We are assessing the case for population growth as a priority. Here is the Google Docs version if you prefer to read it as a document. Charlotte Siegmann and Luis Mota
Recently, population growth as a cause area has been receiving more attention (MacAskill, 2022, PWI, Jones, 2022a, Bricker and Ibbitson, 2019). We outline and discuss arguments making that case, and argue that population growth falls short of being a top cause area under the longtermism paradigm. But our main contribution may be to just outline and explain all considerations and arguments.
We consider three value propositions of population growth, and argue that:
- Long-run population size is likely determined by factors apart from biological population growth rates. More
- Biological reproduction will be replaced if the future is to be large, which should nullify the effects of population growth in timescales larger than a thousand years.
- Assuming biological reproduction continues, long-run population size is only influenced by current population growth in particular scenarios.
- Population size may impact economic growth, but its long-run effects are comparatively small. More
- Population is one of the current drivers of economic growth.
- We argue that the probability and long-term harms of economic stagnation, such as moral regress, are overstated.
- We argue that the effects of economic growth on extinction risk are small.
- Population size has negligible effects on humanity’s resilience to catastrophes. More
We think that the most compelling case for intervening on population growth comes from its effects on economic growth. Overall, while we think that increasing population growth rates is positive, the scale of its benefits appears to be orders of magnitude smaller than those of top cause areas.
Today is the Day of Eight Billion, a day projected by the UN to be roughly the one in which the world human population reaches 8 billion people. The last time the world population increased by a round billion was 11 years ago, and the UN projection suggests that the world population will reach the 9 billion mark in 2037, and the 10 billion mark in the mid-2050s. This projection never gets to 11 billion, as it peaks at 10.4 billion during the 2080s and slowly declines throughout the last decades of this century. Other population projections predict an even earlier peak in the human population, which would happen before there are 10 billion of us (Lutz et al. 2018, Vollset et al. 2020).
There is a reasonable amount of uncertainty about whether the peak will happen by the end of this century. In the UN projection, with a 5% probability, the population size will be larger than 12 billion by 2100. But fertility trends indicate that a peak will eventually happen. The world fertility rate has been declining since the 1960s and is expected to decline to the replacement level of 2.1 children per woman by the mid-century. Once it is below that point, growth would only be due to population momentum. This transition to below replacement levels of fertility is already on its way, as more than half of the world population lives in a country where fertility is already below the replacement rate.
The expectation of a population decline raises the question: should population growth be a priority for actors wanting to do the most good? For one, a higher population might be desirable on its own. For example, according to the total view in population ethics, creating lives with positive wellbeing is good, all else equal. Moreover, population growth may also affect other variables of moral importance, such as economic growth (Jones, 2022b) and social and political norms (MacAskill 2022, Bicker and Ibbitson 2020). At first glance, increasing population growth may have an unusually lasting impact, as the children one causes to exist will eventually come to have children of their own.
The aim of this report is to evaluate the case for prioritising population growth as a priority. We consider the effects of a population increase on three variables: the size of future population, economic growth, and civilizational resilience to catastrophes. For each of these effects, we present a broad picture of the key arguments for population growth and the counterarguments, as well as our conclusion about the relative strengths of these arguments.
Four key assumptions behind our analysis are:
- The longtermism paradigm (any action that is among the best positively impacts the world far away in the future),
- The total view in population ethics,
- Welfarism, and
- Fertility increases or reductions in child mortality are the main channels through which we can increase population size.
It is not obvious that these assumptions hold in practice. However, we believe that they constitute a reasonable basis for the present analysis, and dropping some of them likely weakens the case for population growth. We conclude that the case for population growth as we understand it is not strong enough to make it a priority, that is, to justify considering it a top career path or a major funding opportunity for actors aiming to do the most good.
1. Population increases are unlikely to persist in the long run
Caveat: this section is inherently more speculative than the others, as it deals with very long-run trends, and radically different future possibilities make us much less sure about what will happen then.
If a larger human population is better, and short-run increases in population growth translate into a higher long-run population size, then increasing population growth now could lead to significantly higher total wellbeing in the future. In principle, this effect could be comparable in impact to existential risk reduction: increasing the expected size of the entire future population by a fixed proportion is equivalent to increasing our probability of survival by that same proportion. Thus, if existential risk reduction is one of the most impactful interventions, changing population sizes may be similarly impactful. Furthermore, population increases have been responsible for most of the growth in total human wellbeing (as measured by the log of GDP) in the last century (Klenow et al. 2022), which sets a precedent for this type of effect.
There are a few different ways in which these long-run population increases could take place, and the graphs below illustrate some of these possibilities. In each case, the red line denotes the business-as-usual scenario, and the blue line denotes a possible population trajectory after an intervention. In scenarios (a) and (b), population eventually declines to zero without intervention, as an extrapolation of the UN projection would suggest. An intervention in these cases could either slow down the decline or change the long-run trajectory from a decline to an increase. Alternatively, long-run population may not decline without bound without intervention, as it could increase without bound (scenario c) or stabilise (scenario d). Population interventions could hence increase the long-run rate of growth or increase the level at which it will stabilise.
We argue that the trajectory with and without the intervention converges in the long run. First, we make the case that the futures with artificial people are much bigger in expectation, and that population sizes in those worlds are unaffected by human population size. Second, we contend that even in cases where we do not create artificial people, there are several channels through which the effects of increasing population growth can fade away.
Artificial people would ultimately change population growth dynamics
The creation of artificial people might be feasible through technology such as mind uploading, which would enable us to simulate human people and create people living in virtual reality. The creation of artificial people may allow a substantially larger number of humans to inhabit a given area, experience faster-paced virtual worlds, expand to regions of space inhospitable to biological humans, and have a wide range of other benefits. The advent of digital people would make space settlement more likely, and plausibly the vast majority of people in the long-run future will be artificial.
The transition from biological to artificial people would likely bring about substantial changes in key considerations underlying a reproduction/replication decision, such as a substantial reduction in cost. Even if artificial people can reproduce sexually, and their fertility preferences are similar to those of biological people at the time of their creation, higher population growth could be achieved by simply creating more simulations. Moreover, artificial people could be immortal, in which case additional people would only increase population. Hence, assuming that creating artificial people will eventually be possible, it seems unlikely that a higher biological population would have any significant direct effect on the population size of artificial beings. We put >50% credence that such technology will be developed in the next thousand years (more reasoning in this guesstimate model). It follows that we believe biological population growth interventions should not have effects lasting more than a few centuries.
One has to ask about the moral status of artificial people, and when we would have such technology. We believe that sentience is sufficient for moral patienthood status. Moreover, if beings are conscious, they are likely also sentient. Most consciousness researchers believe that machines can have consciousness, and most consciousness theories allow for this possibility.
An overwhelmingly high probability that artificial beings will exist is not necessary for this argument. Even if they are only 10% likely to ever be developed, the sheer number of potential people in such worlds would make them have a much larger value potential than those where artificial minds do not come into existence. Hence, expected value considerations would lead to the conclusion that we should focus our actions on the worlds where mind uploading is possible.
Although we have focused on the case of artificial people when making this argument, it is worth noting that this line of reasoning is more general. Higher biological population growth would not be the best way to promote long-run increases in population size, as long as the following two conditions are satisfied:
- Future worlds with the highest expected population size are those where we develop technologies that enable us not to depend on biological reproduction for increasing population size; and
- The population trajectory after the development of such technologies is independent of the trajectory before their development.
While the creation of artificial people is a key example of a change enabled by future technology, it is not the only one. For example, advanced AI may manage biological human population size in a way that could also satisfy the two conditions above. A reason for believing that these conditions will hold is that our future will be drastically different from the present, and brings with it the potential for increasing population growth in ways that are far more efficient than traditional biological reproduction.
Short-run population changes only persist under particular circumstances
Even if we assume that biological reproduction will never cease to be the main form of reproduction, there are good reasons to expect that changes to the current population growth trajectory will not have lasting effects.
First, regarding the possibility of a permanent fertility decline (as in scenarios a and b in the figure above), note that there are good reasons to expect fertility rates to rebound and reach a long-run above-replacement level. For a rebound to happen, we would only need a single human group satisfying the following two conditions: long-run above-replacement fertility, and a high enough “retention rate”, that is, a large enough fraction of the descendants of this group continues to belong to the group (Arenberg et al. 2021). There are several ways by which we could get at least one group satisfying these conditions:
- Evolutionary pressure leading to high-fertility genes being selected for in human populations, eventually leading to above replacement fertility and total prevalence of high-fertility genes in the population;
- One country successfully implementing a fertility policy, and having low enough migration (“retention”); or
- A culture within a country developing high fertility and high retention rates.
Thus, for long-run population decline to occur, all of these mechanisms would have to fail, which appears unlikely.
On the other hand, we have observed systematic fertility decline throughout the world. This may suggest that fertility decline continues because of systematic drivers, such as wealth, simply make people much less inclined to have children, and that there is not much that individual actors can do to change that. It follows from this reasoning that population decline is much more likely to continue, but also that it should be much harder to prevent it.
Second, population cannot keep growing indefinitely. We assume in this section that biological humans will never be replaced by another species. As an upper bound, there are only so many atoms in the universe reachable by biological humans, and only so many ways in which we can arrange those atoms. Hence, given enough time, we would reach some steady state where future societies are using all the resources available to them. In the figure below, denotes the steady state population size when society is using the entirety of resources available, denotes the time at which we will reach this population size, and denotes the time at which the universe would become inhospitable for humans. The area between the blue and the red curves denotes the impact of getting to this upper bound in population more quickly. If we expect that time at the upper bound to be much greater than the time it takes for us to get there (), then the gains of increasing current growth rates are a small share of the total value.
Perhaps the condition above does not hold, and perhaps the universe will become inhabitable to humans before we ever reach a steady state (), as illustrated below. In this case, increasing population growth could have massive effects on population size.
Given that the universe will remain inhabitable for a long time, and current economic growth rates would allow us to reach this steady state relatively quickly, this scenario tentatively strikes us as more unlikely than the previous one. Furthermore, expected value considerations point to worlds where we reach this upper bound being more valuable. Without artificial people, it is also possible that humans will choose not to settle the stars, or will only do so to a limited extent, implying that the steady state could happen much sooner than otherwise.
One could think that depends on the trajectory of human population growth. For example, perhaps a given generation decides to act as a cartel, and ensure that the equilibrium population size will henceforth maximise their average wellbeing. If this equilibrium population size is below the size that maximises total wellbeing, then increasing population size at the time that the cartel agreement is made might lead to permanently higher population size. We think that this argument presents a plausible way in which short-run population growth changes could have large long-run implications. We have not engaged with this argument enough to properly assess its merits, but we note here that there are important considerations which might make such a scenario unlikely. More discussion of this argument here .
2. Population increases have a small effect on economic growth
In (semi-)endogenous growth models, a widely used class of economic growth models, population growth is the fundamental driver of long-run economic growth. These models hint that a smaller working-age population in one period reduces GDP per capita—a proxy for average human wellbeing—in all future periods. When evaluating a standard growth model with a negative population growth rate, Jones (2022a) concludes that the GDP per capita growth rate would eventually converge to zero, in contrast to the usual positive long-run economic growth rate.
There are many arguments as to why higher economic growth would be desirable from a longtermist perspective. Influencing population growth has been justified by three key arguments:
- A few centuries of economic stagnation could lead to an existential catastrophe (MacAskill 2022, chapter 7—co-authored with Aschenbrenner);
- Economic stagnation could lead to the erosion of societal values (MacAskill 2022, chapter 7); and
- Increasing standards of living make people care more about preventing extinction (Aschenbrenner, 2020 and Trammell, 2021; see also Aschenbrenner’s discussion of the topic).
In this section, we discuss the link between population and economic growth, and examine the three arguments presented above.
Population growth constitutes a small share of current economic growth
While population growth is the underlying cause of long-run human-driven economic growth, other factors also contribute to current economic growth. Jones (2022b) decomposes the constituents of growth in the United States for the past 60 years, as illustrated in the figure below.
Source: Jones (2022b)
The US GDP per capita has grown, on average, by 2% per year over the past 60 years. The pie chart on the left breaks this growth rate into three components:
- Approximately 0.5 percentage points are attributed to increases in human capital per person (roughly, increases in education);
- Changes to the employment-population ratio account for 0.5 p.p.; and
- Total factor productivity growth (roughly, technological improvements and misallocation reductions) account for 1.3 p.p.
Total factor productivity growth can be broken down into three subcomponents:
- Increases in research intensity (fraction of population working on R&D) account for 0.7 p.p. of total factor productivity growth;
- Decreases in the misallocation of capital (in how much inefficiency there is in the allocation of the labour force) were responsible for 0.3 p.p.; and
- Population growth was responsible for 0.3 p.p.
These estimates suggest that population growth was ultimately only responsible for 15% of economic growth occurring in the period. Population growth is seen as the fundamental driver of economic growth in the long run because the other variables cannot grow indefinitely. For example, the employment-population ratio can be at most 1. In the past 60 years, there was enough room for growth in misallocation reduction, research intensity, human capital, and so on, for these aspects to account for the vast majority of US growth.
When considering the effect on the following centuries, increasing fertility rates does not stand out among growth-promoting interventions. The other factors will still play a major role for at least several decades, as there is still plenty of room for them to increase. Consider research intensity (the main factor responsible for growth in the figure above): today, only 1% of the US population works on research. There is no good reason to expect that this fraction could not grow to 6% of the population, and we would not be too surprised if it eventually went up to 15%. Based on Jones’s (2022b) estimate of a 2% growth in research intensity per year, it would take 90 years until it grew to 6% of the labour force. And these are figures for the US, which means there is more room for growth in research efforts in most other countries, as they have a much lower proportion of researchers in 2022.
Furthermore, on a scale of millennia, economic growth will likely be entirely driven by advanced artificial intelligence or by artificial people. Hence, to our mind, the most compelling cases for biological population growth to be a key factor behind economic growth are those where this process takes several centuries to unfold. In these scenarios, biological population growth still has to be the best candidate for increasing growth over the next centuries.
Moreover, by increasing population, one may also bring the creation of artificial people forward in time as there is more research in each period.
The risk and long-term consequences of stagnation are overstated
In What We Owe the Future (chapter 7), MacAskill and Aschenbrenner (henceforth MA) argue that the possibility that economic growth slows down considerably or stagnates might be particularly dangerous for humanity. The expectation of a significant growth slowdown is justified by the following arguments:
- The population growth rate, a driver of economic growth (as discussed above), is declining, and population will start to decline by the end of the century.
- Other drivers of economic growth cannot keep on generating growth at the current rate for more than a couple of centuries.
- Ideas are getting harder to find, that is, the rate at which a single researcher is discovering new ideas has decreased over time.
Furthermore, they mention that if growth does slow down a lot or completely stops, there are two reasons to believe that stagnation would be harmful:
- If we stagnate at an unsustainable level of technological development, this stagnation period could be particularly likely to cause existential risks. As an illustration, if technological development were to have stopped in the 1920s, we would have no way to control climate change. We would also eventually run out of fossil fuels to burn, eventually leading to collapse.
- The values that humanity would have after the stagnation is over are different from those that we have now, and are more likely than not to be worse than our current values. For instance, because there is a positive correlation between economic growth and better values—see Friedman (2005).
We agree with them about the importance of the harms caused by stagnation, and think that its potential causes mentioned above are worthy of serious consideration. However, as we argue below, we believe that their arguments overstate the plausibility and size of the effects of stagnation.
We are not the first to scrutinise this chapter's arguments. This writeup provides additional reasons for scepticism about their claims. In particular, the argument that technology to drastically increase economic growth might already be available strikes us as particularly strong.
The possibility of stagnation
Regarding the possibility of stagnation happening, we are more optimistic about the timing of factors causing population in the long-run to rebound from a decline. In particular, there is evidence that evolutionary pressure is already leading to increases in fertility (a potential pathway for population increases not considered by MA). We expect that this force alone will likely be able to single-handedly prevent population decline lasting more than a couple of centuries.
Furthermore, it is uncertain whether ideas in the future will be getting harder to find. Bloom et al. (2020), the main source cited for these claims, present figures showing central trends in productivity for various sectors over time. One of us (Charlotte) analysed the data used for this paper and found that, while ideas got harder to find in the past 60 years, the trends and fluctuations in the data does not allow us to confidently extrapolate this trend into the future.
Lastly, and more importantly, even if we take all of MA’s arguments for granted, there are still good reasons for scepticism that attempting to increase population growth would be the best way to bring about high growth. Population growth currently only accounts for 0.3 p.p. of economic growth. Other interventions, such as increasing the proportion of researchers, may be more cost-effective.
The long-term harms of stagnation
Turning to the reasons MA provided as to why stagnation might be bad, we think that a few caveats are in order regarding the existential risks arising from the level of technological development. Current levels of technological development may be unsustainable. The development of synthetic biology today and in the next decades poses a dual-use risk. If the cost of renewable energy had stalled in 2009, a transition to clean energy would have been much less likely. However, at some levels of technological development, stagnating for several centuries would not generate particularly high levels of existential risk.
Second, if a level of technological development is unsustainable, then society also has the option to choose to ban unsustainable technologies. If we were stuck with the technological level of the 1920s, we could collectively abstain from using fossil fuels, and the aforementioned risks would disappear. Although banishment might seem particularly hard in the case of fossil fuels, given how large their role in the 1920s world was, it would be a more attractive option for less important technologies, or less widespread technologies. Such a possibility has precedents: as MA themselves mention, we have opted not to use cloning technology to clone human beings, a technology that could be extremely beneficial to economic growth.
Finally, actors concerned with economic stagnation should also consider options that would improve civilisational prospects conditional on stagnation. Two examples are investing more resources to increase their influence in a stagnant world, or trying to lock in some values.
Increasing economic growth is suboptimal for changing societal preferences
Aschenbrenner (2020) models the relationship between existential risk and economic growth. The key idea behind the results of this model is that, as people become wealthier, their lives become more valuable, and extra money can do ever less to improve their wellbeing. Hence, people start spending their extra wealth to increase their life expectancy over increasing consumption. While at the individual level this implies investing more money in health and safety (Jones, 2016), at the societal level this would include spending more to reduce the risk of existential catastrophes. Under some plausible assumptions about parameter values, speed-ups in economic growth would make people invest in safety enough to reduce the total risk of an existential catastrophe. A temporary increase in the population growth rate would be sufficient to speed growth up in this way, and hence cause existential risk reduction.
We believe this is a useful consideration for the case that increasing population reduces existential risks. However, it does not follow that increasing population size is among the best ways to reduce existential risks. There are seemingly more direct and high-leverage opportunities to increase the number of existential risk safety researchers. For instance, as the paper’s numerical estimates suggest, increasing the patience of important actors by a small amount could lead to an effect on existential risk reduction comparable to a substantial increase in economic growth. Alternatively, private actors can (i) increase awareness of the risks facing humanity or (ii) subsidise existential risk safety research, which could increase the proportion of researchers working on safety.
Moreover, the consideration above regarding risk from changes in technological development also applies here. If one believes that risks arising from increases in technological development are higher when economic growth occurs at a faster pace, then accelerating growth would bring about negative consequences that could be stronger than the gains.
3. Unclear effect of population size on existential risks
We consider the effect of existential risks that are not mediated by effects on economic growth. We tentatively argue that changes to population size do not have a clear effect on human resilience to existential catastrophes.
An implication of Rodriguez’s analysis of civilisational collapse is that a higher population can increase our resilience to disasters. If a nuclear war breaks out, the larger the population at the time, (generally) the higher the probability that enough people will survive the catastrophe, enabling a civilisational rebound. However, it seems implausible that this can make population increases cost-effective. We have to compare population growth increases with interventions that directly target ensuring survival in such scenarios, which would likely generate an equivalent benefit for a fraction of the cost.
When it comes to the direct effects of population growth on other existential risks, the direction of the effect is unclear. For some risks concentrated in the upcoming decades, fertility changes would most likely have little effect, because most of the effect would come from additional adults. For global catastrophic biological risks, some factors—higher social contact between humans and higher risk of cross-species disease transmission—increase the risk, and while others—larger expected immune population—decrease it. When it comes to other risks and risk factors, such as great power conflict and authoritarianism, the effects have an unclear sign and high variance. We are unaware of any attempts at modelling or understanding these dynamics.
4. The main potential harms of population growth accrue in the short-run
So far, we discussed arguments pointing to a positive value of an additional person. One might think such an assumption is not warranted, as there are also negative consequences for bringing people into existence, most importantly climate change and the consumption of animal products. While we agree with these arguments, we do not expect these negative effects to last long enough to make the long-term value of additional lives negative. This assumes population changes have positive long-run effects, which we have been sceptical about in the previous sections.
Concerning climate change, a greater population could reduce the moral value of life on earth because the pollution emitted by the additional people worsens climate change. However, given the current trends in carbon emissions and policies, we ultimately find this argument unconvincing. Kuruc et al. (in progress) estimate that even a reasonably large change in the UN population trajectory would likely have small negative effects on climate. Furthermore, when Kuruc et al. consider the benefits of a larger population in terms of climate change reduction R&D, and find that these benefits seem to outweigh the costs.
The negative effects of animal consumption alone may arguably make the net effect of additional human lives negative. However, animal consumption most likely will only continue for a couple of centuries, while the benefits of population growth that we consider would purportedly last much longer. For more background on the duration of factory farming, we refer to Dullaghan and Zhang’s discussion of cultivated meat forecasts, Brauner and Grosse-Holz’s writeup on the long-run value of extinction risk, and this AMA by Lewis Bollard.
We have considered three ways population increases may benefit the future:
- Increasing the long-run population size,
- Increasing economic growth, and
- Reducing the probability of existential risks.
When it comes to changes to long-run population size, our main argument is that we will likely not depend on biological population growth for very long, as we put 50% credence that artificial people will be developed within this millennium. Even in cases where we still depend on biological population growth on the long-run, higher population growth is unlikely to have substantial effects on long-run population size. Furthermore, expected value considerations indicate that scenarios where population interventions actually increase long-run population size have a much lower potential for value than the other scenarios. Hence, increasing population size may be comparatively less important. Our arguments here rely on very speculative assumptions.
We consider the effects of a greater population on economic growth as the main channel through which population size can increase average wellbeing and reduce existential risks. We outline some arguably more cost-effective ways to increase economic growth over the next centuries. Further, we discuss several limitations in the existing arguments that establish the link between economic growth and existential risk, and propose more targeted ways of achieving similar benefits.
The effect of changes to current population size on increasing probability of survivor after a catastrophe are minor. We expect interventions that directly target survival after a catastrophe to be much more effective. The relationship between population and other existential risks is much less clear, if any, due to the scarcity of previous work on the topic. Changes to population growth will not affect AI risk under short AI timelines and global catastrophic biological risks.
Lastly, we discussed some potentially harmful effects of a higher population on the world. We argued that the case for larger populations worsening climate change is weak, and the negative animal welfare externality should only last for at most a century.
In conclusion, we believe that the current case for population growth falls short of the bar needed to consider it a priority. The arguments discussed above suggest that increasing population growth is beneficial on the margin, but the scale of these benefits does not seem comparable with other areas. That said, we believe that the strongest effect of population growth is on economic growth and population size itself over the next few centuries, as biological population growth might still be important in this timescale.
The analysis has some noteworthy limitations. First, we ignored issues related to the costs of interventions. We still think that the overall case for population growth as a priority seems weak enough that influencing population growth would have to be outstandingly tractable to change our minds—which could be true for developing countries. Second, our analysis of each specific argument was relatively shallow, and a more in-depth analysis of some of these could change our conclusions somewhat.
The robustness argument: We argued that population growth falls short of being a cause area assuming the longtermism paradigm. However, one may argue that it looks promising when compared to other longtermist cause areas in a timescale of 200-400 years, and that while each individual pathway to impact is not convincing enough, their combination passes the bar for a priority. A proper assessment of such claims is outside the scope of this report, as it involves a comparison with other cause areas. Nevertheless, we believe there are likely non-population interventions that increase economic growth in the coming centuries and that are more cost-effective. We also expect prima facie that a promising cause area has at least one robust theory of change, rather than several frail ones that add up to a strong thesis.
The arguments we discussed above suggest various ways in which we could change our minds regarding whether population growth should be a priority. We outline some of these considerations here:
- Evidence that sentient artificial people will not be feasible in the upcoming millennia, or that humanity would never want to create artificial people.
- Evidence that significant population decline is likely to last for many centuries, such as a good model of natural selection for fertility hinting that this process will never succeed to increase fertility above replacement.
- Evidence that population growth is likely to have a particularly large effect on existential risk through a channel other than economic growth.
We thank Gustav Alexandrie, Isabel Juniewicz, Kevin Kuruc, Eli Lifland, Andreas Mogensen, Jonas Sandbrink, Andreas Schmidt, and Philip Trammell for their comments and useful discussion, and Justis Mills for editing an earlier version. They do not necessarily agree with any conclusions or arguments. Any remaining mistakes are our own.
We provide a framework to discuss the conditions under which population growth interventions have significant long-run effects on long-run population effects. We assume that biological humans will not be replaced by other minds.
- is the average utility per person conditional on a certain population size p.
- if people provide on net positive externalities e.g. research and public goods.
- And if negative externalities dominate, e.g. an additional person uses rare resources such as land, and makes others worse off.
- If life is at subsistence level, , a decrease in average utility, would lead to humans starving and dying. The subsistence level s* may have positive, negative or neutral utility.
Population levels could be determined in 3 ways:
- The Malthusian population level :
- In the long-run, evolutionary pressure leads to the domination of those groups that grow the fastest. These may be the Malthusian dynamics.
- The optimal population level : where .
- A social planner sets the population size to maximise total wellbeing.
- The cartel population level , i.e. the one that maximises average experience: .
- The current society protects their average wellbeing at the disadvantage of additional people that could be born.
The long-run effects
No effects if Malthusian state or optimal social planner
- If the long-run population trajectory is p_m or p_o, then short-run population changes will not affect the long-run.
- This assumes that there are not sufficient lags that make it impossible to ever reach p_o even though the social planner would like to reach p_o.
Long-run effects in the cartel situation
- If there is a population cartel, the population size ends up around p_a.
- One may argue it will always choose the value p_a. In this case, population changes do not have long-run outcomes. But others may argue that population size ends up around p_a but it may be a bit bigger or smaller.
- Hence, population changes only have long-run effects if the long-run trajectory is a flexible p_a.
Size of the effect conditional on the cartel situation
- If one doubles p_a, then the additional utility is
- Notice that this additional utility is always smaller than
- If population increases do have very little externalities, then and p_a, are very close together and the effect of population increases is small.
- Hence, if the overall utility in the population cartel, is closer to the maximally achievable utility, then the effect of changing the population is smaller.
- The plausibility of the cartel situation:
- The less likely p_a, the smaller the expected long-run effects of population changes.
- The current world may be cartel-like: having more children would increase total utility but would potentially decrease the utility of some existing people. Hence, maybe a cartel is likely in the future.
- The costs of population increases may be marginal in the future: Today, population increases come at costs to the existing people: upbringing and education, pollution, meat-eaters, and more crowdedness. Most of these arguments may not apply anymore to a space-settling nation with very advanced technologies, e.g. AIs that can educate or mind emulations that can skip childhood or space settlement.
Table for illustrative purposes:
Cartel to maximise average wellbeing
The Day of Seven Billion was in October 2011.
Replacement fertility level is ~2.1 children per woman, instead of 2 children per woman, due to a naturally occurring difference in the sex ratio at birth, whereby slightly more males than females are born.
Only wellbeing has value.
An alternative way in which one could try to increase population growth would be to work on extending the human lifespan.
As an illustration, we put less than 70% credence on each of the first two assumptions.
Note that the longtermism paradigm or AI risk as a priority could equally well hold if one adopts, for instance, the asymmetry view in population ethics (Thomas 2019).
Given constant and positive average wellbeing, then total wellbeing is w = p*n, where p is the probability of long-run survival, and n is the total population size. Hence, ∂log w/∂log p = ∂log w/∂log n. For a more detailed model arriving at the same conclusion, see Alexandrie and Eden (forthcoming), “Is Extinction Risk Prevention the Unique Longtermist Priority? Not in Standard Population Models”.
Once countries complete their demographic transition, the model assumes that their fertility rate will fluctuate around a country-specific level, estimated using local data. This rate has been below replacement for countries having completed the demographic transition thus far. See their methodology report for more details.
While we will not discuss the mechanics and implications of this technology here, some resources that those interested could use to learn more include Digital People Would Be An Even Bigger Deal, Hanson (2016) and Bostrom (2014).
Note that higher population growth might still bring about some other changes that have indirect consequences for the long-run future, such as by speeding up when artificial beings are first created.
Depending on one’s willingness to accept fanatical decision theories, this argument might work even for very low probabilities of the vast majority of the population in the future being artificial, given how large the differences in the expected size of the world are. However, we believe it is unreasonable to attribute such low probabilities to this possibility. See Thomas and Beckstead 2022 and Russell 2021 for a more critical conclusion.
An implicit assumption here is a low mortality rate.
See Collins and Page (2019) for more details on this reasoning, evidence of this process already occurring, and relevant literature. As a counterargument, some models indicate that a reproductive strategy leading to low fertility could be evolutionarily advantageous (e.g., Hill and Reeve, 2005, and Boone and Kessler, 1999). We tentatively believe that models in such papers do not provide a good description of the existing evolutionary environment in most low fertility countries today, but we have not engaged with this literature in-depth.
Naturally, the entire human population descending from a single cultural group has its own issues, which we do not discuss here.
This might be a gradual process, but we denote it by a single point in time for simplicity.
We discuss below why we do not have strong reasons to believe that economic growth should not stagnate for long in the coming centuries.
We thank Kevin Kuruc for pointing out this possibility.
For instance, for such a cartel to want to fix population size for eternity, biological humans in the cartel would likely need to care about population size after they die, or be immortal. The cartel situation is also only one of several possible equilibria for biological humans to reach; other plausible cases are unaffected by preceding population growth, such as a Malthusian scenario and a scenario where population size is optimal for total wellbeing, neither of which.
Paul Romer received the Nobel Prize for his work on endogenous growth theory in 2018 which was subsequently extended to the semi-endogenous growth theory.
E.g. more talented women and people of colour replaced less talented white men in leadership positions.
Of course, as previously discussed, there are physical limits for population growth as well, but these limits are arguably much further away than the limits for the other factors generating growth.
We also tend to believe the stronger claim that fertility is very unlikely to be among the best ways to promote growth, but will not discuss this further.
See OECD chart of fraction of researchers in the US population
On the other hand, as the proportion of the population that works in research increases, the average quality of the researcher should decline.
While we will not get into the details of this discussion here, those interested in knowing more about the probability of the development of AI advanced enough to fully automate research can read this writeup, which links to further resources.
As in the previous section, worlds where delegating research to AI or artificial people is possible are much more valuable in expectation than worlds where delegation is impossible, leading the former to dominate expected value calculations.
This may also affect the probability of existential catastrophes.
They express a 35% credence in stagnation before the end of the century. We think that estimates in the range of 5-20% is more reasonable, where we refer to economic stagnation as a global GDP per capita growth below 0.5%.
We are highly uncertain about the timescales for evolutionary pressure to take place. One estimate, by Collins and Page (2019), suggests that heritability of fertility should lead to above fertility replacement rates in developed countries by the end of the century; however, we believe that this estimate is unreliable, as the fertility paths they project for the US and Europe on the short-run seems at odds with the fertility trends in these regions. We would find a more thorough evaluation of this force, also including a consideration of the countervailing forces, quite valuable for understanding how long we should expect low fertility to last.
Unpublished blogpost. The author can be contacted.
We do not have a good understanding as to how risks should arise from different levels of technological development, which would be particularly valuable for making such an assessment. We think that this would be a valuable research project.
This would be the case due to the possibility of cloning highly talented individuals, see Saint-Paul (2014) for an explanation.
Aschenbrenner develops a semi-endogenous growth model. Trammell (2021) finds that similar results hold, and are indeed strengthened, in the context of an exogenous growth model.
For instance, dedicating resources to developing resilient food or for building refuges.
More specifically, they change the UN population projections so that the labour force stays constant by the end of the century, instead of peaking and declining.
See also these estimates on how much lower the emissions of having children would be if one takes policy into account, and this discussion of overpopulation.
The costs of increasing fertility directly are certainly not that low (Stone 2020), but lobbying governments to implement these measures, or doing research on more cost-effective policies, could turn out to be considerably cheaper.
One may argue that reducing child mortality in developing countries is cheaper than increasing fertility. However, this consideration is not particularly strong. Saving a life by donating to AMF costs between 3500 and 7500 USD. However, if a child dies, parents often have an additional child. Hence, to create an additional child, the costs are likely at least twice as high—if not higher. More discussion here, here and here. Moreover, the effects of increasing the population in poor countries may be mixed, as more children also lead to less human capital investment. In fact, this is part of the reason other charities work on family planning.
We thank Kevin Kuruc for the discussion that inspired that point.
Thanks for writing this! For all the discussion that population growth/decline has gotten recently in EA(/-adjacent) circles, as a potential top cause area--to the point of PWI being founded and Elon Musk going on about it--there hasn't been much in-depth assessment of the case for it, and I think this goes a fair way toward filling that gap.
One comment: you write that "[f]or a rebound [in population growth] to happen, we would only need a single human group satisfying the following two conditions: long-run above-replacement fertility, and a high enough “retention rate”, that is, a large enough fraction of the descendants of this group continues to belong to the group." I think that's a good and underappreciated point, but I also think it's a bit weaker than it sounds at first, since something of a converse also holds. I.e. for permanent population decline to happen, we would only need a single human group satisfying the following two conditions: long-run below-replacement fertility, and a high enough "attraction rate", that is, a large enough fraction of people born outside the group continue to join the group. "Western civilization" has arguably been such a group for the last few generations, and it's not obvious to me that it (or its "descendants") won't continue to be for a very long time.
I've been interested in seeing more research on this, so thank you for writing it.
Your main argument
I'm sympathetic to your claim, namely:
"Population growth is unlikely to be the best way of keeping humanity safe from extinction risks"
From what I remember of reading WWOTF, I didn't get the impression that MacAskill was claiming that it was the best way. (I might be misremembering) Rather, I thought his argument was:
"In a world where many people having fewer children is better for humanity's longterm future, Will thinks the sign is probably the other way round"
Your 50% model
I thought your claim of a 50% chance of artificial people this millennium was interesting. I tried following your guesstimate model but it was a struggle to review it. Perhaps a separate write-up focusing on this would be of value?
Great post, and thank you for looking into this. I have an objection to your analysis, though as you point out, there's a substantial chance that most economic growth in the future will be driven by artificial intelligence or digital people, so this doesn't end up being overly consequential. But your section on the proportion of economic growth from population increases is, I think, a bit insufficiently optimistic on that factor. Economic growth rates are often theorized to be superexponential in terms of population. Increasing population contributes to growing the economy not only because the additional people directly contribute economic activity, but because marginal people also increase the growth rate. The 8,000,000,001st person isn't just another worker or scientist, they're another person all eight billion previous people can bounce ideas off of. One example is prehistoric Australia, which was settled by trans-oceanic explorers, settled into a stable population level in the tens of thousands for an extremely long time, and had stagnant economic and technological growth over this period. Adding another person there wouldn't increase their growth very much, but in the population regime where the world already has ten billion people, it makes a much bigger difference.
Hi Robi, thanks for your response. Are you referring to the endogenous growth models, where additional people have growth effects rather than level effects as in the semi-endogenous growth model (they increase the future growth rate)? or are you referring to historical trends? I am personally not very convinced of both.
"they're another person all eight billion previous people can bounce ideas off of"
This seems to depend on whether people actually have more connections. Even if they have more connections AND you think that research is driven by bouncing off ideas, you might think that this positive effect is smaller than the negative effect of research duplication when the population becomes bigger. But I agree it is plausible that the relevant parameter in the semi-endogneous growth model, lambda, is greater than 1.
I'm sympathetic to Robi's claim that growth rates may be a superexponential function of population. I suspect that the argument around people bouncing ideas off each other might be just one of the mechanisms.
I don't want to try to articulate all of those mechanisms here, but here's another one:
Charlotte sort of already addresses this, but just to clarify/emphasize: the fact that prehistoric Australia, with its low population, faced long-term economic and technological (near-)stagnation doesn't imply that adding a person to prehistoric Australia would have increased its growth rate by less than adding a person to an interconnected world of 8 billion.
The historical data on different regions' population sizes and growth rates is entirely compatible with the view that adding a person to prehistoric Australia would have increased its growth rate by more than adding a person to the world today, as implied by a more standard growth model.
Population growth is an existential risk. The HANDY model shows many regimes where self organizing systems grow to the point of catastrophic failure. These models attempt to explain the fall of prior isolated civilizations, including the complete loss of population such as Easter Island and the loss or civilization such as the collapse of Teotihuacan.
It would be worthwhile to factor in such risks.