I am a generalist quantitative researcher. I am open to volunteering and paid work. I welcome suggestions for posts. You can give me feedback here (anonymously or not).
In this post, "welfare range as a fraction of that of humans" means "individual welfare per fully-healthy-animal-year as a fraction of the welfare per fully-healthy-human-year".
I don't think you can extrapolate standard population models past the invention of artificial wombs and robo-nannies. The "Recovery" premise is much more plausible if creating new lives is less personally costly.
Why would artificial wombs and robo-nannies have very different effects from other innovations that have decreased the time cost of having children (for example, washing machines), or increased the income of prospective parents (for example, increasing income of women)? A lower cost of having children as a fraction of available income would increase fertility if the cost of everything else remained the same, but alternatives would become cheaper too relative to available income. It is unclear to me whether your point strengthens or weakens the conclusions of Alexandrie and Eden (2025) under the population models they analysed.
Drone replacements for pollination services. Bring down the cost of micro-drones that can pollinate almond trees down to be cost-competitive with bee-based pollination services, potentially ending an industry that harms billions of bees each year.
Are you assuming the bees would prefer not existing? I suppose this is very uncertain.
Of course, one could always try to argue on the merits that standard EA views on both global poverty and AI safety are empirically mistaken. That’s fine. All I’m saying here is that one had better not endorse higher-level heuristics, applied prior to any engagement with the specific details, that rule out both rigorously-supported global health charities (simply for being insufficiently ambitious) and speculative longtermist longshots (simply for being insufficiently rigorous). After all, if you prioritize rigor, the prima facie case for global health charities is going to be hard to beat. And if you prioritize reach (potential impact), nobody else is even in the same ballpark as the longtermists.
Are you confident the longtermist interventions you have in mind, like work on AI safety, are more cost-effective accounting for longterm effects than global health interventions? Decreasing the risk of human extinction may be less cost-effective than global health interventions accounting for longterm effects hold under standard population models. So it is unclear to me whether there is the rigor-reach dilemma you describe.
I can even see global health interventions decreasing nearterm extinction risk more cost-effectively than AI safety interventions. I am not aware of any quantitative modelling comparing both options.
As I understand your view, something is good if you prefer it and don't expect that preference to change on reflection. That makes "good" relative to each person.
Right. I would just replace "something is good if you prefer it" with "something being morally good means (under your definition) you prefer it". Many people define "morally good" differently, and definitions are not right or wrong, although they can be more or less in agreement with common usage of the words involved. One should just be clear about definitions to avoid confusion, and not pick definitions which are super far from common usage (like defining "good" as something that is strongly dispreferred, which would easily lead to confusion).
That makes "good" relative to each person. So imagine someone who says: "I like eating chickens from factory farms, and I fully expect to keep liking it." By your definition, factory farming is good for them and bad for you, and that's the end of the discussion.
Nitpick. I have little idea if eating chicken increaes or decreases welfare. So it is unclear to me whether it is good or bad.
Someone thinking eating chicken is morally good or bad (under my definitions of these terms) would not necessarily be the end of the discussion. They could be wrong about how their preferences would evolve given new information. They could be very confident chickens do not feel pain, and that they would also not care about decreasing the pain of chickens anyway, but change their minds after investigating the biology and behaviour of chickens, and wacthing footage of broiler farms.
Why should they care about your preference, any more than you should care about theirs?
They may care more about my preferences if their preferences have been approaching my preferences. Say they initially did not care at all about chickens, but now care about them as I predicted they would if they reflected on it. Suppose they still do not care about fishes. They may trust themselves less than in the past. They may know they were wrong about which preferences about chickens they would have on reflection, that I correctely predicted how their preferences about chickens would evolve, and that I care about fishes.
Some people could continue not caring at all about farmed animals after going through all the relevant evidence. However, I believe they would be a very small fraction of the population, and I do not think engaging with them would be a cost-effective way of increasing the welfare of farmed animals.
You said long-term outcomes don't have to be good, but on your view that only means they don't have to be good by your lights.
Right.
What would make the majority wrong, rather than just different from you?
The majority could be morally wrong by their own lights due to not accurately predicting how their preferences would evolve if they reflected more them. Assuming their preferences were very stable, they would be morally right by their own lights, but they would still be morally wrong by my own lights, by which I just mean their preferences on reflection would differ from mine. I understand many people define "morally wrong" differently.
And a related question: chickens and shrimp aren't moral agents. Do their interests count in themselves on your view, or only because moral agents like you happen to care about them?
I would still want chickens and shrimps (and any other organisms) to satisfy their preferences if there were no moral agents.
If donating to the CCF is bad because of risk aversion + substantial off-target effects, donating to pretty much anything is bad, then. Again, here, you're raising a potential problem that goes well beyond the CCF.
I agree. However, to the extent you want to help donors act in agreement with their values, you could recommend they do not donate to anything at all (CCF or other options) if this is the best option given their values. Some could then realise they have different values, and use your tools to clarify what they should donate to given their updated values. Some others could actually endorse not donating, or donating less than they would have donated if they had not learned their donations could cause a lot of harm.
Or they can develop a preference to keep slaves, and keep animals in factory farms, because these preferences were/are shared by so many, how will you defend that they're unethical?
I do not know whether I would defend they are "unethical". It depends on how this is defined. In any case, is the question above relevant in practice? If I cared about decreasing the number of slaves, or farmed animals, I would do whatever achieves this the most cost-effectively instead of focussing on defending that keeping slaves, and farmed animals is unethical. Discussing methaethics would not cost-effectively decrease the number of slaves, and farmed animals because very few people have preferences for keeping slaves, or eating animals for metaethical reasons.
Let's say you have 100 people enslaved, and 10,000 who think it's a brilliant idea — in your theory, do the strongest preferences win?
I am not sure what you mean by "win". I think preferences which are stronger and more prevalent now will tend to be stronger and more prevalent in the future.
you reduced morality to perspective, so in a society you would just accumulate those and assume whatever this accumulation arrived at is good?
Not under my definition of "good". I consider something is good if I prefer it, and do not expect my preference to change upon further reflection. My preferences depend on the preferences of moral agents, as I want to satisfy the preferences of moral agents all else equal. However, my preferences may differ from the longterm preferences of moral agents. So longterm outcomes do not have to be good.
I did mean actually neutral. I assume that our position would be to tentatively assign a value of 0 to effects where any precise number within a range spanning over 0 seems arbitrary, which approximates bracketing out effects we are clueless about. (I know that you are skeptical of bracketing and might say that 0 is just as arbitrary as any other number.)
Say you want to increase expected human welfare, and there is an action i) which has a probability of 100 % of saving 1 human, a probability p of around 50 % of killing 1 billion people, and a probability 1 - p of saving 1 billion people. Would you evaluate this action as an action ii) which has a probability of 100 % of saving 1 human, and no more effects? I would evaluate i) and ii) very differently. I would say i) could easily increase or decrease human welfare a lot, whereas ii) increases it just a little with certainty (a little compared with how much i) can change human welfare). I would focus on decreasing the uncertainty about probability p instead of brackecting out the uncertain effects of i), and decide between i) and nothing as if I was deciding between ii) and nothing.
Moreover, many people care not only about increasing welfare in expectation, but also about not decreasing welfare. For those people, the uncertain effects can matter a lot. They may well prefer i) over nothing, but much prefer nothing over ii). From the point of view of decreasing the chance of causing large harm, funding CCF is much worse than doing nothing? This could be an uncomfortable conclusion, but it seems worth communicating if you want to empower donors to act in agreement with their values.
I think we overall would disagree with the first part, due to being sympathetic to bracketing out effects we are clueless about, or maybe believing these are not that large to begin with. (I am probably the person on the team most sympathetic to deep uncertainty about this, though.)
What do you mean by clueless? "Being clueless about X" is not a mathematical proposition which is either true or false. So I would think one can be more or less clueless about any effects, and I do not understand why some effects would be 100 % neglected. I also do not know why such effects would be effectd on non-target individuals. There are other factors that affect the level of clueless. It generally makes sense to be more clueless about effects happening further into the future. Based on the "More options" drop-down in question 2 of your Donor Compass, you are modelling effects after 500 years. Are you less clueless about these effects than the effects on the population of soil arthropods next year caused by chicken welfare reforms increasing the production of chicken feed? Thereare academic studies on the effects of chicken welfare reforms on agricultural land use, and on the density of soil arthropods by biome. You can combine these with your research on the moral weight of shrimps and black soldier flies (BSFs) to get a sense of the importance of the effects on soil ants and termites. I think you already agree that we have little idea about whether chicken welfare reforms increase or decrease animal welfare due to effects on non-target individuals, but many people believe these effects are negligible.
But fwiw, we are planning on considering including "meta" funds/projects in the allocation, including those that reduce our key uncertainties. (Our key uncertainties may very well differ from yours, though.)
Good to know.
Fwiw, I don't expect "estimating the soil-arthropod-years affected per electrically stunned shrimp" to be the hardest part, relative to figuring out the sign of this + other potential off-target effects of shrimp stunning (and how to weight all this). And I don't expect the CCF team to figure this out any time soon. The AW team might be more likely to make some progress, but I also don't expect it to figure this out.
I agree decreasing i) uncertainty about the direction of the effects on non-target individuals is harder than decreasing ii) uncertainty about their magnitude. However, I think ii) is valuable even without i). If the effects on non-target individuals are much smaller than those on the target individuals, the effects on non-target individuals can be neglected. If the effects on non-target individuals are much larger than those on the target individuals, there would be other interventions targeting the non-target individuals which would be more cost-effective (for example, interventions targeting soil ants and termites instead of chickens). In either case, there would be no need to decrease the uncertainty about the direction of the effects on the non-target individuals.
In this post, "welfare range as a fraction of that of humans" means "individual welfare per fully-healthy-animal-year as a fraction of the welfare per fully-healthy-human-year".
Why would artificial wombs and robo-nannies have very different effects from other innovations that have decreased the time cost of having children (for example, washing machines), or increased the income of prospective parents (for example, increasing income of women)? A lower cost of having children as a fraction of available income would increase fertility if the cost of everything else remained the same, but alternatives would become cheaper too relative to available income. It is unclear to me whether your point strengthens or weakens the conclusions of Alexandrie and Eden (2025) under the population models they analysed.
That makes sense.
Hi Mal.
Are you assuming the bees would prefer not existing? I suppose this is very uncertain.
Hi Richard.
Are you confident the longtermist interventions you have in mind, like work on AI safety, are more cost-effective accounting for longterm effects than global health interventions? Decreasing the risk of human extinction may be less cost-effective than global health interventions accounting for longterm effects hold under standard population models. So it is unclear to me whether there is the rigor-reach dilemma you describe.
I can even see global health interventions decreasing nearterm extinction risk more cost-effectively than AI safety interventions. I am not aware of any quantitative modelling comparing both options.
Relatedly, I liked the post by 50 things I know by Cate Hall.
Right. I would just replace "something is good if you prefer it" with "something being morally good means (under your definition) you prefer it". Many people define "morally good" differently, and definitions are not right or wrong, although they can be more or less in agreement with common usage of the words involved. One should just be clear about definitions to avoid confusion, and not pick definitions which are super far from common usage (like defining "good" as something that is strongly dispreferred, which would easily lead to confusion).
Nitpick. I have little idea if eating chicken increaes or decreases welfare. So it is unclear to me whether it is good or bad.
Someone thinking eating chicken is morally good or bad (under my definitions of these terms) would not necessarily be the end of the discussion. They could be wrong about how their preferences would evolve given new information. They could be very confident chickens do not feel pain, and that they would also not care about decreasing the pain of chickens anyway, but change their minds after investigating the biology and behaviour of chickens, and wacthing footage of broiler farms.
They may care more about my preferences if their preferences have been approaching my preferences. Say they initially did not care at all about chickens, but now care about them as I predicted they would if they reflected on it. Suppose they still do not care about fishes. They may trust themselves less than in the past. They may know they were wrong about which preferences about chickens they would have on reflection, that I correctely predicted how their preferences about chickens would evolve, and that I care about fishes.
Some people could continue not caring at all about farmed animals after going through all the relevant evidence. However, I believe they would be a very small fraction of the population, and I do not think engaging with them would be a cost-effective way of increasing the welfare of farmed animals.
Right.
The majority could be morally wrong by their own lights due to not accurately predicting how their preferences would evolve if they reflected more them. Assuming their preferences were very stable, they would be morally right by their own lights, but they would still be morally wrong by my own lights, by which I just mean their preferences on reflection would differ from mine. I understand many people define "morally wrong" differently.
I would still want chickens and shrimps (and any other organisms) to satisfy their preferences if there were no moral agents.
I agree. However, to the extent you want to help donors act in agreement with their values, you could recommend they do not donate to anything at all (CCF or other options) if this is the best option given their values. Some could then realise they have different values, and use your tools to clarify what they should donate to given their updated values. Some others could actually endorse not donating, or donating less than they would have donated if they had not learned their donations could cause a lot of harm.
I do not know whether I would defend they are "unethical". It depends on how this is defined. In any case, is the question above relevant in practice? If I cared about decreasing the number of slaves, or farmed animals, I would do whatever achieves this the most cost-effectively instead of focussing on defending that keeping slaves, and farmed animals is unethical. Discussing methaethics would not cost-effectively decrease the number of slaves, and farmed animals because very few people have preferences for keeping slaves, or eating animals for metaethical reasons.
I am not sure what you mean by "win". I think preferences which are stronger and more prevalent now will tend to be stronger and more prevalent in the future.
Not under my definition of "good". I consider something is good if I prefer it, and do not expect my preference to change upon further reflection. My preferences depend on the preferences of moral agents, as I want to satisfy the preferences of moral agents all else equal. However, my preferences may differ from the longterm preferences of moral agents. So longterm outcomes do not have to be good.
Say you want to increase expected human welfare, and there is an action i) which has a probability of 100 % of saving 1 human, a probability p of around 50 % of killing 1 billion people, and a probability 1 - p of saving 1 billion people. Would you evaluate this action as an action ii) which has a probability of 100 % of saving 1 human, and no more effects? I would evaluate i) and ii) very differently. I would say i) could easily increase or decrease human welfare a lot, whereas ii) increases it just a little with certainty (a little compared with how much i) can change human welfare). I would focus on decreasing the uncertainty about probability p instead of brackecting out the uncertain effects of i), and decide between i) and nothing as if I was deciding between ii) and nothing.
Moreover, many people care not only about increasing welfare in expectation, but also about not decreasing welfare. For those people, the uncertain effects can matter a lot. They may well prefer i) over nothing, but much prefer nothing over ii). From the point of view of decreasing the chance of causing large harm, funding CCF is much worse than doing nothing? This could be an uncomfortable conclusion, but it seems worth communicating if you want to empower donors to act in agreement with their values.
What do you mean by clueless? "Being clueless about X" is not a mathematical proposition which is either true or false. So I would think one can be more or less clueless about any effects, and I do not understand why some effects would be 100 % neglected. I also do not know why such effects would be effectd on non-target individuals. There are other factors that affect the level of clueless. It generally makes sense to be more clueless about effects happening further into the future. Based on the "More options" drop-down in question 2 of your Donor Compass, you are modelling effects after 500 years. Are you less clueless about these effects than the effects on the population of soil arthropods next year caused by chicken welfare reforms increasing the production of chicken feed? There are academic studies on the effects of chicken welfare reforms on agricultural land use, and on the density of soil arthropods by biome. You can combine these with your research on the moral weight of shrimps and black soldier flies (BSFs) to get a sense of the importance of the effects on soil ants and termites. I think you already agree that we have little idea about whether chicken welfare reforms increase or decrease animal welfare due to effects on non-target individuals, but many people believe these effects are negligible.
Good to know.
I agree decreasing i) uncertainty about the direction of the effects on non-target individuals is harder than decreasing ii) uncertainty about their magnitude. However, I think ii) is valuable even without i). If the effects on non-target individuals are much smaller than those on the target individuals, the effects on non-target individuals can be neglected. If the effects on non-target individuals are much larger than those on the target individuals, there would be other interventions targeting the non-target individuals which would be more cost-effective (for example, interventions targeting soil ants and termites instead of chickens). In either case, there would be no need to decrease the uncertainty about the direction of the effects on the non-target individuals.