Stephen Clare

Research Fellow @ Forethought Foundation
1659Joined Aug 2021


I'm a Research Fellow at the Forethought Foundation, where I worked on What We Owe The Future with Will MacAskill and the rest of the Forethought team. Now I'm working to assess the probability of great power conflict and its long-term effects.


I don't think this is true:

Most people in EA don’t discount the future much, if at all.

In the latest EA Survey (from 2020, admittedly), global poverty (not a longtermist cause) still ranked as the top cause among respondents. I personally find analyses that compare economic growth to global health interventions in terms of short-term impact valuable for people prioritizing within the global poverty cause area.

All the endnotes are online here. ctrl+f "2.22" to find this one.

Congrats on this report Rani - it's excellent, and an important contribution to the longtermist conflict research literature.

The post is long and detailed, so I want to draw out the big implications for Forum readers who are less immersed in what's going on with this kind of research. The story you tell in this report is something like:

  • Track II diplomacy is particularly interesting among interventions to reduce conflict risks because it's more direct than research and can easily be funded by philanthropists
  • But it's really hard to evaluate the effect of Track II dialogues because they don't happen that often, we can't do RCT or RCT-like studies, and many factors influence both the feasibility of Track II and the outcomes of conflict (endogeneity)
  • We have some hope, though, because as John Halstead points out in "Evaluating Policy Organisations", for policy interventions we just need to know how big the biggest wins were - that's where (almost) all the impact comes from
  • So we can just look at a few cases, and do the messy, journalistic work of reconstructing the theory (and practice) of change to understand what actually happened, and whether the dialogues had an effect
  • Because we're looking for big wins, you do this for the Pugwash Conferences, the paradigmatic example of Track II dialogues

Before you started this project, I was genuinely unsure about what you'd find . The Pugwash Conferences are so narratively compelling (inspired by Einstein-Russell, involving scientists from around the world, etc.) that it was plausible to me that they became famous for being cool, not for having a policy impact. But your conclusion is that it seems pretty likely that they did have an effect! That the dialogues allowed the Americans and Soviets to build relationships and exchange arguments, and that this led some participants to change their mind on important issues. The Soviet scientists then went on to influence policymakers.

Of course, it's hard to be certain. John discusses the problems with reconstructing these case studies for contemporary policy organisations. It's even harder when you're reconstructing a case that happened 50 years ago and involved policy changes within a secretive government that worked in a language you don't speak! But in section 4 you get really stuck into the details and make a relatively compelling case for thinking the dialogues made a difference.

I'm really impressed with your work on this post! I've updated towards thinking Track II dialogues will be worth funding in certain cases, and we should definitely be looking for Pugwash-like opportunities to support (there are clear parallels between the Anti-Ballistic Missile Treaty and various military AI or other emerging tech issues today).

(Disclaimer: I supervised Rani this summer while she wrote this report)

Thanks for this. There's a lot to digest and I'll come back in a day or two with a longer comment. But I wanted to quickly jump in and suggest you revise the second bullet point in your summary. It reads:

In particular, risk of a US-China war is real (Metaculus gives it 50/50 odds of happening by 2050).

This is not true. The Metaculus question you're referring to is this one, in which the median forecaster prediction for the next great power war is 2045. However this is not specifically about a US-China war. Rather, this includes any war between two countries which are in the top 10 by military spending according to SIPRI's index. That includes the US and China, but, crucially,  Russia, Germany, India, Japan, and South Korea (among others) as well.

Given the number of potential combatant-pairs in that set, some of which are probably about as likely to go to war as the US and China, I think your bullet point as written is very misleading. The odds of a serious US-China war by 2050 are closer to 10-20% IMO (a more specific Metaculus question has it at 16% by 2035).

There's also a Metaculus question about the chance of >100 deaths in US-China conflicts before 2050, with the community prediction at 53%. But 100 deaths is well below the threshold for a war and way, way below even your "shorter, limited" war scenario (which assumes 25,000 combined US-China deaths, plus >200,000 Taiwanese deaths). Because war deaths follow a heavy-tailed distribution, I think a lot of the conflict probability mass lies between 100 and 25,000 deaths.

I think a 10-20% chance of war is probably still high enough for many of your arguments to go through. But I think starting your piece off with such a misrepresentation does it a disservice. I recommend you edit that bullet point.

(Like I said - overall I'm really happy to see this piece and will come back with a more substantial comment in a day or two!)

I'm a bit surprised that you didn't discuss international climate finance in the neglectedness section. The OECD estimates that 21% of ~$79 billion of annual climate financing for developing countries goes to adaptation.

I'm not sure the economic losers point is a strong one. It seems to model economic opportunities in an area as fixed, such that someone gets pushed down a rung to worse opportunity when someone else (in this case, someone treated by deworming) enters at a higher rung. But elasticities are complex and the kings of economic opportunities available might also be changed by deworming.

I think it would be better to focus on productivity. And it seems likely that the productivity externalities of deworming, if indeed it makes treated individuals more productive, are negligible or slightly positive. I don't see how it could make untreated individuals less productive, and therefore worse off.

You're so productive! I'd love to read a blog post talking about how you manage multiple projects, structure your day, and organize your work.

I appreciate a lot of what's in this post! I do think it's the case that we as a community have only explored a fraction of the space of development initiatives. It's really plausible to me that there are more impactful approaches or things to fund out there.

That said, I feel this swings too hard against the traditional EA approach of careful analysis and prioritization. After working in development I'm kind of allergic to calls to do things like "establish ecosystems that empower altruistic leaders at scale, and that allow those that build and those that analyze to collaborate to each other's mutual benefit". Many of these programs do exist. Only a few of them are impactful.

I did a quick Google search and found that something kind of like what you propose does actually exist. The Global Shapers Community has a hub in Port-Au-Prince, which "brings together exceptional Haitian youths whose mission is to shape the community to which they belong". I found it useful as an example of what something like this would actually look like in practice. Anyone can glance over their website and form their own view on how promising they think it would be to give this organization (or something like it) $1M.

I don't think this project, or other projects like what you discuss in the post, are un-analyzable by EA evaluators. I do think it's likely that many of them just sound better in theory than in practice.

Yep, totally agree that this would be tricky! There'd be a lot of details to think through. I would note that Vivalt does run regressions where, e.g., the kind of organization implementing the program (government vs NGO) is included as a covariate, and the coefficient on sample size doesn't change much (-0.011 vs -0.013 in the single linear regression; see table 7, p. 31).

On non-specific discount factors: one approach which I was interested in when doing a lot of this work was to use estimates we have of how much effect sizes shrink when more and/or larger studies are conducted.

For example, in this paper Eva Vivalt, using a sample of impact evaluations, regresses effect size on variables like number of studies and sample size. As one would expect, the larger the sample size, the smaller the estimated effect size. I always wondered if you could just use the regression coefficients she presents to estimate how much an effect size would be expected to shrink if one conducted a larger study.

I don't think this strikes at exactly what you or HLI are trying to get at. But I do think it's valuable to ponder how we might get at "principled" discount rates, given all we know about the validity problems these studies exhibit.

Load More