TL;DR

• Most community-building effort currently goes towards creating new, highly-engaged EAs, yet the vast majority of people who can do things to help further EA goals will not wish to be highly-engaged.
• You don’t need to actually be an “EA” to do effectively altruistic things, hence why influencing people with EA ideas can be very useful.
• While we already do some influencing, we might want to do more on the margin, especially if we feel urgency about problems like AI alignment.
• EA is special; we should try sharing the ideas/mental models/insights we have with those who can do good things with them.
• It would be useful to have an idea as to how cost-effective it is to try influencing others relative to creating highly-engaged EAs.

Epistemic status

Quite uncertain. I’m more confident in the sign of these arguments than the magnitude. This post was mostly informed by my impressions from being highly involved within the EA community over the last year or so, as well as the time I spent working at Giving What We Can and my current work at GovAI. All views articulated here are my own and do not reflect the stances of either of these organisations, or any other institutions I’m affiliated with. Finally, I’m hoping to spark a conversation rather than to make any sweeping declarations about what the community’s strategy ought to be.

I was inspired to write this post after reading Abraham Rowe’s list of EA critiques he would like to read.

Introduction

I’m writing this post because I think that gearing the overwhelming majority of EA community building effort towards converting young people to highly-engaged EAs might neglect the value of influence, to the detriment of our ability to get enough people working on difficult problems like AI alignment.

Providing altruistically-inclined young people with opportunities to pursue highly-impactful career paths is great. I’m happy this opportunity was provided to me, and I think this work is incredibly valuable insofar as it attracts more smart, ambitious, and well-intentioned people into the EA community for comparatively little cost. But my impression is that I (alongside other highly-engaged EAs) am somewhat unusual with respect to my willingness to severely constrain the list of potential career paths I might go down, and how much I weigh my social impact relative to other factors like salary and personal enjoyment.

Most people will be unwilling to do this, or might be initially turned off from the seemingly large commitment that comes with being a highly-engaged EA. Does that mean they can’t contribute to the world’s most pressing problems, or only do so via donating? I don’t think so — working directly on pressing problems doesn’t necessarily have to be all or nothing. But my outside impression is that the current outreach approach might neglect the value of influencing the thinking of those who might never wish to become highly involved in the community, and/or those who already have influence over the state of affairs. You know, the people who (by and large) control/will control the levers of power in government, business, academia, thought leadership, journalism, and policy.

To be clear, I don’t think we should deliberately soft sell EA^[1] when communicating externally, but we should be aware that some ideas that sound perfectly reasonable to us might sound off-putting to others. Moreover, we should also be aware that packaging these ideas together as part of the “EA worldview” might risk even just one weird part of the package putting someone off the other ideas (that they might otherwise endorse!) entirely. In this case, it could be better to just strategically push the core thing we want the person/institution to learn about — the importance of cost-effectiveness when allocating charitable resources, or why preventing pandemics is important, for example.

What I mean by influence

My conceptualisation of influence is any outreach that nudges people to do things that are (lower case) ea, rather than trying to guide them directly into the (upper case) EA community. Put more plainly, this does not mean bringing these people into the EA community en masse, nor does it necessarily mean blindly spreading EA memes as far and wide as possible.^[2] All it means is doing things that will cause people outside of the EA community to pursue things the EA community is keen on doing, like preventing pandemics, improving animal welfare, aligning AI, etc.

By this definition, the following things^[3] would count as immediately influential:

• Using an existing relationship with a government official to meet with a UK MP to chat about the importance of crafting legislation that considers future peoples’ interests.
• Hosting a talk at a computing hardware conference on why compute governance could be important for making AI safer.
• Setting up an EA organisation/think-tank in DC that is explicitly focused on influencing US policymakers and politicians.^[4]
• Running for political office.^[5]
• Going on a popular podcast to talk specifically about pandemic prevention.^[6]
• Lobbying congress/other political bodies to increase foreign aid expenditure to low-income countries.

On the contrary, the following things would count as influential over the medium/long term:

• Hosting a talk at a US university on the dangers of gain-of-function research, where one of the student attendees eventually goes on to work in a senior position at the CDC (but not explicitly for “EA” reasons).
• Working with university professors to integrate core EA concepts into undergraduate classes, or creating new classes that cover core ideas in EA entirely.
• Offering funding for PhD students that are tied to conducting research on alternative proteins.

Side note: don’t we already do these things?

You might (reasonably) read the above list and think to yourself “don’t we already do these things?”. I don’t think the ideas I’ve shared above are particularly groundbreaking, nor are they entirely neglected. But my off-the-cuff impression is that the majority of outreach effort in the EA community is focused on other theories of change — find the smartest, most ambitious young people, and get them into the EA community. I would love to see other more targeted and ambitious efforts to influence others where the KPI isn't the number of highly-engaged EAs created.

Value drift, and the tradeoffs of slow growth

I don’t know how to give a precise answer to “how big should EA be” and “how fast should EA grow”, as the tradeoffs involved in thinking about these questions are highly uncertain. While I will briefly touch on these questions, my central claim instead will be that while getting more highly-engaged EAs is great, it might be even better if we focused a bit more of the community’s marginal attention towards influencing others with EA tools, principles, and mental models. To caveat this, I’m more confident in the sign of this claim rather than the magnitude.

I’m sympathetic to the claim that value drift is a serious concern from growing the EA community too quickly. Indeed, value drift could be quite bad. I would hate to see our impact stymied by an erosion of the principles that make EA such a promising avenue for change. But the tradeoffs are real. For example, depending on your views on AI timelines, there are plausible claims to be made that the number of people currently doing useful work on AI safety/governance is far too small, and growing far too slowly given the urgency of the problem.

Beyond influence: getting non-EAs to directly help the community to achieve its goals

Let’s extend this model to the EA movement as a whole. With a total of ~10,000 EAs, and an annual growth rate of 10 to 20 percent (and my best guess is that the bulk of this growth comes from relatively junior EAs), do we really think we can align AI, prevent the next pandemic, transition past factory farming, and eliminate extreme poverty in a reasonable amount of time? I like to lean optimistic, but even I have my limits.

Whether we like it or not, it seems likely to me that we will need people to work on these problems (particularly on the ground level^[7]) who don’t use Bayesian language, never drink Huel, prefer other hobbies more than forecasting, and think “EA” is a gaming company. Again, we might not wish to try growing the EA community with these people. Most will likely never wish to join. But if we can influence these individuals and/or institutions to do things like thinking about scale, neglectedness, and tractability when deciding on what to focus on, that could be extremely valuable — possibly as valuable as adding new highly-engaged EAs to the community. This strategy also partially mitigates some concerns about value erosion within the community by scaling too quickly.

Note: If you find these claims unconvincing, I encourage you to comment with your vision for how we solve these big problems over the next few decades while growing 10 to 20 percent per year and keeping our external influence roughly as it is now. I’m not by any means convinced that I’m right, but I am having a tough time envisioning this theory of change being sufficient for getting the job done. I would love to hear alternative perspectives that I might not have considered.

What makes EA special

One of the most special parts about EA, in my view, is the epistemic culture that has been carefully built in the community. It’s this culture that allows us to take weird ideas seriously; to reason under extreme uncertainty and ambiguity; to strive to be truth seeking; to prioritise, and then try tackling the most important problems head on; and to coordinate a bunch of people with different worldviews under the same umbrella

I think it would be a disservice if we largely kept these ideas to ourselves, or only measured the success of our outreach by the number of highly-engaged EAs we bring into the community. At the end of the day, having more EAs is only instrumentally useful. I don’t care if EAG conferences grow in attendance, or if I have more people to talk about fun thought experiments with; I care about solving AI alignment, ending poverty, and improving animal welfare, whether those who are doing useful work to make progress on these challenges are card-carrying EAs or not. Being insular might allow us to avoid the challenges and risks of outreach, but this strategy can be costly.

To make this argument more concrete, let’s use the numbers from this Twitter thread by Ben Todd. It would be a huge win for the community if we could double the number of people working on AI safety from ~100 to ~200 by 2023. All else equal, I think we should spend a significant portion of our attention and resources on making that happen. But what if we also managed to influence the 100,000 researchers working on capabilities to care even just a little bit more about safety?^[8] Of course, these things aren’t mutually exclusive, and there are spillovers^[9] from one to the other. Moreover, influencing 100,000 capabilities researchers might be quite difficult. But I think it might be risky to hope that we will indirectly influence those outside of the community without explicitly aiming to do so, and I would like to see some analysis on whether or not influence is a cost-effective use of resources.

Conclusion

All in all, I think external influence might be neglected by the EA community. This claim is based on the following seven premises:

1. Most community-building effort currently goes towards creating new, highly-engaged EAs.
2. The vast majority of people who can do things to help further EA goals will not wish to become highly-engaged EAs.
3. You don’t need to be a highly-engaged EA (upper-case) to do ea (lower-case) things.
4. We already do some influence-related things, but we should do more on the margin.
5. Staying both relatively small and insular will make solving our priority problems quite difficult.
6. EA has special epistemic norms, and we should try hard to share them with others who control the levers of power without explicitly trying to bring them into the community.
7. It would be useful to have an idea as to how cost-effective it is to try influencing others relative to creating highly-engaged EAs.

I don’t have the answers to these challenges. My aim with this post is to start a conversation, and to hear from others who have a different view on our need (or lack thereof) to influence others outside of the community. I encourage you to comment below to get the conversation started.

Open questions I’m uncertain about

1. Are there any organisations/projects explicitly focused on influence that I might’ve forgotten, neglected to properly highlight, or am otherwise unaware of? Or am I just completely downplaying how much influence we’re already trying to have?
2. Is there some mechanism that I haven’t considered wherein creating highly engaged EAs could be the best way — or at least a reasonably good way — to influence others?
3. What are some downsides/risks with trying to spread parts of the EA toolkit — e.g., understanding cost-effectiveness, the ITN framework, expanded moral circles, the importance of future generations, the scout mindset — to external stakeholders?
4. It seems obvious that getting people to do robustly good ea things is, well, good. But perhaps it is significantly more costly/difficult to influence people than to create new highly-engaged EAs. If so, how costly is it? What are the bottlenecks, if there are any?
5. If it turns out that (1) influence is important, and (2) the community actually is neglecting influence, what can be done about it?

Acknowledgements

Thank you to Luke Freeman, Grace Adams, Nathan Young, Caleb Parikh, Michael Townsend, Frances Lorenz, Bella Forristal, Trevor Levin, and others who I spoke about this with in person for your thoughts and feedback on this piece.

Appendix 1: The difference between influence and bringing mid-career people into the EA community

I think influencing others and bringing mid-career people into the EA community are different, but I also think these efforts are important and probably interrelated.

My mental model is that it could be highly valuable to influence, say, (1) someone who has 20~ years of experience working for the US Office of Science and Technology Policy, or (2) someone who is an influential public intellectual in the technology space, or (3) some key decision maker in the military, (4) some senior person at the UN, (5) or some member of parliament, etc.

That being said, it would be great if we had some more mid-career people (especially those with excellent managerial skills) to help the community directly scale up projects, or tenured academics switching to valuable research topics. My impression is that there is already some rumbling about this in the community.

Perhaps the people in the former group could end up wanting to work directly at an EA/EA-adjacent org. But if not, I think it is still valuable to influence their worldviews.

1. ^{^}

    H/T to Nathan Young for mentioning this.

2. ^{^}

   That being said, I still think it’s quite good to share robustly good EA ideas through mass media, like moral circle expansion or the importance of donating to charity based on cost-effectiveness. I’m happy to see WWOTF is furthering the longtermism discourse to a much wider audience, for example.

3. ^{^}

   Notably, I’m not making a judgement about the relative pros/cons of these strategies. There is plenty of room to debate whether or not these things would be good in expectation.

4. ^{^}

    I think there are a few organisations that do this, but I’d like to see more.

5. ^{^}

   Of course there was Carrick Flynn’s campaign. I think this was great and I’d be happy to see others make similarly ambitious bets.

6. ^{^}

   See Will MacAskill in August 2022.

7. ^{^}

   We might not need everyone working directly in EA organisations to be entirely value-aligned. Other organisations, both good and harmful, can get people to buy into their goals by doing simple things like paying good salaries. To scale, we will need people to do things like operations, recruiting talent, providing legal counsel, or managing researchers. It might not really matter if these people have joined a virtual program, attended a conference, read The Precipice, or have taken the Giving What We Can Pledge. H/T to Frances Lorenz for pointing this out.

8. ^{^}

   You might (reasonably) think that the marginal AI safety researcher is an order of magnitude more important than influencing capabilities researchers, and/or important decision-makers at AI labs, but I’m not sure this strikes me as obviously true.

9. ^{^}

   It seems reasonable to think that growing the community of people working on AI safety would provide greater credibility to safety arguments, thereby influencing people who are working on capabilities research.