I didn’t understand the section about EA being too centralized and focused on absolute advantage. Can anyone explain?
EA-in-practice is too centralized, too focused on absolute advantage; the market often does a far better job of providing certain kinds of private (or privatizable) good. However, EA-in-practice likely does a better job of providing certain kinds of public good than do many existing institutions.
It’s interesting to conceive of EA principally as a means of providing public goods which are undersupplied by the market. A slightly deeper critique here is that the market provides a very powerful set of signals which aggregate decentralized knowledge, and help people act on their comparative advantage. EA, by comparison, is relatively centralized, and focused on absolute advantage. That tends to centralize people’s actions, and compounds mistakes. It’s also likely a far weaker resource allocation model, though it does have the advantage of focusing on public goods. I’ve sometimes wondered about a kind of “libertarian EA”, more market-focused, but systematically correcting for well-known failures of the market.↩
I think he’s saying something like:
Imagine the Soviet Union producing one model of dishwashers for all its people. We would expect them to do a terrible job. This is even true if the Soviet Union dedicates teams of specialists, like literally a team of 200 engineers, designers, manufacturers, and marketers. It’s obvious that isn’t enough, and it would be a stilted final product. In the US and market economies, many companies rise and fall making dishwashers. The innovation makes for a much better product.
So, by having an intervention focused on provisioning malaria nets, Give Well is making the same error. I think “absolute advantage” here refers to focusing obtaining and distributing the “most and cheapest nets to the most people”. This is in place of a more flexible solution, less top down, which would be more effective somehow. Maybe even more, the writer is saying that, GiveWell’s entire team, by focusing on “maximizing good” in “one office”, is committing the same sin as the Soviet planners. I think this is close to or saying that having that “one cost effective spreadsheet is bad”.
I don’t think this is a good take.
It’s just wrong. We can’t actually provision hundreds of millions of dollars of mosquito nets a better way, or one that would be close to cost efficient, by say, making a bounty in each of the malaria plagued countries or something.
There are many other sets of interventions which require a lot of coordination and where social, political and other capital is essential to setup. Like, there’s no way a major AI safety org would be produced and handle all of the complexities.
Also, as you point out, almost all EA activity involving spending uses market methods.
The other content “A slightly deeper critique here is that the market provides a very powerful set of signals which aggregate decentralized knowledge, and help people act on their comparative advantage...” Is not good.
To be specific, they recap arguments about markets (e.g. libertarian, “Road to Serfdom” style) in theoretical ways that pattern match to something that is ideological (and sort of basic, like dinner party talk). Mentioning externalities and public goods doesn’t help.
The issue is that this colors their entire essay—thinking the market can solve most problems of the world, to the degree, that a focused, talented team trying to find a new global health problem is stilted or marginal, is a really strong statement.
It’s hard to trust that this same model of the world has anything robust to say about much more complicated and esoteric things, like EA or AI safety.
I agree with your basic interpretation, but I think a different conclusion can be drawn. You’re framing the argument as saying “A completely centralised government is bad, so we should use markets to do these things”. Then you explain why markets alone can’t really provide the things people need.
This is an argument on why not to jump from full centralisation to full libertarian decentralisation. But as in economic policy, there’s vast middle ground. One can, as in social democracy, have that centralised guiding organisation but still rely on the power of markets as a tool.
In practice this means, for example, looking for opportunities like AMF where collective organisation allows for much more efficient allocation of goods than would otherwise be possible; but also giving a substantial part of the resources through GiveDirectly for other goods to be allocated more dynamically by the market. Which is someone “we” (i.e. GiveWell) don’t currently do.
I think he’s saying something like:
Imagine the Soviet Union producing one model of dishwashers for all its people. We would expect them to do a terrible job. This is even true if the Soviet Union dedicates teams of specialists, like literally a team of 200 engineers, designers, manufacturers, and marketers. It’s obvious that isn’t enough, and it would be a stilted final product. In the US and market economies, many companies rise and fall making dishwashers. The innovation makes for a much better product.
So, by having an intervention focused on provisioning malaria nets, Give Well is making the same error. I think “absolute advantage” here refers to focusing obtaining and distributing the “most and cheapest nets to the most people”. This is in place of a more flexible solution, less top down, which would be more effective somehow. Maybe even more, the writer is saying that, GiveWell’s entire team, by focusing on “maximizing good” in “one office”, is committing the same sin as the Soviet planners. I think this is close to or saying that having that “one cost effective spreadsheet is bad”.
I don’t think this is a good take.
It’s just wrong. We can’t actually provision hundreds of millions of dollars of mosquito nets a better way, or one that would be close to cost efficient, by say, making a bounty in each of the malaria plagued countries or something.
There are many other sets of interventions which require a lot of coordination and where social, political and other capital is essential to setup. Like, there’s no way a major AI safety org would be produced and handle all of the complexities.
Also, as you point out, almost all EA activity involving spending uses market methods.
The other content “A slightly deeper critique here is that the market provides a very powerful set of signals which aggregate decentralized knowledge, and help people act on their comparative advantage...” Is not good.
To be specific, they recap arguments about markets (e.g. libertarian, “Road to Serfdom” style) in theoretical ways that pattern match to something that is ideological (and sort of basic, like dinner party talk). Mentioning externalities and public goods doesn’t help.
The issue is that this colors their entire essay—thinking the market can solve most problems of the world, to the degree, that a focused, talented team trying to find a new global health problem is stilted or marginal, is a really strong statement.
It’s hard to trust that this same model of the world has anything robust to say about much more complicated and esoteric things, like EA or AI safety.
I agree with your basic interpretation, but I think a different conclusion can be drawn. You’re framing the argument as saying “A completely centralised government is bad, so we should use markets to do these things”. Then you explain why markets alone can’t really provide the things people need.
This is an argument on why not to jump from full centralisation to full libertarian decentralisation. But as in economic policy, there’s vast middle ground. One can, as in social democracy, have that centralised guiding organisation but still rely on the power of markets as a tool.
In practice this means, for example, looking for opportunities like AMF where collective organisation allows for much more efficient allocation of goods than would otherwise be possible; but also giving a substantial part of the resources through GiveDirectly for other goods to be allocated more dynamically by the market. Which is someone “we” (i.e. GiveWell) don’t currently do.