Thanks for the response!
I guess the fact that no country in history has gotten rich while being agrarian gives me a very strong prior against it. And there are clear reasons why; agricultural goods are commodities that are extremely cheap, so even having an advantage in them, you can only have a slim advantage.
My impression is that Argentina became a very rich country, by the standards of today, largely as an agricultural exporter? And just because a product is cheap doesn't mean you can't have a big advantage in them if it is a very scaled business; according to OWID, US agricultural productivity is over 100x higher than in Liberia in dollar term; I don't have the bushels/farmer figures to hand but I suspect there are also orders of magnitude difference.
Theres no assumption that services are not valuable, just that productivity growth (as measured by revenue per worker growth) is much slower.
I'm not assuming it, I'm offering it as an explanation. The 'nontradable' explanation doesn't really make sense to me, because it doesn't explain why people would choose to work in a less productive sector (agriculture or services), where wages are presumably lower, instead of manufacturing. Unless you think that manufacturing has to be combined with labour suppression, where wages will be held lower than productivity, in order to facilitate more investment? I do get the impression that is part of the story behind east asian growth, and it does make sense it would be easier in manufacturing than services.
Maybe if/how his thinking about AI governance has changed over the last year?
Thanks for sharing this, I found it very interesting.
I was curious about the sectoral transformation. Presumably we will always need some people working in agriculture. A lot of this specialisation has occurred between rural and urban areas, but might it not also make sense for some entire countries to focus on agriculture? They could focus their education systems, regulations and so on the industry, which might improve efficiency, rather than having smaller numbers of people in more countries doing agriculture. If this is the case then we could see some countries getting rich entirely off agriculture - just like there are wealthy farmers in the US, Australia, etc. If so pushing for sectoral transformation could be a mistake if some countries really have a comparative advantage in agriculture.
This also connects to your points about agricultural productivity. My impression was that many third world countries have chronic under-utilisation of labour; you literally just have a lot of working age men hanging around doing nothing all day. This is labour slack is implicit in the model for why GiveDirectly might boost economic activity that they described on the 80k podcast. If so, productivity growth that incentivized people to stay in agriculture could be good, especially if it replaced unemployment, even though it would retard the sectoral transformation.
Finally, I was interested in the negative effects of the services transition. Could this be because poor regulatory setup means many of the 'service' jobs are essentially rent-seeking rather than providing socially valuable services? e.g. an increase in lawyers or bureaucrats primarily creating more work for other people.
If it's a material risk risk, companies already have to disclose it as a risk factor. What makes this unusual is 1) it requires a very costly data gathering exercise 2) it opens companies up to very large legal risk about their precise methodology and 3) it is required of all companies, even if the risk is not material to them.
As an example, at least one of the economists in the poll you linked thought it would help investors make decisions, but was still a bad idea:
Probably the costs of a mandate exceed the benefits. The uncertainty is for firms where the impact is small and indirect. Climate is a risk that might be hidden.
It might be useful to consider an analogy with the opposite political valence. Many companies in the US employ, or deal with other companies who employ, immigrants, including illegal immigrants. This causes political risk; there may be changes to immigration rules, or an increase in enforcement and deportations, that could affect their operations. At the moment, companies for which this is material issue disclose it, generally using relatively high level language, and companies for whom it is not material do not. The equivalent of this SEC move would be if all companies had to report the exact number of immigrants they employed, broken down by visa category, national origin, and illegal status, for themselves, their contractors, their suppliers and their customers. This would help investors make decisions! But it would be extremely costly, and the motivation would clearly be political and an abuse of the SEC's remit.
Some previous discussion here.
Would you utter the phrase "animal bodies are held together by flesh instead of skeletons?"
Seems like weird phrasing but ultimately true. Bones give structure but they're not sticky - if you took away all the flesh (muscle, skin, ligaments etc.) the bones would just fall apart, but if you took away all the bones it would be very floppy but still would clump together. A boneless chicken thigh doesn't fall apart, but a skeleton from hundreds of years ago, where only the bones remain, is not held together.
Makes sense to me, thanks for sharing! It seems pretty plausible that the tighter remit is a good choice, both operationally and in terms of communication with donors. And I appreciate the clear examples of what falls inside and out.
One question, not intended as a criticism: you point out that EAIF would no longer function as a catch-all donor-of-last resort to random projects, which makes sense. But I do come across people working on such projects, and it does feel like there should be somewhere I can refer them, where they will be evaluated by cause-agnostic generalist EA evaluators (even if their prior probability of beating AI/GHW/Animals was low). Are you aware of such a venue?
I realize I am saying 'someone should', but it might be interesting to go through this list point-by-point in retrospective and see how it applied to the OpenAI situation.
Thanks very much for sharing, very interesting work.
I was very pleased to see the pre-registration, which I think significantly improves the credibility of the results. Would you be able to help me understand some of the endpoints? I was hoping to see a straightforward 1:1 correspondence between the pre-registration and the paper but found it a bit confusing; it might be helpful to publish a table showing how to compare the two.
For example, the pre-registration lists 'aspirations' as one of the primary endpoints:
The main outcomes of interest are economic status (income, consumption, assets, and food security), anthropometrics (for children in the household between 2-12 years of age), time use (work, education, leisure, community involvement), risk-taking (especially, migrating and starting businesses), gender relations (especially female empowerment), aspirations and mental health.
This is defined in the pre-analysis plan as
the Anderson weighted index of (i) the average of responses to question 1a on aspirations for kids educational attainment, (ii) question 2c on level of assets the respondent wants to achieve in 10 years, and (ii) question 3b on the level of income they want to achieve after 10 years.
but I was not able to find it in the paper. Similarly the plan describes a 'crime index', based on responses to the village elder survey, and a 'social integration index', as other outcomes that will be studied, but I couldn't find these in the paper.
Thanks for writing this. I think people might find it helpful if you made some estimate of the QALY impact of microplastics, and maybe described what you think are some of the more tractable interventions to affect it.