I appreciate you writing this up at the top level, since it feels more productive to engage here than on one of a dozen comment threads.
I have substantive and âstylisticâ issues with this line of thinking, which Iâll address in separate comments. Substantively, on the âSuggestionsâ section:
At the very least, I think GiveWell and Ambitious Impact should practice reasoning transparency, and explain in some detail why they neglect effects on farmed animals. By ignoring uncertain effects on farmed animals, GiveWell and Ambitious Impact are implicitly assuming they are certainly irrelevant.
Why? It seems clear that you arenât GiveWellâs target audience. You know that, and they know that. Unless someone gives me a reason to think that Animal Welfare advocates were expecting to be served by GiveWell, I donât see any value in them clarifying something that seems fairly obvious.
In addition, I encourage people there to take uncertainty seriously, and, before significant further investigation, only support interventions which are beneficial in the nearterm accounting for effects on farmed animals.
Unless the differences on human welfare are incredibly narrow or the impacts on animal welfare are enormous, this seems like a very bad idea. In general, donating $100 to a charity with suboptimal impacts on human welfare but improved impacts on animal welfare is going to be strictly worseâfor both human and animal welfareâthan donating $90 to the best human welfare charity and $10 to the best animal welfare charity.
Similarly, investigating the exact size of the effects mostly seems like a waste of time to me. I wrote this up in more detail a few years ago; was addressing a longtermist cluelessness critique but you can pretty much cut/âpaste the argument. To save a click-through, the key passage is:
Similar thoughts would seem to apply to also other possible side-effects of AMF donations; population growth impacts, impacts on animal welfare (wild or farmed), etc. In no case do I have reason to think that AMF is a particularly powerful lever to move those things, and so if I decide that any of them is the Most Important Thing then AMF would not even be on my list of candidate interventions
GiveWell and Ambitious Impact could also offset the nearterm harm caused to farmed animals by funding the best animal welfare interventions. Icalculate these are over 100 times as cost-effective as GiveWellâs top charities ignoring their effects on animals. If so, and some funding from GiveWell or Ambitious Impact is neutral due to negative effects on animals, these could be neutralised by donating less than 1 % (= 1â100) of that funding to the best animal welfare interventions.
Equally, GiveWell or AIMâs donors can offset if they are worried about this. That seems much better than GiveWell making the choice for all their donors.
Stylistically, some commenters donât seem to understand how this differs from a normal cause prioritisation exercise. Put simply, thereâs a difference between choosing to ignore the Drowning Child because there are even more children in the next pond over, and ignoring the drowning children entirely because they might grow up to do bad things. Most cause prioritisation is the former, this post is the latter.
As for why the latter is a problem, I agree with JWSâs observation that this type of âFor The Greater Goodâ reasoning leads to great harm when applied at scale. This is not, or rather should not be, hypothetical for EA at this point. No amount of abstract reasoning for why this approach is âbetterâ is going to outweigh what seems to me to be very clear empirical evidence to the contrary, both within EA and without.
Beyond that issue, itâs pretty easy to identify any person, grant, or policy as plausibly-very-harmful if you focus only on possible negative side effects, so you end up with motivated reasoning driving the answers for what to do.
For example, in this post Vasco recommends:
In addition, I encourage people there to take uncertainty seriously, and, before significant further investigation, only support interventions which are beneficial in the nearterm accounting for effects on farmed animals.
But why stop at farmed animals? What about wild animals, especially insects? What about the long-term future? If taking Expected Total Hedonistic Utilitarianism seriously as Vasco does, I expect these effects to dominate farmed animals. My background understanding is that population increase leads to cultivation of land for farming and reduces wild animal populations and so wild animal suffering quite a bit.. So I could equivalently argue:
In addition, I encourage Vasco to take uncertainty seriously, and, before significant further investigation, only support interventions which are beneficial in the nearterm accounting for effects on wild animals.
These would then tend to be the opposite set of interventions to the prior set. It just goes round and round. I think there are roughly two reasonable approcahes here:
Pick something that seems like a clear good - âsave livesâ, âend factory farmingâ, âsave the worldâ - and try to make it happen without tying yourself into knots about side-effects.
Really just an extension of (1), but if you come across a side effect that worries you, add that goal as a second terminal goal and split your resources between the goals.
By contrast, if your genuine goal is to pick an intervention with no plausible chance of causing significant harm, and you are being honest with yourself about possible backfires, you will do nothing.
thereâs a difference between choosing to ignore the Drowning Child because there are even more children in the next pond over, and ignoring the drowning children entirely because they might grow up to do bad things.
This is a fantastic summary of why I feel much more averse to this argument than to statements like âanimal welfare is more important than human welfareâ (which I am neutral-to-positive on).
Thanks, Karthik. I think not saving distant (not close) drowning children because they may well (not might) do terrible (not bad) things makes much more sense than not saving them because there are more of them. It is quite reasonable to think that not benefiting some people (drowning children) may be for the greater good (if they would cause lots of suffering to farmed animals). In contrast, I do not see how one can justify not decreasing harm just because one could not eliminate all harm.
But why stop at farmed animals? What about wild animals, especially insects? What about the long-term future?
As you said, I strongly endorse expected total hedonistic utilitarianism, so I do think one should consider effects across all time, space and beings. One reason I prefer interventions improving the conditions of farmed animals instead of ones reducing their consumption is that the former have smaller effects on wild animals. Another is that, although I think reducing the consumption of farmed animals is beneficial nearterm (next few years) because farmed animals have negative lives now, it may be harmful longer term (next few decades) if it is sufficiently permanent because farmed animalsâ lives may become positive.
It is super unclear whether wild animals have positive or negative lives, which means the expected impact on them is lower than it would otherwise be if one could more confidently say they are positive or negative. I believe it is way clearer, although not totally clear, that farmed chickens and shrimp in standard conditions have negative lives, because there is data on the time they spend in pain (which I used in my post), which is not the case for the lives of wild arthropods (the most relevant group to assess the effects on wild animals). For chickens in improved conditions, I would say there is room for disagreement about whether they have positive or negative lives.
My background understanding is that population increase leads to cultivation of land for farming and reduces wild animal populations and so wild animal suffering quite a bit.
Relatedly, I estimated the effects on wild animals of saving human lives in a random country of the beneficiaries of GiveWellâs top charities are 1.15 k times as large as the effects on humans. I welcome estimates for the effects on wild animals of interventions improving the conditions of farmed animals, and I am open to changing my prioritisation based on the results.
By contrast, if your genuine goal is to pick an intervention with no plausible chance of causing significant harm, and you are being honest with yourself about possible backfires, you will do nothing.
I think acting with the goal of trying to decrease the chance of harm reduces this in expectation relative to the counterfactual of doing nothing.
Why? It seems clear that you arenât GiveWellâs target audience. You know that, and they know that. Unless someone gives me a reason to think that Animal Welfare advocates were expecting to be served by GiveWell, I donât see any value in them clarifying something that seems fairly obvious.
Many people who donate to GiveWellâs interventions care about animal welfare, often donating to animal welfare interventions at the same time. Some of these people may want to know about harms caused to animals nearterm due to supporting GiveWellâs interventions. Some of these people may even endorse RPâs median welfare ranges, although still support GiveWellâs interventions due to not wanting to maximise impartial welfare. In general, people have complex preferences about their giving, so I think it is better to be transparent instead of assuming no one would care about the additional information.
In general, donating $100 to a charity with suboptimal impacts on human welfare but improved impacts on animal welfare is going to be strictly worseâfor both human and animal welfareâthan donating $90 to the best human welfare charity and $10 to the best animal welfare charity.
I agree. However, it would still be good to go from your 2nd allocation to one where the 10 $ still go to the best animal welfare organisation, but the 90 $ go to an intervention which is more cost-effective than the best human welfare intervention, which may be one global health and development intervention with improved impacts on animals.
Similar thoughts would seem to apply to also other possible side-effects of AMF donations; population growth impacts, impacts on animal welfare (wild or farmed), etc. In no case do I have reason to think that AMF is a particularly powerful lever to move those things, and so if I decide that any of them is the Most Important Thing then AMF would not even be on my list of candidate interventions
I agree with this prioritisation framing, and commented 4 months ago the meat eating problem is mostly a distraction in this sense. However, many people do not think there is a single most important thing, and so may be open to donating to a global health and development interventions with improved impacts on animals even if donating to animal welfare would be more cost-effective. In addition, it still seems worth analysing the meat eating problem to arrive to more accurate beliefs about the world, and because, in some hard to specify way, many value decreasing the probability of causing harm more than prioritising the most cost-effective interventions.
Equally, GiveWell or AIMâs donors can offset if they are worried about this. That seems much better than GiveWell making the choice for all their donors.
GiveWell has made many other choices for all of their donors, and the ones related to how much they value saving lives (as a function of age), and increasing income influence way more money than what would be needed to offset potential negative impacts on animals.
In general, people have complex preferences about their giving, so I think it is better to be transparent instead of assuming no one would care about the additional information.
I think GiveWell is sufficiently transparent hereâits value proposition is that donating a few thousand dollars will, in expectancy, save the life of a child under five in the developing world. Whether or not this is a good thing is largely left as an exercise to the reader. I do not expect GiveWell to do my moral philosophy homework for me.
I also think itâs fairly obvious that people tend to eat meat and cause carbon emissions, that more children in a heavily resource-constrained country means spreading available resources more thinly across the countryâs children, and so on. Because these things are fairly obvious, donors who are concerned about the sign value of the saving-lives output are free to conduct their own research.
If GiveWell dwelled a ton on the upside collateral effects of saving a lifeâsuch as harping on the possibility that the life you can save will cure cancerâthen I would be more favorably inclined to a view that it was inappropriately selective in its presentation of second-order effects.
I appreciate you writing this up at the top level, since it feels more productive to engage here than on one of a dozen comment threads.
I have substantive and âstylisticâ issues with this line of thinking, which Iâll address in separate comments. Substantively, on the âSuggestionsâ section:
Why? It seems clear that you arenât GiveWellâs target audience. You know that, and they know that. Unless someone gives me a reason to think that Animal Welfare advocates were expecting to be served by GiveWell, I donât see any value in them clarifying something that seems fairly obvious.
Unless the differences on human welfare are incredibly narrow or the impacts on animal welfare are enormous, this seems like a very bad idea. In general, donating $100 to a charity with suboptimal impacts on human welfare but improved impacts on animal welfare is going to be strictly worseâfor both human and animal welfareâthan donating $90 to the best human welfare charity and $10 to the best animal welfare charity.
Similarly, investigating the exact size of the effects mostly seems like a waste of time to me. I wrote this up in more detail a few years ago; was addressing a longtermist cluelessness critique but you can pretty much cut/âpaste the argument. To save a click-through, the key passage is:
Equally, GiveWell or AIMâs donors can offset if they are worried about this. That seems much better than GiveWell making the choice for all their donors.
Stylistically, some commenters donât seem to understand how this differs from a normal cause prioritisation exercise. Put simply, thereâs a difference between choosing to ignore the Drowning Child because there are even more children in the next pond over, and ignoring the drowning children entirely because they might grow up to do bad things. Most cause prioritisation is the former, this post is the latter.
As for why the latter is a problem, I agree with JWSâs observation that this type of âFor The Greater Goodâ reasoning leads to great harm when applied at scale. This is not, or rather should not be, hypothetical for EA at this point. No amount of abstract reasoning for why this approach is âbetterâ is going to outweigh what seems to me to be very clear empirical evidence to the contrary, both within EA and without.
Beyond that issue, itâs pretty easy to identify any person, grant, or policy as plausibly-very-harmful if you focus only on possible negative side effects, so you end up with motivated reasoning driving the answers for what to do.
For example, in this post Vasco recommends:
But why stop at farmed animals? What about wild animals, especially insects? What about the long-term future? If taking Expected Total Hedonistic Utilitarianism seriously as Vasco does, I expect these effects to dominate farmed animals. My background understanding is that population increase leads to cultivation of land for farming and reduces wild animal populations and so wild animal suffering quite a bit.. So I could equivalently argue:
These would then tend to be the opposite set of interventions to the prior set. It just goes round and round. I think there are roughly two reasonable approcahes here:
Pick something that seems like a clear good - âsave livesâ, âend factory farmingâ, âsave the worldâ - and try to make it happen without tying yourself into knots about side-effects.
Really just an extension of (1), but if you come across a side effect that worries you, add that goal as a second terminal goal and split your resources between the goals.
By contrast, if your genuine goal is to pick an intervention with no plausible chance of causing significant harm, and you are being honest with yourself about possible backfires, you will do nothing.
This is a fantastic summary of why I feel much more averse to this argument than to statements like âanimal welfare is more important than human welfareâ (which I am neutral-to-positive on).
Thanks, Karthik. I think not saving distant (not close) drowning children because they may well (not might) do terrible (not bad) things makes much more sense than not saving them because there are more of them. It is quite reasonable to think that not benefiting some people (drowning children) may be for the greater good (if they would cause lots of suffering to farmed animals). In contrast, I do not see how one can justify not decreasing harm just because one could not eliminate all harm.
Thanks, Alex.
As you said, I strongly endorse expected total hedonistic utilitarianism, so I do think one should consider effects across all time, space and beings. One reason I prefer interventions improving the conditions of farmed animals instead of ones reducing their consumption is that the former have smaller effects on wild animals. Another is that, although I think reducing the consumption of farmed animals is beneficial nearterm (next few years) because farmed animals have negative lives now, it may be harmful longer term (next few decades) if it is sufficiently permanent because farmed animalsâ lives may become positive.
It is super unclear whether wild animals have positive or negative lives, which means the expected impact on them is lower than it would otherwise be if one could more confidently say they are positive or negative. I believe it is way clearer, although not totally clear, that farmed chickens and shrimp in standard conditions have negative lives, because there is data on the time they spend in pain (which I used in my post), which is not the case for the lives of wild arthropods (the most relevant group to assess the effects on wild animals). For chickens in improved conditions, I would say there is room for disagreement about whether they have positive or negative lives.
Relatedly, I estimated the effects on wild animals of saving human lives in a random country of the beneficiaries of GiveWellâs top charities are 1.15 k times as large as the effects on humans. I welcome estimates for the effects on wild animals of interventions improving the conditions of farmed animals, and I am open to changing my prioritisation based on the results.
I think acting with the goal of trying to decrease the chance of harm reduces this in expectation relative to the counterfactual of doing nothing.
Thanks, Alex.
Many people who donate to GiveWellâs interventions care about animal welfare, often donating to animal welfare interventions at the same time. Some of these people may want to know about harms caused to animals nearterm due to supporting GiveWellâs interventions. Some of these people may even endorse RPâs median welfare ranges, although still support GiveWellâs interventions due to not wanting to maximise impartial welfare. In general, people have complex preferences about their giving, so I think it is better to be transparent instead of assuming no one would care about the additional information.
I agree. However, it would still be good to go from your 2nd allocation to one where the 10 $ still go to the best animal welfare organisation, but the 90 $ go to an intervention which is more cost-effective than the best human welfare intervention, which may be one global health and development intervention with improved impacts on animals.
I agree with this prioritisation framing, and commented 4 months ago the meat eating problem is mostly a distraction in this sense. However, many people do not think there is a single most important thing, and so may be open to donating to a global health and development interventions with improved impacts on animals even if donating to animal welfare would be more cost-effective. In addition, it still seems worth analysing the meat eating problem to arrive to more accurate beliefs about the world, and because, in some hard to specify way, many value decreasing the probability of causing harm more than prioritising the most cost-effective interventions.
GiveWell has made many other choices for all of their donors, and the ones related to how much they value saving lives (as a function of age), and increasing income influence way more money than what would be needed to offset potential negative impacts on animals.
I think GiveWell is sufficiently transparent hereâits value proposition is that donating a few thousand dollars will, in expectancy, save the life of a child under five in the developing world. Whether or not this is a good thing is largely left as an exercise to the reader. I do not expect GiveWell to do my moral philosophy homework for me.
I also think itâs fairly obvious that people tend to eat meat and cause carbon emissions, that more children in a heavily resource-constrained country means spreading available resources more thinly across the countryâs children, and so on. Because these things are fairly obvious, donors who are concerned about the sign value of the saving-lives output are free to conduct their own research.
If GiveWell dwelled a ton on the upside collateral effects of saving a lifeâsuch as harping on the possibility that the life you can save will cure cancerâthen I would be more favorably inclined to a view that it was inappropriately selective in its presentation of second-order effects.