Local blogger Celeste just released a post claiming ‘earning to give’ should be ~the default path for EAs who are not geniuses or saints. But I think Celeste overlooked a pretty clear reason to sidestep earning to give for now:
The singularity is coming
The first reason why people might want to do ‘direct work’ rather than ‘earning to give’ is that the singularity is coming.
First, there will be lots of local singularities: Jeff Kaufman is trying to give money away sooner because EA-aligned Anthropic employees will soon have several billion free. As we unlock new technologies, society is going to become a lot wealthier. The funding available will grow drastically—and may outstrip our knowledge of how to spend it. I suspect Situational Awareness LP will also have billions available; there are other such bets in the works.
Earning potential is heavy-tailed
The Scott Alexander ‘genius or saints’ argument seems as follows:
Because contributing to intellectual / moral progress is so heavy-tailed, you should only pursue this if you’re extremely good at it.
Because earning potential is so heavy-tailed (see the examples above), I think you should only optimize for ‘earn to give’ if you’re shooting to donate above some order of magnitude over a lifetime, and not otherwise.
Celeste describes a diligent EA earning $55-200k/year in a London tech job. But when you’re you’re a young person considering earning to give, you should be thinking about your lifetime donations / impact, not just what you can accomplish this year. And while this person might be striving to donate ~$1 million over their life, there’s a reason they might not get to:
Human earning potential will be driven to 0
Celeste’s diligent EA is only going to get to donate from her tech job for ~2-6 years, I think, before that job is replaced by cheaper AI labor.
Are the <$100-300k across ~2-6 years worth it, when she could’ve been a) making bigger bets, b) steering the singularity, c) preparing for the singularity? ‘Preparing for the singularity’ is a whole post of its own, but here are some things an EA might want to have during: good judgement, leadership / direction / management / PI ability, being a well-integrated person with a coherent ethical system, deep familiarity with frontier AI systems, stakeholder coordination ability, Askell-like ability to build robust, stable, compassionate longterm regimes, systems thinking, equity.
If her tech job gives her those things, I think it might be worth doing. But if it’s just getting a wage from doing work that will soon be automated, I’m skeptical.
What I suggest to the diligent EA who wants to help end human suffering
Help us unlock more vitae per person by focusing on inputs that are complementary to artificial intelligence. Remember, AI will be able to do a lot of scientific prototyping; we’ll have humanoid robots for incredibly cheap labor. What won’t they do by default, and how might you supply it?
Incidentally, ‘flipping non-EA jobs into EA jobs’ and ‘creating EA jobs’ both seem much more impactful than ‘taking EA jobs’. That could be e.g. taking an academic position that otherwise wouldn’t have been doing much and using it to do awesome research / outreach that others can build on, or starting an EA-aligned org with funding from non-EA sources, like VCs.
There is lots to do to increase vitae per person.
The billions of anthropic argument is convincing. The working in AI safety thing is convincing.
The problem is all the prior presuppose a singularity coming soon. I am not convinced. Furthermore, the dilligent techie could fund careers in AI safety. Finance earns a lot of money. This is especially important if she does not consider effective cause areas interesting. But I mean it's inherently pascal-wager esque, so it kind of matters automatically because of how impactful it will be and how neglected it is.
> Incidentally, ‘flipping non-EA jobs into EA jobs’ and ‘creating EA jobs’ both seem much more impactful than ‘taking EA jobs’. That could be e.g. taking an academic position that otherwise wouldn’t have been doing much and using it to do awesome research / outreach that others can build on, or starting an EA-aligned org with funding from non-EA sources, like VCs.
That's a super good argument tbh. Embarrassed I didn't think of it. I will consider this if I were to do an ML PhD. Which I'm not sure of yet.
> Celeste describes a diligent EA earning $55-200k/year in a London tech job. But when you’re you’re a young person considering earning to give, you should be thinking about your lifetime donations / impact, not just what you can accomplish this year. And while this person might be striving to donate $O(10^6) over their life, there’s a reason they might not get to:
I was doing this by the way. The americain cannot comprehend salary capping at $200k. (in belgium most seniors never reach $100k)
> I think Celeste is very smart and can do a lot if she sets her mind to it.
thank you, I hope so! I might just by accident get into AI safety. I have very similar MLish interests to you of just wanting to know how everything works (learning neuralese as you put it) and just through following curiosity I may end up in technical AI safety even when not convinced of existential risk.
nitpick: I am confused at your usage of asymptotic notation in this post. How is big O appropriate here at all when the constant is kind of everything?
It seems to me that the two are never really mutually exclusive in the first place; a lot of people who E2G end up working in EA and vice versa. I can even imagine they are correlated vectors - E2G means you're more familiar with EA arguments means you're more likely to end up working in AI safety.