Hi everyone! Below is a summary of the chapter that Maya Eden and I wrote for Essays on Longtermism.
What socially beneficial causes should altruists prioritize if they give equal ethical weight to the welfare of current and future generations? Many have argued that, because human extinction would result in a permanent loss of all future generations, extinction risk mitigation should be the top priority. We call this the long-run argument for extinction risk mitigation.
In Alexandrie and Eden (2025), we evaluate the long-run argument for extinction risk mitigation through the lens of population models. Below we outline what we take to be the two key takeaways of the paper.
Takeaway 1: The long-run argument for extinction risk mitigation relies on the assumption that the global population would partly recover after a non-extinction catastrophe We present a theoretical framework for quantifying the cost-effectiveness of interventions aimed at preventing negative shocks to the size of the global population. A heuristic implied by this framework is that the undiscounted cost-effectiveness of reducing the risk of a negative population shock is proportional to the ratio of future lives lost (in percentage terms) to current lives lost (in percentage terms). We call this the long-term value ratio.
Let us compare the cost-effectiveness of reducing extinction risk with that of reducing the risk of a catastrophe that would kill 10% of the population. Since human extinction would result in a loss of 100% of current lives and 100% of future lives, the long-term value ratio of extinction risk mitigation is 1. Now, a catastrophe that would kill 10% of current lives may result in a loss of less than 10% of future lives (if there is recovery) and more than 10% of future lives (if there is amplification). Therefore, if there is recovery, the long-term value ratio of mitigating the 10%-catastrophe is higher than 1; if there is amplification, it is less than 1. This example shows that (partial) recovery is necessary for extinction risk mitigation to be more cost-effective than other types of catastrophic risk mitigation.
Takeaway 2: Population models suggest that nothing guarantees recovery after a negative population shock We distinguish between two different negative shocks to the size of the global population. A pure population shock is an event that kills some fraction of the world population without having much direct impact on other factors of production (e.g., a pandemic that kills people, but doesn't destroy physical capital). An all-factor shock is an event that destroys all factors of production in equal proportion (e.g., an asteroid that kills the same fraction of people as it destroys physical capital).
Pure population shocks and all-factor shocks have different implications for recovery dynamics after the shock. We consider three models of such fertility dynamics: the social determinants model, the Barro-Becker model, and the Malthusian model. For pure population shocks, only the Malthusian model unambiguously implies that the population would recover after a shock. For all-factor shocks, none of the models imply that such recovery would occur, at least if natural resources are destroyed in the same proportion as population and physical capital. This is because the population models we consider imply that the economic determinants of fertility are invariant to the scale of the economy, which is all that is affected by all-factor shocks.
Conclusion The long-run argument for prioritizing extinction risk mitigation relies on the assumption that the global population would partly recover after a non-extinction catastrophe (Takeaway 1). However, population models suggest that nothing guarantees that such recovery would occur after a non-extinction catastrophe (Takeaway 2). Together, these two takeaways provide a challenge to the long-run argument for prioritizing extinction risk mitigation.
Thanks for the comment! I'll try to explain briefly what I mean:
I take "pain" (in the morally relevant sense) to be defined by something like "a sensation that you dislike when experiencing it". In this terminology, all pain is disliked. Whether a person finds a particular experience painful is subjective, but our moral judgement about pain need not be subjective. In fact, it seems to me and many (most?) people that underserved pain (in the aforementioned sense) is bad regardless of what attitudes other people have about such pain. The fact that some proposition seems true obviously does not imply that it is true. However, it does provide evidence in its favour the proposition. Indeed, I think this is how we get most of our knowledge (e.g., we have reasons to believe that there is an objective physical reality because that seems to be the case). (See phenomenological conservatism for a more general statement of this type of epistemology).
Some moral judgements (e.g., pain is bad) certainly strike me as prima facie objectively true. Moreover, I find most arguments against the objectivity of morality very weak (e.g., argument from disagreement). However, I do think that epistemic worries of various kinds (e.g., epistemic access and evolutionary debunking) count against the objectivity of morality. Overall, I would assign somewhat less than 50% probability to the proposition that morality is objective, but the view should nonetheless be taken very seriously.
Do you know if there are any good writings on axiological-but-not-deontic realism? I have always found this view quite plausible, but I haven't seen much discussion of it.
Hi David! GPI is currently in the process of updating the research agenda. The document you link to above is a draft that was posted on the GPI website earlier this month and is likely to be updated again very soon. It is the most recent agenda to be linked on the website, but it is only a draft at this point.
Feel free to email me at [email protected], if you have further questions!
Minor comment: A few years ago, I looked into estimates of the ratio of animal lives lost to a kilogram of animal protein. One of the facts that were really striking to me was how much the ratio has changed over time in the US for many animal protein products (e.g., dairy cows produce significantly more milk now than they used to). Given how much the ratio has changed over time, it seems likely that there is also a fair bit of heterogeneity between countries. For the OWID charts that display "Animal lives lost per kilogram of product", "Animal lives lost per kilogram of product, including indirect deaths", and "Kilograms of meat produced per animal", it might therefore be worth adding one more sentence in the description clarifying whether the estimates are for the US, the world, or something else. One could of course find this information by clicking on the source, but it may not occur to everyone that the ratios may differ between countries.
Thanks for this writeup – I imagine that it will be very helpful to some other EA groups! When I was the president of EA Stockholm University we organised something similar in several welcome fairs at Stockholm University. Some experiences from this:
The most important thing that I learned was that it is very useful to have some introductory event planned in the next week or even next few days after the fair, so that one can encourage people to attend that event.
For the giving game we used real money, which was laying openly on the table where we were standing. I think this attracted more people to the table than we otherwise would have gotten, so I agree with you that doing this when possible is probably a good idea. If I remember correctly, we had 3000 SEK in 100-bills. Once 30 people had participated, we counted the money in each bucket and then took it out so that the same bills could be used once again. At the end of the game we summed up everything and donated it. Usually we only had something like 50 participants, so it amounted to donating roughly 5000 SEK in total.
For one fair we had charities from different cause areas (global development, animal welfare and something else that I can't remember) instead of charities within a cause area. This allows for more discussion about cause selection, but probably invites people to think less critically (i.e., just pick what feels like a good cause to them).
We tried doing speed giving games at some welcome fairs and not doing it at others. Our experience was that doing speed giving games helped attract people to the table, but made it somewhat harder to engage in depth as a lot of the focus of the conversation often revolved around explaining the rules of the giving game as opposed to explaining what EA is and how to get involved. When I was at EA SU, we eventually decided to not do speed giving games at welcome fairs for this reason, but I think we were uncertain whether this was the right decision and it probably depends a lot on the particulars of the welcome fair in question etc.
Thanks – this is a very important point and I am glad that you raised it! Overall, I think we should be very uncertain about what the long-run population dynamics might be after a catastrophe. I am not sure how much we disagree, but I tried to add some thoughts below (Note: Maya might not necessarily agree with my thoughts on this issue).
As you point out, we write in the paper that Malthusian population dynamics may reemerge in the long run and that evolutionary pressures are one of the main reasons to expect that this might happen. We do not directly argue that such reemergence won’t happen, but in the footnote that follows immediately after that passage you quoted we write:
"In contrast, Arenberg et al. (forthcoming) argue that “empirical facts and models of heritability do not provide reason to conclude that positive population growth is bound to continue via the dynamics of a higher-fertility type making up an ever-increasing share of the global population”.
Here's an attempt to quickly explain Arenberg et al.’s argument: Arenberg et al. point out that we “should not conflate higher fertility within a heterogeneous population with high or above-replacement fertility: it is an empirical question whether future higher-fertility sub-populations will have above replacement fertility”. They then argue that “there is strong historical and global evidence that even higher-fertility groups will eventually trend to near or below replacement fertility”. Drawing on these insights, they introduce a model indicating that “long-term population growth can be negative even with both strong heritability and an above-replacement-fertility sub-population”.
Now, it could reasonably be argued that evolutionary pressures will nonetheless determine fertility rates on 100,000+ year timeframes. However, even if this is correct, such timeframes are only relevant if there is at least a decent probability that humanity will still be around in 100,000 years (conditional on surviving this century). This is not obvious; for instance, if the background probability of human extinction is 1% in each century, then the probability that humanity is still around after 100,000 years is only 0.004%. It is therefore not clear to me that the Malthusian model is correct, so it seems sensible to take other models of fertility seriously.
I just want to second the point that some others have made that it seems more accurate to say only that Harsanyi's result supports utilitarianism (rather than total utilitarianism). Adding the word "total" suggests that the result rules out other version of utilitarianism (e.g. average, critical-level and critical-range utilitarianism), which as you point out is not correct. More generally, I think "utilitarianism" (without the "total") nicely signals that Harsanyi's result concerns fixed-population settings.
It is also worth noting that Harsanyi himself accepted average utilitarianism rather than total utilitarianism in variable-population settings (see the letter exchange between him and Yew-Kwang Ng reported in the appendix of Ng, Y. K. (1983). Some broader issues of social choice. In Contributions to Economic Analysis (Vol. 145, pp. 151-173). Elsevier.).
Hi everyone! Below is a summary of the chapter that Maya Eden and I wrote for Essays on Longtermism.
What socially beneficial causes should altruists prioritize if they give equal ethical weight to the welfare of current and future generations? Many have argued that, because human extinction would result in a permanent loss of all future generations, extinction risk mitigation should be the top priority. We call this the long-run argument for extinction risk mitigation.
In Alexandrie and Eden (2025), we evaluate the long-run argument for extinction risk mitigation through the lens of population models. Below we outline what we take to be the two key takeaways of the paper.
Takeaway 1: The long-run argument for extinction risk mitigation relies on the assumption that the global population would partly recover after a non-extinction catastrophe
We present a theoretical framework for quantifying the cost-effectiveness of interventions aimed at preventing negative shocks to the size of the global population. A heuristic implied by this framework is that the undiscounted cost-effectiveness of reducing the risk of a negative population shock is proportional to the ratio of future lives lost (in percentage terms) to current lives lost (in percentage terms). We call this the long-term value ratio.
Let us compare the cost-effectiveness of reducing extinction risk with that of reducing the risk of a catastrophe that would kill 10% of the population. Since human extinction would result in a loss of 100% of current lives and 100% of future lives, the long-term value ratio of extinction risk mitigation is 1. Now, a catastrophe that would kill 10% of current lives may result in a loss of less than 10% of future lives (if there is recovery) and more than 10% of future lives (if there is amplification). Therefore, if there is recovery, the long-term value ratio of mitigating the 10%-catastrophe is higher than 1; if there is amplification, it is less than 1. This example shows that (partial) recovery is necessary for extinction risk mitigation to be more cost-effective than other types of catastrophic risk mitigation.
Takeaway 2: Population models suggest that nothing guarantees recovery after a negative population shock
We distinguish between two different negative shocks to the size of the global population. A pure population shock is an event that kills some fraction of the world population without having much direct impact on other factors of production (e.g., a pandemic that kills people, but doesn't destroy physical capital). An all-factor shock is an event that destroys all factors of production in equal proportion (e.g., an asteroid that kills the same fraction of people as it destroys physical capital).
Pure population shocks and all-factor shocks have different implications for recovery dynamics after the shock. We consider three models of such fertility dynamics: the social determinants model, the Barro-Becker model, and the Malthusian model. For pure population shocks, only the Malthusian model unambiguously implies that the population would recover after a shock. For all-factor shocks, none of the models imply that such recovery would occur, at least if natural resources are destroyed in the same proportion as population and physical capital. This is because the population models we consider imply that the economic determinants of fertility are invariant to the scale of the economy, which is all that is affected by all-factor shocks.
Conclusion
The long-run argument for prioritizing extinction risk mitigation relies on the assumption that the global population would partly recover after a non-extinction catastrophe (Takeaway 1). However, population models suggest that nothing guarantees that such recovery would occur after a non-extinction catastrophe (Takeaway 2). Together, these two takeaways provide a challenge to the long-run argument for prioritizing extinction risk mitigation.
Thanks for the comment! I'll try to explain briefly what I mean:
I take "pain" (in the morally relevant sense) to be defined by something like "a sensation that you dislike when experiencing it". In this terminology, all pain is disliked. Whether a person finds a particular experience painful is subjective, but our moral judgement about pain need not be subjective. In fact, it seems to me and many (most?) people that underserved pain (in the aforementioned sense) is bad regardless of what attitudes other people have about such pain. The fact that some proposition seems true obviously does not imply that it is true. However, it does provide evidence in its favour the proposition. Indeed, I think this is how we get most of our knowledge (e.g., we have reasons to believe that there is an objective physical reality because that seems to be the case). (See phenomenological conservatism for a more general statement of this type of epistemology).
Some moral judgements (e.g., pain is bad) certainly strike me as prima facie objectively true. Moreover, I find most arguments against the objectivity of morality very weak (e.g., argument from disagreement). However, I do think that epistemic worries of various kinds (e.g., epistemic access and evolutionary debunking) count against the objectivity of morality. Overall, I would assign somewhat less than 50% probability to the proposition that morality is objective, but the view should nonetheless be taken very seriously.
Do you know if there are any good writings on axiological-but-not-deontic realism? I have always found this view quite plausible, but I haven't seen much discussion of it.
Thanks – I appreciate it! As I said before, the draft is very much work in progress so I'm sure these comments will be useful.
Hi David! GPI is currently in the process of updating the research agenda. The document you link to above is a draft that was posted on the GPI website earlier this month and is likely to be updated again very soon. It is the most recent agenda to be linked on the website, but it is only a draft at this point.
Feel free to email me at [email protected], if you have further questions!
Great work – this looks really useful!
Minor comment: A few years ago, I looked into estimates of the ratio of animal lives lost to a kilogram of animal protein. One of the facts that were really striking to me was how much the ratio has changed over time in the US for many animal protein products (e.g., dairy cows produce significantly more milk now than they used to). Given how much the ratio has changed over time, it seems likely that there is also a fair bit of heterogeneity between countries. For the OWID charts that display "Animal lives lost per kilogram of product", "Animal lives lost per kilogram of product, including indirect deaths", and "Kilograms of meat produced per animal", it might therefore be worth adding one more sentence in the description clarifying whether the estimates are for the US, the world, or something else. One could of course find this information by clicking on the source, but it may not occur to everyone that the ratios may differ between countries.
Thanks again for posting this!
Thanks for this writeup – I imagine that it will be very helpful to some other EA groups! When I was the president of EA Stockholm University we organised something similar in several welcome fairs at Stockholm University. Some experiences from this:
Again, thanks for writing this up!
Thanks – this is a very important point and I am glad that you raised it! Overall, I think we should be very uncertain about what the long-run population dynamics might be after a catastrophe. I am not sure how much we disagree, but I tried to add some thoughts below (Note: Maya might not necessarily agree with my thoughts on this issue).
As you point out, we write in the paper that Malthusian population dynamics may reemerge in the long run and that evolutionary pressures are one of the main reasons to expect that this might happen. We do not directly argue that such reemergence won’t happen, but in the footnote that follows immediately after that passage you quoted we write:
Here's an attempt to quickly explain Arenberg et al.’s argument: Arenberg et al. point out that we “should not conflate higher fertility within a heterogeneous population with high or above-replacement fertility: it is an empirical question whether future higher-fertility sub-populations will have above replacement fertility”. They then argue that “there is strong historical and global evidence that even higher-fertility groups will eventually trend to near or below replacement fertility”. Drawing on these insights, they introduce a model indicating that “long-term population growth can be negative even with both strong heritability and an above-replacement-fertility sub-population”.
Now, it could reasonably be argued that evolutionary pressures will nonetheless determine fertility rates on 100,000+ year timeframes. However, even if this is correct, such timeframes are only relevant if there is at least a decent probability that humanity will still be around in 100,000 years (conditional on surviving this century). This is not obvious; for instance, if the background probability of human extinction is 1% in each century, then the probability that humanity is still around after 100,000 years is only 0.004%. It is therefore not clear to me that the Malthusian model is correct, so it seems sensible to take other models of fertility seriously.
Thanks again for engaging with the paper!
I just want to second the point that some others have made that it seems more accurate to say only that Harsanyi's result supports utilitarianism (rather than total utilitarianism). Adding the word "total" suggests that the result rules out other version of utilitarianism (e.g. average, critical-level and critical-range utilitarianism), which as you point out is not correct. More generally, I think "utilitarianism" (without the "total") nicely signals that Harsanyi's result concerns fixed-population settings.
It is also worth noting that Harsanyi himself accepted average utilitarianism rather than total utilitarianism in variable-population settings (see the letter exchange between him and Yew-Kwang Ng reported in the appendix of Ng, Y. K. (1983). Some broader issues of social choice. In Contributions to Economic Analysis (Vol. 145, pp. 151-173). Elsevier.).
Anyway, thanks for this post!
[Edited comment to remove grammatical error]