Thanks for writing this up! I agree it’s really important to clarify that a lot of ‘spending’ is also investing. I think I should have been clearer about this in the podcast, and I worry that if patient longtermism becomes more popular without this being appreciated, it might be negative.
When I think about it to myself, I divide ways to use resources into three categories:
Object-level spending with the aim of impact.
Meta spending that increases the resources and knowledge of aligned people
Investments in financial assets or career capital.
These terms are not ideal because “meta” sounds like I only mean spending on activities that are explicitly aimed at increasing resources (e.g. EA community building or GPR), when many things that look like object level work are also ‘meta’ on this category – for instance, publishing research papers on a new topic might look like directly trying to solve the problem, but often also helps to get more researchers working on the area.
When I was saying 0.5% to 4%, I was only talking about the object-level component. I completely agree that a patient longtermist would likely ‘spend’ more on the meta category.
PS One smaller thing is that it sounds like I might be a bit more skeptical than you of the typical movement building benefits of general object level work. I agree they exist, but I think they are typically smaller than explicit meta work, unless someone is pretty strategic (e.g. the paper Concrete Problems in AI Safety), or in especially good cases. This is partly due to a prior of ‘directly focusing on X leads to more of X’. So, I could image a purely patient longtermist portfolio still being pretty different (though I agree much less different than it first looks).
I think that these categories make some sense to gesture with, and describe reasonable paradigm cases, but actually the lines between the categories are super blurry, such that it’s hard to use them too much as the basis for subsequent analysis.
For instance, as you point out some things that “look like object level work are also ‘meta’”. But some other weird cases:
Some things that sound like “meta” might not be justifiable as long-term investments. For instance running an early-career programme for people to get into field X that seems undersupplied, if the people entering field X won’t do so with a good understanding and motivation that linked to the reasons for it being a priority in the first place.
Since a lot of work will have both object-level and meta-level effects it seems hard to draw a line between them such that we could even start counting “0.5%”
I basically don’t know how to do this for current spending
You’re talking about it in terms of “aims”, which I think is getting at something real, but also gives a lot of weird cases:
I think it could mean that the same activity counts as “object” or “meta” depending on who’s funding it, and what their aims are in doing so
I think lots of time people won’t have a clean idea that one of these is “the aim”; they’ll have a sense that the activity is good (which will connect to impressions about its various effects)
I think “aim of impact” is kind of a weird way of putting it, since almost all activities in the longtermist space only hope to have impact quite indirectly factored through other people’s actions, so it’s hard even in principle to know where the line should be
If Alice uses her savings to do a PhD for career capital reasons, that counts as investment, but if I give her a scholarship for the same reasons, does that count as meta rather than investment?
Overall I think I’d prefer to think about “how good are various opportunities as investments in the longtermist community?”, as well as “how good are various opportunities at making progress towards other proxies-for-good that we’ve identified?”. Activities can score well on either, both, or neither of these, rather than being classed as one type or the other.
Overall I think I’d prefer to think about “how good are various opportunities as investments in the longtermist community?”, as well as “how good are various opportunities at making progress towards other proxies-for-good that we’ve identified?”. Activities can score well on either, both, or neither of these, rather than being classed as one type or the other.
That seems like a good way of putting it, and I think I was mainly thinking of it this way (e.g. I was imagining that an opportunity could further all three categories), though I didn’t make that clear (e.g. should call them ‘goals’ rather than ‘categories’).
Hi Owen,
Thanks for writing this up! I agree it’s really important to clarify that a lot of ‘spending’ is also investing. I think I should have been clearer about this in the podcast, and I worry that if patient longtermism becomes more popular without this being appreciated, it might be negative.
When I think about it to myself, I divide ways to use resources into three categories:
Object-level spending with the aim of impact.
Meta spending that increases the resources and knowledge of aligned people
Investments in financial assets or career capital.
These terms are not ideal because “meta” sounds like I only mean spending on activities that are explicitly aimed at increasing resources (e.g. EA community building or GPR), when many things that look like object level work are also ‘meta’ on this category – for instance, publishing research papers on a new topic might look like directly trying to solve the problem, but often also helps to get more researchers working on the area.
When I was saying 0.5% to 4%, I was only talking about the object-level component. I completely agree that a patient longtermist would likely ‘spend’ more on the meta category.
PS One smaller thing is that it sounds like I might be a bit more skeptical than you of the typical movement building benefits of general object level work. I agree they exist, but I think they are typically smaller than explicit meta work, unless someone is pretty strategic (e.g. the paper Concrete Problems in AI Safety), or in especially good cases. This is partly due to a prior of ‘directly focusing on X leads to more of X’. So, I could image a purely patient longtermist portfolio still being pretty different (though I agree much less different than it first looks).
I think that these categories make some sense to gesture with, and describe reasonable paradigm cases, but actually the lines between the categories are super blurry, such that it’s hard to use them too much as the basis for subsequent analysis.
For instance, as you point out some things that “look like object level work are also ‘meta’”. But some other weird cases:
Some things that sound like “meta” might not be justifiable as long-term investments. For instance running an early-career programme for people to get into field X that seems undersupplied, if the people entering field X won’t do so with a good understanding and motivation that linked to the reasons for it being a priority in the first place.
Since a lot of work will have both object-level and meta-level effects it seems hard to draw a line between them such that we could even start counting “0.5%”
I basically don’t know how to do this for current spending
You’re talking about it in terms of “aims”, which I think is getting at something real, but also gives a lot of weird cases:
I think it could mean that the same activity counts as “object” or “meta” depending on who’s funding it, and what their aims are in doing so
I think lots of time people won’t have a clean idea that one of these is “the aim”; they’ll have a sense that the activity is good (which will connect to impressions about its various effects)
I think “aim of impact” is kind of a weird way of putting it, since almost all activities in the longtermist space only hope to have impact quite indirectly factored through other people’s actions, so it’s hard even in principle to know where the line should be
If Alice uses her savings to do a PhD for career capital reasons, that counts as investment, but if I give her a scholarship for the same reasons, does that count as meta rather than investment?
Overall I think I’d prefer to think about “how good are various opportunities as investments in the longtermist community?”, as well as “how good are various opportunities at making progress towards other proxies-for-good that we’ve identified?”. Activities can score well on either, both, or neither of these, rather than being classed as one type or the other.
That seems like a good way of putting it, and I think I was mainly thinking of it this way (e.g. I was imagining that an opportunity could further all three categories), though I didn’t make that clear (e.g. should call them ‘goals’ rather than ‘categories’).