I love this! Reminds me of Vavilov guarding the seed bank.
One of the worries I’ve heard shared about longtermism is that it’s hard to find interventions that keep on being cost effective when there’s a lot of funding to throw at them—like where the last dollar is as valuable as the first dollar? How do you think that would play out here? To what extent are there diminishing returns?
I think its right that this reminds you of Vavilov! There are a handful of biobanking initiatives that gesture at this kind of reasoning, but I’ve never seen someone apply a longtermist framework to it.
Do you mean dimishing returns in terms of $ per species saved (i.e. it would end up being more expensive once you’d recorded all the low-hanging fruit), or in terms of utility per species saved (i.e. there’s only so many moth genomes you can record before the value of each starts getting diluted).
To answer the first question, I think it would functionally never be an issue. There are so many species that haven’t been described all over the world. Especially very small and uncharismatic species. The fact that we don’t know most of the species on Earth is actually one of the central obstacles in conservation, usually referred to as the Linnean shortfall. This is particularly true when you consider that the evolutionary biologists have come to the conclusion that new species are arising much faster than we expected, especially as climate change causes lots of pressure on species to adapt. So by the time that a biobanking initiative finished cataloging even just all the beetles on earth (beetles account for 25% of animal species!), there would be new beetles to document that didn’t exist when they started.
To answer the second question, there are two possible answers. One is to say that there might be utility in having a large corpus of data itself. Think of todays LLMs—a crucial barrier to training them is the lack of human-produced text, as they’ve essentially already imbibed everything that has ever been written. If you were thinking of a machine learning context in the future (perhaps an AI trying to understand evolution) that needed lots of genetic data (perhaps more than was available on earth at any one time), then large historical genetic databases could be very useful. The second answer to this question would just be the classic longtermist appeal to Very Large Numbers. If you take seriously that 10^28 humans will eventually exist, there is essentially no amount of information you could produce that would start actually saturating that demand. So I don’t think that diminishing returns on this front would really be an issue.
I love this! Reminds me of Vavilov guarding the seed bank.
One of the worries I’ve heard shared about longtermism is that it’s hard to find interventions that keep on being cost effective when there’s a lot of funding to throw at them—like where the last dollar is as valuable as the first dollar? How do you think that would play out here? To what extent are there diminishing returns?
Hey JM,
I think its right that this reminds you of Vavilov! There are a handful of biobanking initiatives that gesture at this kind of reasoning, but I’ve never seen someone apply a longtermist framework to it.
Do you mean dimishing returns in terms of $ per species saved (i.e. it would end up being more expensive once you’d recorded all the low-hanging fruit), or in terms of utility per species saved (i.e. there’s only so many moth genomes you can record before the value of each starts getting diluted).
To answer the first question, I think it would functionally never be an issue. There are so many species that haven’t been described all over the world. Especially very small and uncharismatic species. The fact that we don’t know most of the species on Earth is actually one of the central obstacles in conservation, usually referred to as the Linnean shortfall. This is particularly true when you consider that the evolutionary biologists have come to the conclusion that new species are arising much faster than we expected, especially as climate change causes lots of pressure on species to adapt. So by the time that a biobanking initiative finished cataloging even just all the beetles on earth (beetles account for 25% of animal species!), there would be new beetles to document that didn’t exist when they started.
To answer the second question, there are two possible answers. One is to say that there might be utility in having a large corpus of data itself. Think of todays LLMs—a crucial barrier to training them is the lack of human-produced text, as they’ve essentially already imbibed everything that has ever been written. If you were thinking of a machine learning context in the future (perhaps an AI trying to understand evolution) that needed lots of genetic data (perhaps more than was available on earth at any one time), then large historical genetic databases could be very useful. The second answer to this question would just be the classic longtermist appeal to Very Large Numbers. If you take seriously that 10^28 humans will eventually exist, there is essentially no amount of information you could produce that would start actually saturating that demand. So I don’t think that diminishing returns on this front would really be an issue.
Hopefully that makes sense!