Should Effective Altruists Have Kids?

Yes.

A friend of mine is an effective altruist, and quite serious about it. I think myself an effective altruist too, but he puts me to shame. Recently, we argued over whether he should have kids. He took the view that raising the child would be expensive, and that he would be better served saving the lives of children elsewhere. I disagreed – the child would themselves produce output, and would have a very good chance of donating as well. He pointed out that the same money could be used to recruit people to the cause – convincing two people should surely outweigh the creation of one additional child. I agreed that this could be true, and would be reliant on specific facts I did not have access to at the time. I would research it, and write an article on it.

I have since concluded, beyond a shadow of a doubt, that having kids is a highly effective cause area if you are unconcerned about AI. If you are concerned about AI, then it may be optimal to not have kids, just as you would be unconcerned with any investment which takes a similarly long time to mature. In scenarios where AI does not lead to the destruction of the world, it is unlikely that humans become an irrelevant factor of production on time scales sufficient to not make having children optimal.

I wrote this post to convince one person. (I did the utilitarian calculus – it checks out). Many of you, however, may be in the same situation, and want to know how best to improve the world. I hope that this might convince you to have kids. This essay is in two sections. The first will establish why people matter in the context of endogenous growth, and will try and find what matters with different AI timelines. The second will be a cost-benefit analysis.

And last, I say to my friend: I know you to be an unfailingly kind and caring person. You would be a more wonderful parent than anyone could hope for. I hope you have children, as they – and the world – would be better for it.

i. An Empty Planet

There are many ways in which more people existing in the world leads to a higher level of income. People matter because they create ideas, and the value of their ideas are not fully captured by their inventors. If we assume that there are increasing marginal costs to giving people with human capital, a population of a billion people is going to find far more ideas than a million people, who in turn will find more ideas than a thousand people will. Kremer (1993) empirically tests this, with the argument being that in a Malthusian world, any increase will be reflected in the growth rate of population, and that we did have a faster rate of population growth when the population was higher.

What is more, we invest far too little into research. While it is conceivably possible for there to be excessive investment into research,¹ empirical estimates are overwhelmingly in agreement that there is an insufficient amount of research. Chad Jones and John Williams (1998) estimate that the lower bound on the social returns of R&D in the United States is in excess of a 30% annual return. Ben Jones and Larry Summers (2020) use slightly different methods, and estimate a minimum social return of $10 for every $1 put into investment. Nordhaus (2004) estimates what percentage of the gains from an innovation are captured by the inventor, and arrives at an estimate of 2.2%. There is simply nothing else close to mattering than the externalities from ideas.

What empirical research has also shown is that there are declining returns in innovation. Ideas are becoming harder to find. Bloom, Jones, Van Reenen, and Webb (2020) show that in order to maintain the same level of growth, firms have to dedicate more and more resources – in particular, more and more researchers – to find new ideas. Research productivity has fallen by an average of 5% per year, every year. A possible explanation is that, if research is driven by the very best researchers, and we are good at determining who the very best researchers are, throwing more researchers at the problem should decrease productivity without improving returns. Ekerdt and Wu (2024) is the best explication of this view. They estimate that average researcher skill has declined 49% since 1960, due to the change in composition.

Of course, most research is done by profit-seeking companies, and we wouldn’t expect unprofitable behavior to long persist across the entire economy. Companies choose between employing workers in research and production, and there’s little reason to think that there should be a smooth transition of workers from production to research. Even Eckert and Wu do not think that there have been no declines in the ease of finding ideas – falling research productivity need not mean that research and development is socially inefficient.

If research into finding ideas has declining returns to scale or declining returns over time, then we must turn to Jones (1995), who proves that endogenous growth models return to being determined in the long run by parameters commonly taken to be exogenously given, such as population growth. Suppose that investment remains a constant fraction of total consumption, and it goes into either accumulating and maintaining capital, or into doing research. As the return from research falls, eventually all investment goes into capital, and we’re right back to the Solow model. Jones (2022) maps out the consequences of a fall in population. The level of economic well-being rises slightly as the ratio of labor to capital rises, and then stops. We get no growth, while the planet empties. Humanity goes extinct. I believe this to be bad; I assume you do too. If we wait too long, it may even be the best possible outcome. If we take the standard view that parents get utility from their children’s utility, and the world where humanity is going extinct is really bad, the optimal course is to not keep them around! From our position, avoiding this is the best possible outcome, and we must strive to prevent it happening.

This does not even consider the effect of market size on productive and allocative efficiency, although these surely exist. If opening trade leads to an increase in productive efficiency due to increased market size, then increasing the population will do the same.

ii. A Cost-Benefit Analysis

Having established the importance of children for economic growth, we should construct a simple model for describing the costs and benefits, and estimate the parameters. We will assume that we have a discount rate of zero, both for convenience, and because many believe that there is no moral difference between those who exist now, and those who are to come. I think this is implicit in caring about the welfare of humans who are not yet extant, which is the entire reason we would be opposed to extinction. Adding a discount rate would favor projects occurring now, rather than occurring later. We will also assume for convenience that the cost to save a life will remain constant in real terms over time. An assumption that the cost to save a life will rise in the future is isomorphic to a positive discount rate; an assumption that it will be cheaper in the future isomorphic to a negative discount rate.

A difficult consideration is taking into account the value of the spillovers that the saved lives in the developing world would produce. Saving the life of someone in the developed world is more valuable than someone in the developing world, because they will end up producing more things which benefit others. One can see this clearly in a hypothetical trolley problem. Imagine a train were on track to kill two people, who are identical in every respect, except that one will discover a new antibiotic which will save millions of lives, and the other will not. The clear choice is the former. Because researchers are far more likely to come from the developed world, saving their lives is a better use of resources, all else equal. The present emphasis on saving lives in the developing world is simply a question of cost-effectiveness.

I presume that it is correct to have a child if the value they produce plus the expected value of future donations exceeds the lives which could be saved. I have created a short software program into which you can plug in your personal estimates (just copy-paste the code into VSCode), and explore how much different specifications affect the outcome; in the next section, I will argue for my personal preferences. I will have the analytical trick, at every point where things are uncertain, try to take the pessimistic side, so that I am underestimating the benefits of children.

The effective altruism survey which rethink priorities puts out does not contain important information about the surveyed population, such as income, intelligence, or education. To get around this, I used the Slatestarcodex survey from 2024, and included only those respondents who answered “Yes” to the question of EA identification. Slightly over a tenth identified as such, giving us a sample of 660 people.

EAs are extremely intelligent, educated, and rich. The median SAT score² was 1540, and the mean 1517. Because of people selecting their own social circles, you may not fully grasp how elevated this is; the median person is above the 99th percentile in ability. This gives us a rough estimate of a median IQ of 138, which is really, really high. By income, controlling for education being complete gives us an average salary of $145,000. This will be understating income, as it rises over time and EAs are disproportionately young; I did not correct for this because the Census reports age brackets 15-24, 25-34, and the SSC survey reports it 10-19, 20-29, and correcting this would be extremely aggravating for a very small benefit.

These things are persistent over generations. We do not care about the heritability, but rather the persistence alone. It is true that there is regression to the mean, but this can be misleading, as people are not actually drawn from the same distribution. If someone has smart parents, it is likely that their distribution is higher. To account for this, I am going to say that, on average, each person declines 20% to the mean. At the same time, economic growth will occur, such that people of equivalent ability in the future will earn more. Estimating this precisely would be misleading, as it may convey the impression that there is certainty over either of the parameters. We can expect people’s children to do about as well as they did.

Someone earning $150,000 a year, after paying taxes, has a net pay of $100,000. The average cost of raising a single child is $15,500 after adjusting for inflation; the price will fall as people have more children, and doubtless children could be raised for less, but it’s a reasonable baseline. This gives us a total cost of $279,000 to raise a child to adulthood. With a career lasting 40 years, if they have a 70% chance of being an effective altruist, then it works out without having to consider the surplus that they would generate for others.

Estimating the surplus generated is actually much harder than I thought at the outset. If the United States has 4,452 researchers per million people, or .4 percent, and inventors gets 2.2% of the social value (a high estimate, since Nordhaus is including a 10% discount rate per year), then assuming their likelihood of being a researcher is totally random, the child is adding an expected value of $27,275 per year in social surplus. (And of course, we know that they are more likely than random to be a research worker). What they pay in taxes clearly also has some social value, even if some of it is wasted. It would take $133,333 per year to raise a child to adulthood for it to not be worthwhile. Since it isn’t necessarily being donated to charity, we do not have the ease of comparing the relative merits of consumption between developed and developing world. Think of it as if Givedirectly had to waste some percentage of money. If you would not donate if it had a 9 to 1 ratio, then you be best having kids.

Calibrating the parameters of the model is quite uncertain and difficult. Nevertheless, I hope to convince you that the value of having a child is so large as to not be in any doubt as to its cost-effectiveness. This is for a single child, and it is worth noting that the marginal cost of additional children will be substantially lower. Even under a totalizing duty to help others, having kids is clearly worthwhile.

1. For example, in the Aghion-Howitt two-period model, individuals choose between research and earning a skilled worker’s wage. An innovation in a prior period secures monopoly rents, but at the expense of denying the previous inventors their rents. Everyone anticipates this, and so no one invests in research at all, because any gain in the first period will be obliterated by the second period.

2. Some people put their whole scores, rather than sub-section scores. I simply divided any scores above 800. Some people listed their ACT scores, so I dropped all scores below 100. (There was also someone who was clearly joking, putting their IQ as 60 and their SAT score as 40, so I dropped them too).