-bucket separately using other sorts of heuristics.

(If you like watching videos rather than reading blog posts, Holden also discussed this approach in his fireside chat at EAG 2018: San Francisco.)

Comment author: Kerry_Vaughan 18 August 2018 12:25:39AM *  11 points [-]

Thanks Sam! This is really helpful. I'd be interested in talking on Skype about this sometime soon (just emailed you about it). Some thoughts below:

Is longtermism a cause?

One idea I've been thinking about is whether it makes sense to treat longtermism/the long-term future as a cause.

Longtermism is the view that most of the value of our actions lies in what happens in the future. You can hold that view and also hold the view that we are so uncertain about what will happen in the future that doing things with clear positive short-term effects is the best thing to do. Peter Hurford explains this view nicely here.

I do think that longtermism as a philosophical point of view is emerging as an intellectual consensus in the movement. Yet, I also think there are substantial and reasonable disagreements about what that means practically speaking. I'd be in favor of us working to ensure that people entering the community understand the details of that disagreement.

My guess is that while CEA is very positive on longtermism, we aren't anywhere near as positive on the cause/intervention combinations that longtermism typically suggests. For example, personally speaking, if it turned out that recruiting ML PhDs to do technical AI-Safety didn't have a huge impact I would be surprised but not very surprised.

Threading the needle

My feeling as I've been thinking about representativeness is that getting this right requires threading a very difficult needle because we need to optimize against a large number of constraints and considerations. Some of the constraints include:

  • Cause areas shouldn't be tribes -- I think cause area allegiance is operating as a kind of tribal signal in the movement currently. You're either on the global poverty tribe or the X-risk tribe or the animal welfare tribe and then people tend to defend the views of the tribe they happen to be associated with. I think this needs to stop if we want to build a community that can actually figure out how to do the most good and then do it. Focusing on cause areas as the unit of analysis for representativeness entrenches the tribal concern, but it's hard to get away from because it's an easy-to-understand unit of analysis.
  • We shouldn't entrench existing cause areas -- we should be aiming for an EA that has the ability to shift its consensus on the most pressing problems as we learn more. Some methods of increasing representativeness have the effect of entrenching current cause areas and making intellectual shifts harder.
  • Cause-impartiality can include having a view -- cause impartiality means that you do an impartial calculation of impact to determine what to work on. Such a calculation should lead to developing views on what causes are most important. Intellectual progress probably includes decreasing our uncertainty and having stronger views.
  • The view of CEA staff should inform, but not determine our work -- I don't think it's realistic or plausible for CEA to take actions as if we have no view on the relative importance of different problems, but it's also the case that our views shouldn't substantially determine what happens.
  • CEA should sometimes exercise leadership in the community -- I don't think that social movements automatically become excellent. Excellence typically has to be achieved on purpose by dedicated, skilled actors. I think CEA will often do work that represents the community, but will sometimes want to lead the community on important issues. The allocation of resources across causes could be one such area for leadership although I'm not certain.

There are also some other considerations around methods of improving representativeness. For example, consulting established EA orgs on representativeness concerns has the effect of entrenching the current systems of power in a way that may be bad, but that gives you a sense of the consideration space.

CEA and cause-impartiality

Suggestion: CEA should actively champion cause impartiality

I just wanted to briefly clarify that I don't think CEA taking a view in favor of longtermism or even in favor of specific causes that are associated with longtermism is evidence against us being cause-impartial. Cause-impartiality means that you do an impartial calculation of the impact of the cause and act on the basis of that. This is certainly what we think we've done when coming to views on specific causes although there's obviously room for reasonable disagreement.

I would find it quite odd if major organizations in EA (even movement building organizations) had no view on what causes are most important. I think CEA should be aspiring to have detailed, nuanced views that take into account our wide uncertainty, not no views on the question.

Making people feel listened to

I broadly agree with your points here. Regularly talking to and listening to more people in the community is something that I'm personally committed to doing.

Your section on representatives feels like you are trying to pin down a way of finding an exact number so you can say we have this many articles on topic x and this many on topic y and so on. I am not sure this is quite the correct framing.

Just to clarify, I also don't think trying to find a number that defines representativeness is the right approach, but I also don't want this to be a purely philosophical conversation. I want it to drive action.

Comment author: Justis 19 August 2018 09:08:31PM 6 points [-]

Disclosure: I copyedited a draft of this post, and do contract work for CEA more generally

I don't think that longtermism is a consensus view in the movement.

The 2017 EA Survey results had more people saying poverty was the top priority than AI and non-AI far future work combined. Similarly, AMF and GiveWell got by far the most donations in 2016, according to that same survey. While I agree that someone can be a longtermist and think that practicality concerns prioritize near-term good work for now anyway, I don't think this is a very compelling explanation for these survey results.

As a first pass heuristic, I think EA leadership would guess correctly about community-held views more often if they held the belief "the modal EA-identifying person cares most about solving suffering that is happening in the world right now."