I agree the incentives matter, and point in the direction you indicate.
However, there's another effect pointing in the opposite direction that also matters and can be as large or bigger: feedback loops.
It's very easy in the nonprofit space to end up doing stuff that doesn't impact the real world. You do things that you hope matter and that sound good to funders, but measurement is hard and funding cycles are annual so feedback is rare.
In contrast, if you have a business, you get rapid feedback from customers, and know immediately if you're getting traction. You can iterate rapidly and quickly get better at something because of rapid feedback.
So in addition to thinking about abstract incentives, think about what kind of product feedback you will get and how important that is. For policy work, maybe it's not that important. For things that look more like auditing, testing, etc where your effectiveness is in significant part transactional, think hard about being for profit.
Source: long engagement with NGO sector on boards and funders, work at private companies.
Sure, but you get feedback for whether it helps customers with their immediate problems. You don't get feedback on whether it helps with AI safety.
It's the direction vs speed thing again. You'll get good at building widgets that sell. You won't get good at AI notkilleveryoneism.
It depends on what you're trying to do, right? Like, if you build a great eval to detect agent autonomy, but nobody adopts it, you haven't accomplished anything. You need to know how to work with AI labs. In that case, selling widgets (your eval) is highly aligned with AI safety.
IME there are an extremely large number of NGOs with passionate people who do not remotely move the needle on whatever problem they are trying to solve. I think it's the modal outcome for a new nonprofit. I'm not 100% sure that the feedback loop aspect is the reason but I think it plays a very substantial role.
Fair enough. There are some for profits where profit and impact are more related than others.
But it's also quite likely your evals are not actually evaluating anything to do with x-risks or s-risks, and so it just feels like it's making progress, but isn't.
I'm assuming here people are trying to prevent AI from killing everyone. If you have other goals, this doesn't apply.
there are an extremely large number of NGOs with passionate people who do not remotely move the needle on whatever problem they are trying to solve. I think it's the modal outcome for a new nonprofit
I'd say this is the same thing for AI for-profits from the perspective of AI notkilleveryoneism. Probably, the modal outcome is slightly increasing the odds AI kills everyone. At least the non-profits the modal outcome is not doing anything, rather than making things worse.
This is one of the most common questions I get when people use my AI safety career advice service.
My short answer is: yes, you can start a for-profit AI safety org, but you should try non-profit first.
It mostly comes down between a trade-off between distorted incentives and amounts of funding.
Non-profits slightly distort your incentives, trying to do things that please funders rather than directly doing what most improves the world.
But usually the funders also want to improve the world, so it’s mostly just “distorting” it to their judgement of what’s impactful, rather than moving away from impact entirely.
For-profits massively distort your incentives.
Your customers usually don’t care about their purchases making the world better. They care about making their lives immediately better.
Yes, for-profits you have a higher potential upside of raising far more funding.
However, broadly speaking, there are two important inputs into how much impact you have:
Going fast in the wrong direction doesn’t matter. This would be having a lot of money to do things, but the things don’t actually help with AI safety.
Especially given how relatively easy it is in this field to think you’re helping but actually making things worse, this should be a very strong consideration.
Of course, conversely, knowing the right direction but not moving anywhere is pointless. And doing AI safety research can be cheap (if it’s mostly theoretical work and you can find good people who are willing to work for little), but often it’s very expensive. So expensive that only a handful of funders can realistically fund your work. So if you’ve tried applying to them and they’ve taken a disliking to you for whatever reason, often going for-profit is the only real option left.
In that case, it can be worth trying for-profit. Just spend a lot of time and energy figuring out ways to make sure that it doesn’t distort your end goal and end up making you a slave to Moloch.
This is a cross-post from my Substack