Hey Vasco! I just read the summary here, so my apologies if I'm missing something important.
I think I disagree with the line of reasoning you're following. For instance, you say:
>>I estimated its HSI has been 0.0292 % as cost-effective as HIPF accounting for effects on the target beneficiaries, and soil ants, termites, springtails, mites, and nematodes.
It would be very surprising to me if we can know with any degree of confidence that exploratory research into highly speculative areas is >3000X more cost effective than Shrimp Welfare Project's main program—a program which seems to be one of the most cost effective on the planet!
I don't doubt that every step in specific chain of reasoning is mostly correct. Rather, I think where I get off the boat here is that I think the uncertainty in these flowthrough effects tends to dominate after a certain point. I get the sense you are willing to take these much more seriously and literally.
Another area of disagreement is the general framing of suggesting that we should always just fund the one most impactful thing. I agree that this makes sense on the margin, but I expect we're generally more likely to get to a better world in the end if we take a portfolio approach and fund and encourage lots of projects that pass a certain bar. A large part of this is driven by the fact that you get seriously diminishing returns on projects (e.g. I'd be surprised if CEARCH could cost effectively deploy SWP's full annual budget).
So I'd probably frame all of this as more "Both CEARCH's program and SWP's program perform very well on these metrics, and we should consider funding them."
You say SWP's HSI "seems to be one of the most cost effective [programs] on the planet". Is this mostly because it has helped lots of shrimp per $ (I estimated 15 k shrimp per $)? If so, and your best guess is that soil animals have negative lives in expectation, you should consider CEARCH's HIPF much more cost-effective because it affects way more animals (I estimated it decreases 5.07 billion soil-animal-years per $)? For SWP's HSI to have increased the welfare of shrimp more cost-effectively than HIPF increases welfare, HSI having helped 1 shrimp would have to be better than HIPF decreasing 338 k soil-animal-years (= 5.07*10^9/(15*10^3)). Assuming that welfare per animal-year is proportional to "number of neurons as a fraction of that of humans"^"exponent of the number of neurons"[1], HIPF is always way more cost-effective than HSI has been.
I am very uncertain about whether soil animals have positive or negative lives, but it is common for people to have a best guess that wild invertebrates have negative lives. I supposed the welfare per animal-year of soil ants/termites/springtails/mites/nematodes is -25 % that of fully happy soil ants/termites/springtails/mites/nematodes. I assumed this holds for all biomes, but I guess there is variation in reality. Karolina Sarek, Joey Savoie, and David Moss estimated-42 % for the “wild bug” in 2018, which is more negative than what I assumed.
I believe recommending on equal foot many interventions with similar marginal cost-effectiveness would definitely make sense. However, I estimate HSI has been only 0.0292 % as cost-effective as HIPF, which is a very large difference. I suspect we mostly disagree about the differences in cost-effectiveness, not about what to recommend conditional on the differences in cost-effectiveness.
I describe this formula in the 2nd bullet of the summary here. For an exponent of 0.188, the formula explains78.6 % of the variance in the welfare ranges in Bob Fischer's book about comparing welfare across species. The results I presented in the top post are for my preferred exponent of 0.5, which results in a welfare per soil-animal-year much closer to 0 than the exponent of 0.188 that explains very well the welfare ranges in Bob's book.
Cost-effectiveness of electrically stunning rainbow trout
I estimate a welfare loss from rainbow trout’s air asphyxia of 0.0105 QALY/fish and 0.0150 QALY/fish-kg, of which 100 % come excruciating pain.
I determine electrically stunning rainbow trout reduces their welfare loss by 0.00895 QALY/fish or 0.0128 QALY/fish-kg given air asphyxia as the alternative. The former is 6.20 times my estimate for electrically stunning shrimp of 0.00144 QALY/shrimp.
I estimate a cost-effectiveness of electrically stunning rainbow trout accounting only for these of 0.182 QALY/$, whichis:
26.0 % of my estimate for the cost-effectiveness of cage-free corporate campaigns accounting only for chickens of 0.701 QALY/$.
71.5 % of my estimate for the cost-effectiveness of broiler welfare corporate campaigns accounting only for chickens of 0.255 QALY/$.
I favour improving the slaughter of shrimp over fish:
I estimate SWP has increased the welfare of shrimp much more cost-effectively than directly paying for electrically stunning rainbow trout increases the welfare of these.
There may be interventions more cost-effective than buying electrical stunners, but this holds for both fish and shrimp.
I think accounting for how changes in the price of the target animals affects their consumption, and therefore land use and soil animals would increase the cost-effectiveness of electrically stunning shrimp due to this increasing agricultural land, whereas it may make electrically stunning fish harmful due to this decreasing agricultural land.
In this post, I estimate the cost-effectiveness of electrically stunning rainbow trout accounting only for these (the target beneficiaries). Here are my calculations.
Welfare loss from rainbow trout’s air asphyxia
I calculate the welfare loss in QALYs from rainbow trout’s air asphyxia linked to each category of pain defined by the Welfare Footprint Institute (WFI) multiplying the following:
The time experiencing the pain in years. I set this to the mean between the lower and upper bound from Fig. 1 of Schuck-Paim et al. (2025).
The intensity of the pain in humans as a fraction of that of a fully happy human life. I guess:
Annoying pain in humans is 10 % as intense as a fully happy human life, which implies 10 days (= 1/0.1) of annoying pain neutralise 1 day of fully happy life.
Hurtful pain in humans is as intense as a fully happy human life, which implies 1 day of hurtful pain neutralises 1 day of fully happy life.
Disabling pain in humans is 10 times as intense as a fully happy human life, which implies 2.4 hours (= 24/10) of disabling pain neutralises 1 day of fully happy life.
Excruciating pain in humans is 100 k times as intense as a fully happy human life, which implies 0.864 seconds (= 24*60^2/10^5) of excruciating pain neutralises 1 day of fully happy life.
The intensity of a fully happy rainbow trout’s life as a fraction of that of a fully happy human life of 1.26 %, which is the square root of Rethink Priorities’ (RP’s) number of neurons of carp as a fraction of that of humans of 0.0160 %. That fraction is 5.25 % (= 0.0126/0.24) of the welfare range of carp as a fraction of that of humans presented in Table 8.6 of Bob Fischer’s book about comparing welfare across species.
I obtain the total welfare loss adding the components of the 4 categories of pain. I infer the welfare loss per unit mass at slaughter assuming a slaughter weight of 0.700 fish-kg, which is the mean between the lower and upper bound from Schuck-Paim et al. (2025).
The results are below. I estimate a welfare loss from rainbow trout’s air asphyxia of 0.0105 QALY/fish and 0.0150 QALY/fish-kg, of which 100 % come from excruciating pain.
Category of pain
Time in pain (min/fish)
Time in pain (min/fish-kg)
Welfare loss (QALY/fish)
Welfare loss (QALY/fish-kg)
Welfare loss as a fraction of the total
Any
12.7
18.1
0.0105
0.0150
100%
Annoying
0.850
1.21
2.04E-09
2.92E-09
1.94E-07
Hurtful
3.04
4.34
7.31E-08
1.04E-07
6.94E-06
Disabling
4.38
6.26
1.05E-06
1.50E-06
0.0100%
Excruciating
4.38
6.26
0.0105
0.0150
100%
Cost-effectiveness
I stipulate electrical stunning decreases the welfare loss from air asphyxia by 85.0 %, which is the mean between the lower and upper bound of 70 % and 100 % used in Schuck-Paim et al. (2025). So I determine electrically stunning rainbow trout reduces their welfare loss by 0.00895 QALY/fish or 0.0128 QALY/fish-kg given air asphyxia as the alternative. The former is 6.20 times my estimate for electrically stunning shrimp of 0.00144 QALY/shrimp, which assumes the welfare range of shrimp is equal to the square root of their number of neurons as a fraction of that of humans.
I suppose electrically stunning rainbow trout costs 0.06 €/fish-kg, as considered in Schuck-Paim et al. (2025), which was 0.0702 $/fish-kg on 7 July 2025. This together with the above result in a cost-effectiveness of electrically stunning rainbow trout without accounting for changes in consumption of 0.182 QALY/$, whichis:
26.0 % of my estimate for the cost-effectiveness of cage-free corporate campaigns accounting only for chickens of 0.701 QALY/$.
71.5 % of my estimate for the cost-effectiveness of broiler welfare corporate campaigns accounting only for chickens of 0.255 QALY/$.
0.885 % of my estimate for the past cost-effectiveness of SWP’s HSI accounting only for shrimp of 20.6 QALY/$.
Note the cost-effectiveness of cage-free and broiler welfare corporate campaigns varies.
My recommendations
I favour improving the slaughter of shrimp over fish:
I estimate SWP has increased the welfare of shrimp much more cost-effectively than directly paying for electrically stunning rainbow trout increases the welfare of these.
There may be interventions more cost-effective than buying electrical stunners, but this holds for both fish and shrimp.
I think accounting for how changes in the price of the target animals affects their consumption, and therefore land use and soil animals would increase the cost-effectiveness of electrically stunning shrimp due to this increasing agricultural land, whereas it may make electrically stunning fish harmful due to this decreasing agricultural land.
Electrically stunning animals makes their consumption more expensive, thus decreasing it, and increasing that of other foods.
I estimate changes in the consumption of food impact soil animals much more than directly affected animals, and that increasing agricultural land benefits soil animals for my best guess that they have negative lives.
Farmed shrimp require less agricultural land per food-kg than other animal-based foods, and legumes. So farmed shrimp being replaced with these would tend to increase agricultural land, and therefore benefit soil animals.
Farmed fish require more agricultural land per food-kg than farmed shrimp, eggs, and legumes. So farmed fish being replaced with these would tend to decrease agricultural land, and therefore harm soil animals.
I recommend funding CEARCH’s HIPF instead of SWP. I estimated its HSI has been 0.0292 % as cost-effective as HIPF accounting for effects on the target beneficiaries, and soil ants, termites, springtails, mites, and nematodes.
AI Use Note: Main body text entirely human written. Claude (Opus 4.8) helped develop models of animal life histories in the appendix.
Cross-posted from Good Structures.
Executive Summary
* Animal advocates sometimes make claims like “there are X of this animal...
Adapted from my Substack, Funding Anthropalypse.
Short version: if you want a share of the coming Anthropic and OpenAI windfall - the $37bn+ that could be in play next year - the way in is to become 'legibly excellent', so the evaluators and donors that frontier lab staff already trust point them to yo...
Disclaimer: Although I work on the Groups Team at CEA, I’m writing this in a personal capacity, and this post does not constitute an endorsement by CEA.
Agency - the realisation that you really can just do things.
TL;DR
Biosecurity needs people (of any background) who are agentic and have a high execution velocity and track record....
Hey Vasco! I just read the summary here, so my apologies if I'm missing something important.
I think I disagree with the line of reasoning you're following. For instance, you say:
>>I estimated its HSI has been 0.0292 % as cost-effective as HIPF accounting for effects on the target beneficiaries, and soil ants, termites, springtails, mites, and nematodes.
It would be very surprising to me if we can know with any degree of confidence that exploratory research into highly speculative areas is >3000X more cost effective than Shrimp Welfare Project's main program—a program which seems to be one of the most cost effective on the planet!
I don't doubt that every step in specific chain of reasoning is mostly correct. Rather, I think where I get off the boat here is that I think the uncertainty in these flowthrough effects tends to dominate after a certain point. I get the sense you are willing to take these much more seriously and literally.
Another area of disagreement is the general framing of suggesting that we should always just fund the one most impactful thing. I agree that this makes sense on the margin, but I expect we're generally more likely to get to a better world in the end if we take a portfolio approach and fund and encourage lots of projects that pass a certain bar. A large part of this is driven by the fact that you get seriously diminishing returns on projects (e.g. I'd be surprised if CEARCH could cost effectively deploy SWP's full annual budget).
So I'd probably frame all of this as more "Both CEARCH's program and SWP's program perform very well on these metrics, and we should consider funding them."
Thanks for the comment, Haven.
You say SWP's HSI "seems to be one of the most cost effective [programs] on the planet". Is this mostly because it has helped lots of shrimp per $ (I estimated 15 k shrimp per $)? If so, and your best guess is that soil animals have negative lives in expectation, you should consider CEARCH's HIPF much more cost-effective because it affects way more animals (I estimated it decreases 5.07 billion soil-animal-years per $)? For SWP's HSI to have increased the welfare of shrimp more cost-effectively than HIPF increases welfare, HSI having helped 1 shrimp would have to be better than HIPF decreasing 338 k soil-animal-years (= 5.07*10^9/(15*10^3)). Assuming that welfare per animal-year is proportional to "number of neurons as a fraction of that of humans"^"exponent of the number of neurons"[1], HIPF is always way more cost-effective than HSI has been.
I am very uncertain about whether soil animals have positive or negative lives, but it is common for people to have a best guess that wild invertebrates have negative lives. I supposed the welfare per animal-year of soil ants/termites/springtails/mites/nematodes is -25 % that of fully happy soil ants/termites/springtails/mites/nematodes. I assumed this holds for all biomes, but I guess there is variation in reality. Karolina Sarek, Joey Savoie, and David Moss estimated -42 % for the “wild bug” in 2018, which is more negative than what I assumed.
I believe recommending on equal foot many interventions with similar marginal cost-effectiveness would definitely make sense. However, I estimate HSI has been only 0.0292 % as cost-effective as HIPF, which is a very large difference. I suspect we mostly disagree about the differences in cost-effectiveness, not about what to recommend conditional on the differences in cost-effectiveness.
I describe this formula in the 2nd bullet of the summary here. For an exponent of 0.188, the formula explains 78.6 % of the variance in the welfare ranges in Bob Fischer's book about comparing welfare across species. The results I presented in the top post are for my preferred exponent of 0.5, which results in a welfare per soil-animal-year much closer to 0 than the exponent of 0.188 that explains very well the welfare ranges in Bob's book.