With EA global coming up next week, I thought it timely to bring up an issue with effective altruism (EA) I’ve thought of for some time but struggled to articulate. The issue I have, which I’ll deem ‘the repugnant conclusion about effective altruism’, is that effective altruism, for most people, mainly comes down to building career capital.

EA is commonly defined as the project of finding the best way to help others and putting them into practice. This is the definition formulated on the website for the Centre for Effective Altruism. EA is both a research field, which aims to identify the world’s most pressing problems and the best solutions to them, and a practical community that aims to use those findings to do good.

The standard reference for anything career-related within EA is the website of 80.000 hours, a well-established organisation within EA, and it advices aspiring effective altruists to first explore their options and then build career capital – skills, connections and credentials that make them valuable on the job market. In the third and final stage of your career, according to 80.000 hours, you should switch to a career which focuses on solving any of the world’s most pressing problems, listed here.

However, this framing seems inaccurate to me: if you truly want to maximise positive impact, should you ever stop building career capital? Even if you have the option to work on any of the world’s most pressing problems, surely you should still continue climbing the career ladder. Whoever has the most power is arguably in the best position to do good, and you gain power through career capital.

The per-person distribution of positive impact has thick tails, however you define positive impact: a tiny subset of the population has positive impact orders of magnitude greater than the ordinary person. Beyond luck, what unites most people on the positive tail end is their career capital.

I remember Neel Nanda voicing a similar opinion at last year’s EAG London. During the fireside chat, he said that there was a huge difference in the research output coming from his top one-two mentees and that of remaining mentees – something like 10x, if I remember it correctly. Mind you, MATS has an acceptance rate between 4–7%, and Neel Nanda’s stream is among the most competitive, so the baseline level of competence is pretty high.

The area of AI safety offers a case in point. While it’s impossible measuring impact accurately, the average person working at a frontier AI lab or in a well-known organisation like Redwood, METR or Apollo will likely have a bigger positive impact than an independent researcher. Without adequate mentorship, a big compute budget and good collaborators – which is what frontier labs and organisations provide – your research will be mediocre, and certainly not attract the attention needed to catalyse policy change. But to land a job at any of these companies, you need some insane career capital.

So, unless you’ll make it to the top percentile with respect to career capital – sadly, career capital is relative – it makes little sense to enter the third, directly utility-maximising phase of an EA career, unless it comes at no personal cost whatsoever. The best you can do is perhaps to work on something which isn’t obviously meaningless for your entire career and be a good family member and friend. Or, for a more conservative version: the best you can do is to build some career capital your entire career, so you can become a good role model, in addition to being kind. I suspect this is how the utilitarian calculus checks out for most people.

To conclude, I think the vast majority of the population cannot follow the classic EA career trajectory – they can just be nice and try excelling professionally. In fact, I suspect a large fraction of self-proclaimed effective altruists, at least 70%, also fall into this category of people.

So, beyond the standard moral philosophy (‘be nice’), maybe there’s not that much to EA. The main merit of EA is maybe that it makes being nice cool – it makes finance dudes care deeply about future beings and non-human animals. But besides making people more ethical, what’s left of EA, if anything, is a repugnant conclusion.

This post grew out of thought-provoking conversations with more senior people I met at EA and AI safety conferences. Thank you for sharing and listening.