People have commented relatively little on the digital tech cooperation element of the agreement. There's currently less substance to go on compared with the submarine element. But it strikes me that the impacts here might be more important in the long run, in the extent to which they shape the development of AI. Not going to make a judgement here (although happy to discuss offline).

Some thoughts from a UK civil servant.

Question 1 - how do civil service careers compare to other options e.g. earning to give?

It varies hugely by position. We should expect impact within the civil service to be really unevenly distributed, as it is in other domains. 

If I had to guess, I'd suppose that something like 1% of civil servants in policy-relevant roles (i.e. including policy, analysis, diplomacy etc.) have more personal impact than your benchmark £30k per year to high-impact charities. Within the EA civil service community, I'd expect that could be an order of magnitude higher e.g. 10%.

As in many areas, I expect much of the value is in a small number of very high-impact roles/events. The person who happens to be in the right place at the right time with the right values, skills and contacts. I'd expect the chances of this happening would be increased by: being in a more senior role, working in a policy area with relevance to a high-impact cause area (though not necessarily working on the cause area itself), and being highly competent.

To get an idea of scale let's look at government's role as a 'grant-maker'. In 2018/19 the UK government spent around £850bn. If we assume that 99% of that spending is locked in by politics etc. that would leave £8.5bn to 'play for'. Assuming ~25k policy professionals across government that's £340k influenceable per civil servant. In practice, some civil servants will have a disproportionate influence on this. For example, there are ~200 spending officials in HM Treasury. That's £3.4bn of government budget, or £34m of influenceable budget, per spending official. Of course the tricky thing is figuring out how much value can be created by influencing that grant-making but I think these numbers give a useful sense of scale.

Obviously government does a lot of impactful things besides grant-making: regulation, foreign policy, social policy etc. 

Question 2 - what's the route to impact?

I think improving talent supply is very relevant at junior grades, where it's quite common to find some civil servants 2-5x more effective than others. I think there's less variance in the competence of people in more 'powerful' areas of government and in more senior roles. Given most value is probably in these positions, that could mean it's harder to have impact via better talent. On the other hand, at the top end, seemingly small differences in talent can have very large impacts. So I think the talent route to impact depends on personal fit - those who have an especially strong personal fit can probably have an impact this way.

But I expect the other routes to impact to be the greater comparative advantage of EA civil servants, with competence acting as a multiplier. One key route to impact starts through better 'prioritisation' - which would typically be based on better knowledge, values, skills, experience and contacts. 

For example, impact could come from knowing that, at the margin, government should be focusing more on high-impact, low-probability risks, and pushing to spend more time (whether your own or others) and resources on preventing, monitoring and mitigating them. It would depend on caring enough to go out of your way to advocate for that. It would depend on having the skills to be able to influence others and achieve change. It would depend on the experience of having seen how change works in government. And it would depend on knowing who to talk to, within and outside government.

So in a way, the fundamental thing is knowing and caring about an important problem, which is very correlated with the EA approach. Then you need to rely on your talent to be able to change anything. But then for knowledge specific to the policy context, you'll benefit from network of people you've built around you. EA can definitely help with this network.

Thanks for asking the question!