Thanks for the comment, Patrick! That makes sense. I suspect status quo bias is an important blocker to transfer learning with respect to applying arithmetic and adjacent methods to figure out how to improve the world. Quantification is unquestionable in engineering projects, but often absent in charitable projects. There is quantification in the effective altruism community, but people still like to conform to societal norms about what it means to contribute to a better world, including about which methods are legitimate to evaluate interventions.

Thanks for the post, Lizka and Ben!

Our biggest recommendation is: to the extent that you're targeting animal welfare improvements in the normal era, you probably should discount heavily. Discounting to zero work which pays off in more than ten years is maybe one viable but very simplistic model.

I like Ege Erdil's median time of 20 years until full automation of remote work (bets are welcome), and I would not neglect impact after that. So I think a typical discount rate of 3 % implying 33.3 years of impact is fine. However, I agree with your recommendation conditional on your timelines.

Among the organisations working on invertebrate welfare I recommended, I would only recommend the Shrimp Welfare Project (SWP) ignoring impact after 5 or 10 years. From Vetted Causes' evaluation of SWP, "SWP also informed us that it typically takes 6 to 8 months for SWP to distribute a stunner and have it operational once an agreement has been signed". Assuming 0.583 years (= (6 + 8)/2/12) from agreements to impact, and 1 year from donations to agreements, there would be 1.58 years (= 0.583 + 1) from donations to impact. All my other invertebrate welfare recommendations involve research, which I guess would take significantly longer than 1.58 years from donations to impact. For SWP, there would be 3.42 (= 5 - 1.58) and 8.42 years (= 10 - 1.58) of impact neglecting impact after 5 and 10 years. I assumed "10 years" of impact to estimate the past cost-effectiveness of SWP's Humane Slaughter Initiative (HSI). As a result, supposing SWP's current marginal cost-effectiveness is equal to the past one of HSI including all years of impact, SWP's current marginal cost-effectiveness neglecting impact after 5 and 10 years would be 34.2 % (= 3.42/10) and 84.2 % (= 8.42/10) of the past cost-effectiveness of HSI.

Among the interventions for which I estimated the cost-effectiveness accounting for effects on soil nematodes, mites, and springtails, I would only recommend buying beef ignoring impact after 5 or 10 years. I estimated it is 63.8 % as cost-effective as HSI has been, and I guess it would only take 2 years or so from buying beef to its benefits materialising via an increased agricultural land. For my AI timelines, cost-effectively decreasing human mortality also increases agricultural land cost-effectively. However, for a life expectancy of 70 years, ignoring impact after 5 and 10 years would make the cost-effectiveness 7.14 % (= 5/70) and 14.3 % (= 10/70) as high.

Thanks for the follow-up, Jan! I agree with all your points except one. I think the best guess distribution about the energy density should be roughly continuous, which implies an infinitesimal probability of any specific value, including that implying a flat universe (which should be very close to the mode of the distribution).

Thanks for the context, Joel!

Hi Joel,

Wang et al. (2022) estimated 250 mL/day more of SSBs is associated with a 7 % higher chance of death. What is your best guess for which fraction of this is causal? I have also asked the 1st 2 authors. I would like to that to estimate the expected decrease in human-years due to consuming SSBs.

Thanks for the follow-up, Nick. I put 100 % weight on expectational total hedonistic utilitarianism as I understand it. However, l am still more uncertain about at least some aspects of cause prioritisation than Marcus, like whether saving human lives is benefial or harmful.

Thanks for clarifying. I still think the variation in the mean temperature is useful because they constrain the seasonal variation. Each season lasts for 0.25 years (= 1/4), so a season becoming e.g. 10 ºC cooler will make the year 2.50 ºC (= 0.25*10) cooler. Assuming the effect of AMOC's collapse on Spring and Summer and is negligible, and that the effect on Autumn and Winter is similar, these would cool 2 (= 4/2) times as fast as I estimated above. So by 0.0600 (= 0.0300*2) to 0.400 ºC/year (= 0.200*2) in Western Europe, and up to 0.200 (= 0.100*2) to 1.33 ºC/year (= 0.667*2) in northern latitudes. I know you pointed out that the extremes matter, but I think the seasonal variations are still relevant. At least now, deaths from moderate cold are much larger than from extreme cold.

I did not calculate the probability of the European population decreasing by more than 10 %. I simply speculated it is lower than 0.01 %. For context, the deaths from non-optimal temperature as a fraction of the population in Europe in 2021 were 0.0416 %. The respective death rate across time is below. For the European population to become at least 10 % smaller, that death rate would have to become at least 240 (= 0.1/(4.16*10^-4)) times as large, which seems a lot considering how little is has varied across time.

Thanks for the post, Marcus!

I fully endorse expectational total hedonistic utilitarianism, so I only consider other moral theories as useful heuristics, and only care about empirical uncertainty. However, uncertainty about the welfare of different beings still makes me very uncertain about which interventions are the most cost-effective at the margin. I think all of the following are plausible candidates. Interventions:

• Targeting humans to increase (decrease) agricultural land  if the welfare ranges I derived for soil nematodes, mites, and springtails, based on Rethink Priorities' (RP's) mainline welfare ranges, are roughly right, and these soil animals have negative (positive) lives.
• Targeting humans to increase human welfare if the welfare ranges of animals are much smaller than RP's mainline welfare ranges.
• Targeting invertebrates if the welfara ranges of the soil animals I mentioned just above are much smaller than I estimated, and RP's mainline welfare ranges are roughly right.
• Targeting non-human vertebrates if the welfare ranges of invertebrates are much smaller than RP's mainline welfare ranges of intervebrates, but their mainline welfare ranges of vertebrates are roughly right.

It’s relatively easy to argue, all else equal, that it’s better to save 100 lives rather than 10

I worry you are underestimating the uncertainty here. I think interventions saving human lives at a low cost are among the most cost-effective, but it is still unclear to me whether they are beneficial or harmful. I estimate GiveWell’s top charities increase the welfare of soil nematodes, mites, and springtails 87.6 k times as much as they increase the welfare of humans based on Rethink Priorities' (RP's) mainline welfare ranges, and my best guess than those soil animals have negative lives. However, I believe they could roughly as easily have positive lives^[1], in which case I would consider saving human lives harmful.

1. ^{^}

   I calculate those soil animals have negative lives with a probability of 58.7 %, 55.8 %, and 55.0 %.

Thanks, Simon!

I am confident increasing the consumption of at least beef increases agricultural land, thus decreasing the animal-years of soil nematodes, mites, and springtails, despite decreasing the human population due to being less healthy. Eating 85 g of red meat is associated with losing 1 microlife, 30 min. I speculate the causal effect is 1/3 as large, which implies eating 100 g of beef shortens one's life by 11.8 min (= 30*1/3*100/85), 2.24*10^-5 year (= 11.8/60/24/365.25). For the global agricultural land per capita in 2022 of 0.60 ha, that implies a decrease in agricultural land of 0.134 m²-year (= 2.24*10^-5*0.60*10^4), or 1.34 m²-year/beef-kg (= 0.134*1/0.1). The urban land per capita in 2015 was 257 m²-year (= 1.91*10^12/(7.44*10^9)), so the decrease in urban land would be 0.00576 m²-year (= 2.24*10^-5*257), or 0.0576 m²-year/beef-kg (= 0.00576*1/0.1), only 4.30 % (= 0.0576/1.34) of the decrease in agricultural land, and therefore negligible. The decrease in agricultural land due to decreasing the human population is only 0.411 % (= 1.34/326) of the increase in it needed to consume the beef. So increasing the consumption of beef increases agricultural land.

Thanks, Toby.

I definitely agree that the subjetive guesses related to RP's mainline welfare ranges are on shaky ground. However, I feel like they are justifiably on shaky ground. For example, RP used 9 models to determine their mainline welfare ranges, giving the same weight to each of them. I have no idea if this makes sense, but I find it hard to imagine which empirical evidence would inform the weights in a principled way.

In contrast, there is reasonable empirical evidence that effects of interventions decay over time. I guess quickly enough for the effects after 100 years to account for less than 10 % of the overall effect, which makes me doubt astronomical longterm impacts.

I would also say there is reasonable evidence that the risk of human extinction is very low. A random mammal species lasts 1 M years, which implies an annual extinction risk of 10^-6. Mammals have gone extinct due to gradual or abrupt climate change, or other species, and I think these sources of risk are much less likely to drive humans extinct. So I conclude the annual risk of human extinction is lower than 10^-6. I guess the risk 1 % as high, 10^-7 (= 10^(-6 - 2 + 1)) over the next 10 years. I do not think AI can be interpreted as other species because humans have lots of control over its evolution.