Summary

• I think GiveWell’s top charities may be anything from very harmful to very beneficial accounting for the effects on terrestrial arthropods (e.g. insects^[1]).
• This warrants further research on:
  • The moral weight of terrestrial arthropods.
  • The change in forest area caused by GiveWell’s top charities.
  • The sign of the welfare of terrestrial arthropods.
• Cluelessness about the (short and long term) effects of GiveWell’s top charities may motivate one to prefer longtermist interventions^[2], as Hilary Greaves suggests here, and Alex HT describes here.

Acknowledgements

Thanks to Anonymous Person, Michael Dickens, Michael St. Jules, and Ramiro.

Methods

I estimated the relative variation in the cost-effectiveness of GiveWell’s top charities due to terrestrial arthropods from the product between:

• Net change in forest area per capita in 2015^[3] (m²/person), in each of the countries analysed by GiveWell for its top charities here^[4]. I calculated this from the ratio between:
  • Net change in forest area in 2015 by country (ha), based on these data from Our World in Data (OWID).
  • Population in 2015 by country, based on these data from OWID.
• Variation in welfare of terrestrial arthropods as a fraction of that of a human due to deforestation (1/ha). I calculated this from the negative of the product between:
  • Welfare of a terrestrial arthropod as a fraction of that of a human.
  • Decrease in density of terrestrial arthropods due to deforestation (1/ha).

Welfare of a terrestrial arthropod as a fraction of that of a human

I estimated this from the product between:

• Moral weight of terrestrial arthropods relative to humans. I set this to:
  • MW($N_{ta}$)/MW($N_h$), where:
    • MW(N) is the moral weight as a function of the number of neurons^[5]. I considered the ones available in foodimpacts.org:
      • Logarithmic: log(N).
      • Square root: N^0.5.
      • Linear: N.
      • Quadratic: N^2.
    • $N_{ta}$ is the number of neurons of a terrestrial arthropod, which I set to 404 k. This is the mean of a lognormal distribution with 5th and 95th percentiles equal to the lower and upper bound of 100 k and 1 M given here by Animal Ethics^[6].
    • $N_h$ is the number of neurons of a human, which I set to 86 G based on Herculano-Houzel 2012.
  • Rethink Priorities’ (RP’s) median welfare range estimates, given in this post from Bob Fischer, for:
    • Black soldier flies, 0.013.
    • Silkworms, 0.002.
• Welfare of a terrestrial arthropod as a fraction of that of a human not adjusting for moral weight. I computed this from the ratio between the total welfare score of a “wild bug” and “human in a low middle-income country” according to the Weighted Animal Welfare Index (WAWI) of Charity Entrepreneurship (CE), which was presented here.

Decrease in density of terrestrial arthropods due to deforestation

I estimated this from the product between:

• Decrease in density of terrestrial arthropods due to deforestation as a fraction of that in global land. I computed this from the difference between the density of terrestrial arthropods in forested and deforested areas as a fraction of that in global land, which I set to 1.55 and 0.899. These are the means of lognormal distributions with 2.5th and 97.5th percentiles equal to the lower and upper bounds of the 95 % confidence intervals (CIs) given here by Brian Tomasik for:
  • Rainforest, 1.02 and 5.
  • Cerrado, 0.7 and 3.
• Density of terrestrial arthropods in global land (1/ha). I computed this from the ratio between:
  • Number of terrestrial arthropods, which I set to 4.04*10^18. This is the mean of a lognormal distribution with 5th and 95th percentiles equal to the lower and upper bound of 10^18 and 10^19 given here by Brian Tomasik.
  • Global land area (ha), which I set to 13.0 Gha based on these data from The World Bank.

Results

The calculations and full results are in this Sheet. The tables below contain the moral weight of terrestrial arthropods relative to humans, variation in welfare of terrestrial arthropods as a fraction of that of a human due to deforestation^[7], net change in forest area per capita in 2015, and relative variation in cost-effectiveness of GiveWell’s top charities due to terrestrial arthropods. I present the results by country for simplicity, but they would ideally be shown for each top charity of GiveWell. $N_{ta}$ and $N_h$ are the number of neurons of terrestrial arthropods and humans.

Discussion

The results suggest changes in the number of terrestrial arthropods may be anything from crucial to negligible. The relative variation due to terrestrial arthropods for the mean country is:

• 386 k (crucial) if the moral weight function is logarithmic.
• 16.6 μ (negligible) if the moral weight function is quadratic.

Terrestrial arthropods completely dominate the cost-effectiveness of GiveWell’s top charities for RP’s median welfare range estimates. Those for black soldier flies and silkworms lead to a cost-effectiveness 9.78 k and 1.50 k times higher.

I think RP’s median welfare range estimates are more accurate than those solely based on the number of neurons. As Adam Shriver put it here:

Neuron counts likely provide some useful insights about how much information can be processed at a particular time, but it seems unlikely that they would provide more useful information individually than a function that takes them into account along with other plausible markers of sentience and moral significance.

I have not accounted for the welfare of farmed animals, but I guess the effect on terrestrial arthropods dominates. Based on:

• This, the total number of neurons of terrestrial arthropods is 20 k times that of farmed animals excluding arthropods.
• CE’s WAWI, the ratio between the total welfare score of factory-farmed animals and the “wild bug” is much lower than 20 k. It ranges from 47.6 % (= 20/42) for beef cows, to 136 % (= 57/42) for both turkeys and United States’ laying hens.

The present analysis is very shallow, and therefore the results are not resilient. However, they point to the value of further research on:

• The moral weight of terrestrial arthropods.
  • The estimates I provided based solely on the number of neurons suggest the effects on terrestrial arthropods can be anything from crucial to negligible.
  • RP’s median welfare range estimates for bees, black soldier flies, and silkworms imply terrestrial arthropods dominate, but the 5th percentile is 0 for all those species (see here).
• The change in the number of terrestrial arthropods caused by GiveWell’s top charities.
  • I supposed it to be directly proportional to net change in forest area per capita, but this is simplistic.
  • For example, I guess the consumption of the beneficiaries of GiveWell’s top charities is lower than that of the mean citizens of their countries, so I arguably overestimated the magnitude of the net change^[11].
• The sign of the welfare of terrestrial arthropods.
  • I assumed it to be negative in agreement with CE’s WAWI, but it is unclear whether this is the case. From this post of Karolina Sarek, “there is lots of room to improve these [WAWI’s] numbers, particularly with deeper investigation into the lives of wild animal[s]”.
  • For pushback on the view that net welfare in the wild is negative, I mostly recommend checking this preprint from Heather Browning and Walter Weit. In addition, I also liked the posts linked here by Michael St. Jules:
    • How Much Do Wild Animals Suffer? A Foundational Result on the Question is Wrong. by Zach Freitas-Groff.
    • Life History Classification and Insect herbivores, life history and wild animal welfare by Kim Cuddington.
    • The Unproven (And Unprovable) Case For Net Wild Animal Suffering. A Reply To Tomasik by Michael Plant.
  • Before looking into the above, I leaned towards the sign being negative. Now, I am essentially agnostic.

The uncertainty about the sign of the welfare of terrestrial arthropods, and the possibility of effects on these dominating those on humans imply GiveWell’s top charities may be anything from very harmful to very beneficial from a neartermist point of view.

Even neglecting animal welfare, I am not confident GiveWell’s top charities are robustly good. For example, the effect of distributing bednets^[12] on the population size is unclear. According to Wilde 2019, that increases fertility in the short term, but decreases it in the long term:

The effect on fertility is positive only temporarily – lasting only 1-3 years after the beginning of the ITN distribution programs – and then becomes negative. Taken together, these results suggest the ITN distribution campaigns may have caused fertility to increase unexpectedly and temporarily, or that these increases may just be a tempo effect – changes in fertility timing which do not lead to increased completed fertility.

I do not know whether increasing the population size is good or bad, but I think it may well be the major driver for both the nearterm (e.g. next 100 years) and total effect of GiveWell’s top charities. For example, it influences not only the change in forested area, but also greenhouse gas emissions^[13], policy, and economic growth more broadly.

To be honest, I personally do not think the best solution to the cluelessness about the (short and long term) effects of GiveWell’s top charities is to make their analyses more sophisticated. I would say focussing on longtermist interventions is better, as their (longterm) effects are more predictable. Regardless of your preferred solution, I highly recommend watching this talk from Hilary Greaves, and reading this post from Alex HT^[14] (if you have not done so).

1. ^{^}

    Fun fact, in portuguese, my last name Grilo means cricket (which is a terrestrial arthropod).

2. ^{^}

    By which I mean increasing the share of resources going towards longtermist interventions, not that neartermist ones should not receive any resources. Relatedly, I liked this post from Jan Kulveit and Gavin Leech.

3. ^{^}

    This is only accurate to the extent the annual impact on net forest area of the people saved by GiveWell’s top charities is similar to that of the mean citizens of their countries in 2015.

4. ^{^}

    GiveWell’s top charities are Malaria Consortium, Against Malaria Foundation (AMF), Helen Keller International, and New Incentives.

5. ^{^}

    As illustrated by William MacAskill in What We Owe to the Future.

6. ^{^}

    I calculated the mean of the lognormal distributions using this Sheet, which is described here.

7. ^{^}

    A value of 1/ha means the suffering of terrestrial arthropods prevented by deforesting 1 ha equals the welfare of 1 person.

8. ^{^}

    Data about the net change in forest area for the other countries analysed by GiveWell for its top charities were not available.

9. ^{^}

    There was no net change in forest area for Mali, so the results are both 0.

10. ^{^}

     Ideally, the mean would be weighted by the funding GiveWell has directed to each of the countries.

11. ^{^}

     I believe this does not affect the conclusions. The uncertainty in the moral weight is much larger than that in the consumption of the beneficiaries, so the relative variation in cost-effectiveness would still be anything from crucial to negligible.

12. ^{^}

     This is done by Against Malaria Foundation, which is one of GiveWell’s top charities.

13. ^{^}

     These have short term effects such as increasing heat-related mortality (0.226 mlife/t based on Bressler 2021), apart from increasing the existential risk due to climate change (0.273 bp/Tt based on this).

14. ^{^}

     For pushback on longtermism, Michael St. Jules suggested checking the following:

    - The Future Might Not Be So Great by Jacy.

    - A longtermist critique of “The expected value of extinction risk reduction is positive” by Anthony DiGiovanni.

    - Why I am probably not a longtermist by Denise Melchin.

    - This thread started by Michael.

    - Which World Gets Saved by Philip Trammell.

    - Why does (any particular) AI safety work reduce s-risks more than it increases them? by Michael.

    - The motivated reasoning critique of effective altruism by Linchuan Zhang.

    - This thread started by Luke Muehlauser.