Thanks to everyone for the discussion here. A few replies to different strands.
First, I agree with Vasco that transparency matters. However, transparency isn’t the only good—and, unfortunately, it often competes with others. (Time is limited. Optics are complicated. Etc.) So, by Vasco’s own lights, it’s only plausible that organizations should devote scarce resources to answering this particular cause prioritization question—and then post their answer publicly on the Forum—if they think (or should think) that the expected value of so doing is positive. It isn’t obvious that anyone in these organizations thinks (or should think) that’s true.
Second, you can use our work on welfare ranges without buying into naive expected utility maximization. I assume that many people who use our welfare ranges are averse to being mugged and, as a result, adopt one of the many strategies for avoiding that outcome. So, it can be true that (a) the expected value of impacts on some group of animals is very large in expectation and (b) you aren’t rationally required, by your own lights, to care much about that fact (and, by extension, investigate it in depth or engage on it publicly).
Third, our models have a narrow theoretical and pragmatic purpose: we wanted to improve the community’s thinking about cause prioritization regarding a group of animals where we took there to be good evidence of sentience. We don’t think you can take our models and apply them generally, nor do we think you can ignore the specific purpose for which they were developed. Put differently, once some animals have crossed some threshold of plausibility for sentience, we support using our models with trepidation, largely because we don’t have better options. But you shouldn’t apply the model beyond that and, if you have any other principled ways to make decisions, that’s probably better. (Principled: “We think that any theory of change for the smallest animals begins with key victories for larger animals.” Unprincipled: “We don’t like thinking about the smallest animals.”)
Fourth, we disagree with @NickLaing characterization of the Moral Weight Project as stacking the deck in favor of high welfare range estimates. There are two reasons why. One of them is that the MWP does not say, “Sum the number of proxies found for a species and divide by the total number of proxies to get the welfare range.” If that were true, then the number of proxies would straightforwardly determine the maximum difference in welfare ranges. But that isn’t correct. We have models (like the cubic model) where you need to have lots of proxies before you have a “highish” welfare range. However, we have lots of models, with uncertainty across them. Predictably, then, more moderate estimates emerge rather than any extreme (whether high or low). Someone is free to say: “A better methodology wouldn’t have been so uncertain about the models; it would have just included animal-unfriendly options.” That’s clearly tendentious, though, and we think we made the right call in including a wider range of theoretical options. That being said, we’ll reiterate that those who are interested in the details of the project should examine the particulars of each model and its conclusions rather than just taking the overall estimates straightforwardly. You can find each model’s results here.
The second reason we disagree with Nick’s characterization of the MWP is that, even if you isolate a particular model, you don’t automatically get high welfare ranges. Suppose, for instance, that there are 80 proxies total and that a model uses them all. If there were N that were as simple as “any pain-averse behavior,” then, for the core models of the MWP, saying “likely yes” to each of them would give you a sentience-conditioned welfare score of 0.875*N/80 on average. We didn’t consider animals as simple as nematodes in the MWP because we didn’t think that the methods were robust for that type of animal. (See above.) But say you think there’s a 0.5% chance of sentience for nematodes. Then, the sentience-conditioned welfare range would have been approximately 0.005*0.875*N/80. If the average model had 5 proxies that are as simple as “any pain averse behavior” and we gave “likely yes” to nematodes on all five, that would generate a mean welfare range of 0.005*0.875*5/80 = 0.00027. Again, we don’t endorse using the MWP for animals with that small of a probability of sentience, but 0.00027 isn’t a particularly high welfare range. (And as we’ve said many times, we’re just talking about hedonic capacity, not “all things considered moral weights,” which don’t assume hedonism. That number would be lower still.)
Should we find funding for a second version of the project, we’re likely to take a different approach to aggregating the proxies to produce welfare ranges, aggregating welfare ranges across models, and communicating the results. Still, we hope the first version of MWP contributes to more informed and systematic thinking about how to prioritize among different interventions.
I agree transparency is not the only good. However, I think there is a high bar for not commenting on effects which in expectation seem way larger than the effects being covered.
I am not sure what you mean by “naive expected utility maximisation”. I agree my analyses of the effects on soil nematodes, mites, and springtails have lots of room for improvement, but at least I am trying to consider these effects instead of assuming they do not change prioritisation even if they look much larger in expectation than the effects on the target beneficiaries.
I fully endorse expectational total hedonistic utilitarianism (ETHU), but I think this is far from required for effects on soil nematodes, mites, and springtails to matter a lot. These are much larger than the effects on the target beneficiaries under RP’s mainline welfare ranges, so they would still be driver of the overall effect even putting just 10 % weight on ETHU.
What is so different between RP’s probabilities of sentience of nematodes and silkworms of 6.8 % and 8.2 %? RP seems quite worried about farming black soldier fly larvae and mealworms, which I guess are roughly as likely to be sentient as silkworms, and therefore only 1.20 (= 0.082/0.068) times as likely to be sentient as nematodes by RP’s own lights. Why is the methodoly used to obtain RP’s mainline welfare ranges supposed to apply to many invertebrates, but not necessarily to nematodes, mites, and springtails?
I think the following point from @NickLaing is spot on. With the methodology used to obtain RP’s mainline welfare ranges, “if a creature have any pain averse behavior (like just withdrawing from anything), it is guaranteed a highish welfare range”. A single behavioural proxy likely to be absent, meaning 12.5 % (= (0 + 0.25)/2) likely to be present, implies a welfare range conditional on sentience, and the rate of subjective experience of humans under the pleasure-and-pain-centric model of at least 0.00339 (= 0.125/37) regardless of the simplicity of the organism (including bacteria).
I noted the sheet I just linked above is no longer public. I encourage you to make it public again such that people can examine the assumptions underlying RP’s mainline welfare ranges.
You say “0.00027 isn’t a particularly high welfare range [of nematodes]”, but it is 41.7 (= 2.7*10^-4/(6.47*10^-6)) times myestimate, and this already implies the expected effects on soil nematodes are way larger in expectation that those on the target beneficiaries.
Thanks for all your work on comparing welfare across species. I have found it super valuable!
We didn’t do the welfare range calculations for plants, protists, nematodes, etc, because we don’t think the methodology is appropriate for organisms that lack a complex brain and/or nervous system. There are a lot of methodological complexities with even applying them to complex farmed animals like chickens, and if we were to try to do something similar for very simple organisms, we might take a quite different approach.
We don’t really put much stock in the probability of sentience estimates, which weren’t the focus of the project and are subject to much more uncertainty than the welfare range estimates themselves conditional on sentience (which themselves are highly uncertain). If you read the welfare range report’s footnotes, you’ll find that the 6.8% probability of sentience estimate for c elegans is driven substantially by my interpretation there that “probably not sentient” meant 10-35%, which was really just an off-the-cuff judgment. The other people whose views were included in that assessment gave under 1% or under 2% probabilities of sentience, and updating based on the proxies didn’t budge the priors much. On reflection, I think lower numbers are more appropriate than 6.8%, and I really would not anchor on that as “RP’s own lights”.
I think part of this stems from a misunderstanding about the spreadsheets that I mistakenly linked to in the welfare range report. The vast majority of the calculations in the spreadsheet you were working off of were from a very early draft of the project, before we had ironed out a methodology and which animals we thought the methodology could apply to. Since they were a first draft and lack the full context of decisions we made along the way, I really would not consider them as our official position. I am sorry for any confusion that may have caused with respect to our methodology, opinions, or the scope of the project. Here are updated tables containing the proxies: Public Welfare Range Data and Public sentience table (Though, please note that the sentience proxies do very little work at all in the sentience probability assessments, which, again, we don’t put that much stock in, particularly for simple animals)
On the 0.00027 welfare range being high: 1) this was just an example to illustrate that Nick isn’t correct about the structure of the model guaranteeing high numbers, not to show that it’s a suitable welfare range estimate for nematodes per se. We’re not claiming that it’s actually what we would arrive at if we did assess nematodes under a more appropriate framework. And 2) it’s only high if you think you can multiply very small numbers by very big numbers and then act on that, which is a separate point.
I think it’s fine if you or others have a different approach to weighing lean/likely no proxies, that was a judgment call. All of the code is public if you’d like the run it, and I created the ability for you to weigh likely/lean nos differently. That being said, they’re not creating very high estimates for many animals because there were relatively few “lean/likely no” judgments, we have many more models than just the pain/pleasure model that give lower welfare ranges, and we were quite conservative by assuming that all “Unknown” proxies were in fact absent. We’d love to have the chance to come up with new models using a more Bayesian framework, and in doing so, we might make different choices. But the point still holds that the models currently do not guarantee high welfare ranges.
Speaking personally (though I know some others on the team agree), I also reject approaches to meta-normative uncertainty that can easily lead you to be dominated by one fanatical theory. If you resolve meta-normative uncertainty by maximizing choiceworthiness, you’re equally susceptible to Pascal’s mugging. So, if (like me) you don’t want to go all-in on expected value maximization because of the Pascal’s mugging worry, you aren’t going to accept strategies for resolving meta-normative uncertainty that recreate that exact problem. In this case, then, the argument that we should still think that nematode welfare dominates our calculations even if we put a small credence on total hedonic utilitarianism doesn’t move me that much.
Overall, I encourage you and others on the EA Forum to not view our first version of the welfare range estimates as our final word on this. The book version, Weighing Animal Welfare, is more systematic, and we hope to improve on the methods in the future. But even still, I don’t think that the original version commits one to the view that very simple animals should dominate our calculations absent other highly controversial normative and meta-normative assumptions.
Could you clarify why your methodology is supposed to apply to silkworms, but not nematodes? I agree nematodes have a less complex brain and nervous system, but silkworms are also less complex than other animals, so I do not understand how you are deciding when your methodology is supposed to be applicable. Did you preregister the animals to which you thought your methods should apply to?
My understanding is that estimating RP’s mainline welfare ranges involved tens of judgement calls similar to the one you made to get the probability of sentience of nematodes. My broader point is that I do not know what majorly distinguishes nematodes from silkworms for you to think only the latter are reasonably covered by your methodology.
I had understood the estimates in the sheet which is no longer public were preliminary. However, it is still the case that the welfare range conditional on sentience under the pleasure-and-pain-centric model is proportional to the sum of the probabilities of the respective proxies being present? If so, for RP’s probability of sentience of nematodes of 6.8 %, a single behavioural proxy likely to be absent results in a welfare range of nematodes conditional on the rate of subjective experience of humans of at least 2.31*10^-4 (= 0.068*0.00339), which both me and Nick consider high.
The conversion from qualitative probabilistic descriptions to probabilities adds uncertainty, but I do not think it the driver of disagreement. In my mind, and I guess Nick’s, the major issue is that the effect of the presence of behaviours on the welfare range is not moderated by neural complexity. RP’s mainline welfare ranges consider “one-ninth weight to the possibility that an organism’s welfare range [conditional on sentience, and the rate of subjective experience of humans] is equal to the number of neurons it possesses relative to humans”. So an organism having 0 neurons only decreases its welfare range conditional on sentience, and the rate of subjective experience of humans by 1⁄9. I understand having no neurons at all would also lead to a lower probability of sentience, but I think it should directly imply a much larger decrease in the welfare range conditional on sentience.
What is your best guess for the probability of sentience of nematodes? It could be lower than 6.8 %, but still very far from Pascalianly low. I think reasonable approaches to deal with meta-normative uncertainty (about how to aggregate the recommendations of different moral theories) should not dismiss a 6.8 % or slightly lower chance of causing huge amounts of suffering.
In the book, “10 percent [weight is assined] to the equality model”, under which the welfare range conditional on sentience is 1. So the final welfare ranges conditional on sentience are at least 0.1 (= 0.1*1). Do you endorse the estimates presented in Table 8.6 of the book over RP’s mainline welfare ranges?
When Bob was selecting the species, he was thinking of adult insects as the edge cases for the model (bees, BSF). He included juveniles to see what the model implies, not because he really thought the model should be extended to them. You’ll notice that, in the book, the species list narrows considerably partly for this reason.
On the points related to sentience-conditioned welfare ranges, e.g. “So an organism having 0 neurons only decreases its welfare range conditional on sentience, and the rate of subjective experience of humans by 1⁄9. I understand having no neurons at all would also lead to a lower probability of sentience, but I think it should directly imply a much larger decrease in the welfare range conditional on sentience.” I think it’s a mistake to point to a hypothetical sentience-conditioned welfare range, which is an intermediate step in the calculations, for an animal that has zero neurons as indicative of an issue with the methodology overall for animals with complex brains.
Put straightforwardly, if an animal has zero neurons, it would have a welfare range of 0 overall, because I would give it a zero percent chance of being sentient, which affects all the models.
I also am not going to put a precise probability of sentience on nematodes, but I do think it’s much much closer to zero and crosses the threshold of being Pascal’s mugged.
I’m finding these discussions very draining and not productive at this point, so will not be engaging further in this debate.
I encourage you to disclaim in the post with RP’s mainline welfare ranges that Bob does not think the methodology used to produce them is applicable to silkworms. In practice, what does this mean? Would it be reasonable to neglect beings to which your methodology is not supposed to apply? Why is the methodology applicable to black soldier flies (BSFs), but not silkworms? I understand a methodology can be more or less applicable, but I still do not understand which concrete criteria you are using. I also think the applicability of the methodology should ideally be taken into account in the estimates such that these are more comparable.
I suggest people account for the lower applicability of your methodology to less complex organisms by using welfare ranges equal to the geometric mean between RP’s mainline welfare ranges, and the number of neurons as a fraction of that of humans. Does this seem reasonable?
I am not certain that neurons are required for an organism to have a non-constant welfare, so I think organisms without neurons have welfare ranges above 0. I guess you mean that organisms without neurons have a welfare range of roughly 0, but exactly how close to 0 matters. As I say in the post, “Rounding to 0 a probability of sentience, or welfare per animal-year close to 0 introduces an infinite amount of scope insensitivity. Regardless of the number of beings affected, the change in their welfare will be estimated to be exactly 0”.
Could you elaborate on why you seem to believe the probability of sentience of nematodes is Pascalianly low, and therefore arguably much lower than RP’s mainline estimate of 6.8 %? I feel like one can reasonably argue from this that the probability of sentience of silkworms is also Pascalianly low, and therefore not worry about improving the conditions of BSFs and mealworms, which RP estimates will be 417 billion in 2033.
Feel free to follow up later if you are finding this discussion draining, and not productive. I think it would be good for RP to write a post clarifying the extent to which the methodology used to produce RP’s mainline welfare ranges apply to the animals covered and not covered, and why.
If you read the welfare range report’s footnotes, you’ll find that the 6.8% probability of sentience estimate for c elegans is driven substantially by my interpretation there that “probably not sentient” meant 10-35%, which was really just an off-the-cuff judgment. The other people whose views were included in that assessment gave under 1% or under 2% probabilities of sentience, and updating based on the proxies didn’t budge the priors much. On reflection, I think lower numbers are more appropriate than 6.8%, and I really would not anchor on that as “RP’s own lights”.
I would say the probability of sentience of nematodes is higher than 6.8 %. From Andrews (2024):
Given the determinate development of their nervous systems, 30-some years ago it was taken as given that C. elegans are too simple to learn. However, once researchers turned to examine learning and memory in these tiny animals, they found an incredible amount of flexible behavior and sensitivity to experience. C. elegans have short-term and long-term memory, they can learn through habituation (Rankin et al., 1990), association (Wen et al., 1997), and imprinting (Remy & Hobert, 2005). They pass associative learning tasks using a variety of sensory modalities, including taste, smell, sensitivity to temperature, and sensitivity to oxygen (Ardiel & Rankin, 2010). They also integrate information from different sensory modalities, and respond differently to different levels of intoxicating substances, “support[ing] the view that worms can associate a physiological state with a specific experience” (Rankin, 2004, p. R618). There is also behavioral evidence that C. elegans engage in motivational trade-offs. These worms will flexibly choose to head through a noxious environment to gain access to a nutritious substance when hungry enough (Ghosh et al., 2016)—though Birch and colleagues are not convinced this behavior satisfies the marker of motivational trade-offs because it appears that one reflex is merely inhibiting another (Birch et al., 2021, p. 31).
C. elegans are a model organism for the study of nociceptors, and much of what we now know about the mechanisms of nociception comes from studies on this species (Smith & Lewin, 2009). Behavioral responses to noxious stimuli are modulated by opiates, as demonstrated by a study finding that administration of morphine has a dose-dependent effect on the latency of response to heat (Pryor et al., 2007). And, perhaps surprisingly, when the nerve ring that comprises the C. elegans brain was recently mapped, researchers found that different regions of the brain support different circuits that route sensory information to another location where they are integrated, leading to action (Brittin et al., 2021).
Even if we grant the author’s low confidence in nematodes’ having marker five (motivational trade-offs), current science provides ample confidence that nematodes have markers one (nociceptors), two (integrated brain regions), four (responsiveness to analgesics), and seven (sophisticated associative learning). Given high confidence that nematodes have even three of these markers, the report’s methodology [Birch et al. (2021)] would have us conclude that there is “substantial evidence” of sentience in nematodes.
Love points one to three from Bob! Perhaps unsurprisingly once he starts disagreeing with me I have some issues.
1. I think I’ve been misrepresented somewhat. I never claimed that the moral weights project did this “Sum the number of proxies found for a species and divide by the total number of proxies to get the welfare range.”
What I said in the comment was “BOTH their sentience ranges and their behavior scores rely heavily on the presence of pain response behavior”.
And In a previous post I did comment that Median final welfare ranges are fairly well approximated by the simple formula Behavioural Proxy sum x Sentience (see graph).
So indeed headline numbers did actually turn out pretty close to the rsult your statement..… “If that were true, then the number of proxies would straightforwardly determine the maximum difference in welfare ranges” . I might be misinterpreting what you mean by this though.
(Behavioral proxy percent) x (Probability of Sentience) = Median Welfare range
2. I stand by (for the moment) my opinion that both the behavioral proxies and Sentient probabilities DO seem to guarantee pretty high final moral weight numbers, although we all will have very different opinions on what ‘high’ means.
I don’t understand how you chose the 0.5% chance of sentience for your low-end calculation? Its far lower than any number in your model Thelowest number in your sentience modelling for a nematode is 6.8%, and the silkworm which was included in the MWP is 8.5%. Why pick a number for the example 13x lower than you model actually generated? The 6.8% number from your model would bring a calculation of more like 0.068x0.875x 5⁄80 which equals 0.0037, or 0.37% as a low end number. This by my lights at least isn’t a very low baseline moral weight, but I understand if some would consider that a decently low baseline.
I agree that you have individual models with a low baseline but I’m discussing your overall process. Using your original overall process I still think that high numbers are guaranteed. Also if your method decides to combine bunch of models where some of them are close to P = 1, balanced with other models which are P=0.00001, then you’re going to get something in-the-middle-ish (say 0.2-0.8) which also seems high to me.
Also as a side note (less important) I think that 5⁄80 for behavioural proxies is pretty hard to get for anything that moves around. Anything that has evolved to move is likely from an evolutionary standpoint to be attracted to things, withdraw from things and have some kind of way to remember that—otherwise they wouldn’t survive. Maybe that does mean that anything that has evolved to move has a high chance of being sentient though, its an interesting question I know has been discussed before (Can’t remember where).
I was surprised to hear “We don’t really put much stock in the probability of sentience estimates, which weren’t the focus of the project and are subject to much more uncertainty than the welfare range estimates themselves conditional on sentience”. Given that the sentience number is half the final calculation for your headline numbers, which are used freely and widely for expected value calcluations, the sentience number seems pretty important. It also does seem like you put a lot of work into estimating them. Given this statement “On reflection, I think lower numbers are more appropriate than 6.8%, and I really would not anchor on that as “RP’s own lights” I wonder whether reasonable options might be
1. Review the sentience numbers from the project and adjust them to where your thinking is now 2. Not publish a sentience-adjusted moral weight—Instead publish your unadjusted welfare ranges and let people choose their own best-guess sentience multiplier.
But you shouldn’t apply the model beyond that and, if you have any other principled ways to make decisions, that’s probably better. (Principled: “We think that any theory of change for the smallest animals begins with key victories for larger animals.” Unprincipled: “We don’t like thinking about the smallest animals.”)
Assuming i) expectational total hedonistic utilitarianism, ii) nematodes are sentient, and iii) other animals are not sentient, would you still “think that any theory of change for the smallest animals begins with key victories for larger animals” is principled? I only endorse i), not ii) and iii). However, I think I should act roughly as if ii) and iii) are true, as Iestimate effects on nematodes account for the vast majority of the effects of interventions on animals.
Thanks to everyone for the discussion here. A few replies to different strands.
First, I agree with Vasco that transparency matters. However, transparency isn’t the only good—and, unfortunately, it often competes with others. (Time is limited. Optics are complicated. Etc.) So, by Vasco’s own lights, it’s only plausible that organizations should devote scarce resources to answering this particular cause prioritization question—and then post their answer publicly on the Forum—if they think (or should think) that the expected value of so doing is positive. It isn’t obvious that anyone in these organizations thinks (or should think) that’s true.
Second, you can use our work on welfare ranges without buying into naive expected utility maximization. I assume that many people who use our welfare ranges are averse to being mugged and, as a result, adopt one of the many strategies for avoiding that outcome. So, it can be true that (a) the expected value of impacts on some group of animals is very large in expectation and (b) you aren’t rationally required, by your own lights, to care much about that fact (and, by extension, investigate it in depth or engage on it publicly).
Third, our models have a narrow theoretical and pragmatic purpose: we wanted to improve the community’s thinking about cause prioritization regarding a group of animals where we took there to be good evidence of sentience. We don’t think you can take our models and apply them generally, nor do we think you can ignore the specific purpose for which they were developed. Put differently, once some animals have crossed some threshold of plausibility for sentience, we support using our models with trepidation, largely because we don’t have better options. But you shouldn’t apply the model beyond that and, if you have any other principled ways to make decisions, that’s probably better. (Principled: “We think that any theory of change for the smallest animals begins with key victories for larger animals.” Unprincipled: “We don’t like thinking about the smallest animals.”)
Fourth, we disagree with @NickLaing characterization of the Moral Weight Project as stacking the deck in favor of high welfare range estimates. There are two reasons why. One of them is that the MWP does not say, “Sum the number of proxies found for a species and divide by the total number of proxies to get the welfare range.” If that were true, then the number of proxies would straightforwardly determine the maximum difference in welfare ranges. But that isn’t correct. We have models (like the cubic model) where you need to have lots of proxies before you have a “highish” welfare range. However, we have lots of models, with uncertainty across them. Predictably, then, more moderate estimates emerge rather than any extreme (whether high or low). Someone is free to say: “A better methodology wouldn’t have been so uncertain about the models; it would have just included animal-unfriendly options.” That’s clearly tendentious, though, and we think we made the right call in including a wider range of theoretical options. That being said, we’ll reiterate that those who are interested in the details of the project should examine the particulars of each model and its conclusions rather than just taking the overall estimates straightforwardly. You can find each model’s results here.
The second reason we disagree with Nick’s characterization of the MWP is that, even if you isolate a particular model, you don’t automatically get high welfare ranges. Suppose, for instance, that there are 80 proxies total and that a model uses them all. If there were N that were as simple as “any pain-averse behavior,” then, for the core models of the MWP, saying “likely yes” to each of them would give you a sentience-conditioned welfare score of 0.875*N/80 on average. We didn’t consider animals as simple as nematodes in the MWP because we didn’t think that the methods were robust for that type of animal. (See above.) But say you think there’s a 0.5% chance of sentience for nematodes. Then, the sentience-conditioned welfare range would have been approximately 0.005*0.875*N/80. If the average model had 5 proxies that are as simple as “any pain averse behavior” and we gave “likely yes” to nematodes on all five, that would generate a mean welfare range of 0.005*0.875*5/80 = 0.00027. Again, we don’t endorse using the MWP for animals with that small of a probability of sentience, but 0.00027 isn’t a particularly high welfare range. (And as we’ve said many times, we’re just talking about hedonic capacity, not “all things considered moral weights,” which don’t assume hedonism. That number would be lower still.)
Should we find funding for a second version of the project, we’re likely to take a different approach to aggregating the proxies to produce welfare ranges, aggregating welfare ranges across models, and communicating the results. Still, we hope the first version of MWP contributes to more informed and systematic thinking about how to prioritize among different interventions.
Thanks for the comment, Bob!
I agree transparency is not the only good. However, I think there is a high bar for not commenting on effects which in expectation seem way larger than the effects being covered.
I am not sure what you mean by “naive expected utility maximisation”. I agree my analyses of the effects on soil nematodes, mites, and springtails have lots of room for improvement, but at least I am trying to consider these effects instead of assuming they do not change prioritisation even if they look much larger in expectation than the effects on the target beneficiaries.
I fully endorse expectational total hedonistic utilitarianism (ETHU), but I think this is far from required for effects on soil nematodes, mites, and springtails to matter a lot. These are much larger than the effects on the target beneficiaries under RP’s mainline welfare ranges, so they would still be driver of the overall effect even putting just 10 % weight on ETHU.
What is so different between RP’s probabilities of sentience of nematodes and silkworms of 6.8 % and 8.2 %? RP seems quite worried about farming black soldier fly larvae and mealworms, which I guess are roughly as likely to be sentient as silkworms, and therefore only 1.20 (= 0.082/0.068) times as likely to be sentient as nematodes by RP’s own lights. Why is the methodoly used to obtain RP’s mainline welfare ranges supposed to apply to many invertebrates, but not necessarily to nematodes, mites, and springtails?
I think the following point from @NickLaing is spot on. With the methodology used to obtain RP’s mainline welfare ranges, “if a creature have any pain averse behavior (like just withdrawing from anything), it is guaranteed a highish welfare range”. A single behavioural proxy likely to be absent, meaning 12.5 % (= (0 + 0.25)/2) likely to be present, implies a welfare range conditional on sentience, and the rate of subjective experience of humans under the pleasure-and-pain-centric model of at least 0.00339 (= 0.125/37) regardless of the simplicity of the organism (including bacteria).
I noted the sheet I just linked above is no longer public. I encourage you to make it public again such that people can examine the assumptions underlying RP’s mainline welfare ranges.
You say “0.00027 isn’t a particularly high welfare range [of nematodes]”, but it is 41.7 (= 2.7*10^-4/(6.47*10^-6)) times my estimate, and this already implies the expected effects on soil nematodes are way larger in expectation that those on the target beneficiaries.
Thanks for all your work on comparing welfare across species. I have found it super valuable!
Hi Vasco,
I just want to make a few points:
We didn’t do the welfare range calculations for plants, protists, nematodes, etc, because we don’t think the methodology is appropriate for organisms that lack a complex brain and/or nervous system. There are a lot of methodological complexities with even applying them to complex farmed animals like chickens, and if we were to try to do something similar for very simple organisms, we might take a quite different approach.
We don’t really put much stock in the probability of sentience estimates, which weren’t the focus of the project and are subject to much more uncertainty than the welfare range estimates themselves conditional on sentience (which themselves are highly uncertain). If you read the welfare range report’s footnotes, you’ll find that the 6.8% probability of sentience estimate for c elegans is driven substantially by my interpretation there that “probably not sentient” meant 10-35%, which was really just an off-the-cuff judgment. The other people whose views were included in that assessment gave under 1% or under 2% probabilities of sentience, and updating based on the proxies didn’t budge the priors much. On reflection, I think lower numbers are more appropriate than 6.8%, and I really would not anchor on that as “RP’s own lights”.
I think part of this stems from a misunderstanding about the spreadsheets that I mistakenly linked to in the welfare range report. The vast majority of the calculations in the spreadsheet you were working off of were from a very early draft of the project, before we had ironed out a methodology and which animals we thought the methodology could apply to. Since they were a first draft and lack the full context of decisions we made along the way, I really would not consider them as our official position. I am sorry for any confusion that may have caused with respect to our methodology, opinions, or the scope of the project. Here are updated tables containing the proxies: Public Welfare Range Data and Public sentience table (Though, please note that the sentience proxies do very little work at all in the sentience probability assessments, which, again, we don’t put that much stock in, particularly for simple animals)
On the 0.00027 welfare range being high: 1) this was just an example to illustrate that Nick isn’t correct about the structure of the model guaranteeing high numbers, not to show that it’s a suitable welfare range estimate for nematodes per se. We’re not claiming that it’s actually what we would arrive at if we did assess nematodes under a more appropriate framework. And 2) it’s only high if you think you can multiply very small numbers by very big numbers and then act on that, which is a separate point.
I think it’s fine if you or others have a different approach to weighing lean/likely no proxies, that was a judgment call. All of the code is public if you’d like the run it, and I created the ability for you to weigh likely/lean nos differently. That being said, they’re not creating very high estimates for many animals because there were relatively few “lean/likely no” judgments, we have many more models than just the pain/pleasure model that give lower welfare ranges, and we were quite conservative by assuming that all “Unknown” proxies were in fact absent. We’d love to have the chance to come up with new models using a more Bayesian framework, and in doing so, we might make different choices. But the point still holds that the models currently do not guarantee high welfare ranges.
Speaking personally (though I know some others on the team agree), I also reject approaches to meta-normative uncertainty that can easily lead you to be dominated by one fanatical theory. If you resolve meta-normative uncertainty by maximizing choiceworthiness, you’re equally susceptible to Pascal’s mugging. So, if (like me) you don’t want to go all-in on expected value maximization because of the Pascal’s mugging worry, you aren’t going to accept strategies for resolving meta-normative uncertainty that recreate that exact problem. In this case, then, the argument that we should still think that nematode welfare dominates our calculations even if we put a small credence on total hedonic utilitarianism doesn’t move me that much.
Overall, I encourage you and others on the EA Forum to not view our first version of the welfare range estimates as our final word on this. The book version, Weighing Animal Welfare, is more systematic, and we hope to improve on the methods in the future. But even still, I don’t think that the original version commits one to the view that very simple animals should dominate our calculations absent other highly controversial normative and meta-normative assumptions.
Thanks for sharing your thoughts, Laura!
Could you clarify why your methodology is supposed to apply to silkworms, but not nematodes? I agree nematodes have a less complex brain and nervous system, but silkworms are also less complex than other animals, so I do not understand how you are deciding when your methodology is supposed to be applicable. Did you preregister the animals to which you thought your methods should apply to?
My understanding is that estimating RP’s mainline welfare ranges involved tens of judgement calls similar to the one you made to get the probability of sentience of nematodes. My broader point is that I do not know what majorly distinguishes nematodes from silkworms for you to think only the latter are reasonably covered by your methodology.
I had understood the estimates in the sheet which is no longer public were preliminary. However, it is still the case that the welfare range conditional on sentience under the pleasure-and-pain-centric model is proportional to the sum of the probabilities of the respective proxies being present? If so, for RP’s probability of sentience of nematodes of 6.8 %, a single behavioural proxy likely to be absent results in a welfare range of nematodes conditional on the rate of subjective experience of humans of at least 2.31*10^-4 (= 0.068*0.00339), which both me and Nick consider high.
The conversion from qualitative probabilistic descriptions to probabilities adds uncertainty, but I do not think it the driver of disagreement. In my mind, and I guess Nick’s, the major issue is that the effect of the presence of behaviours on the welfare range is not moderated by neural complexity. RP’s mainline welfare ranges consider “one-ninth weight to the possibility that an organism’s welfare range [conditional on sentience, and the rate of subjective experience of humans] is equal to the number of neurons it possesses relative to humans”. So an organism having 0 neurons only decreases its welfare range conditional on sentience, and the rate of subjective experience of humans by 1⁄9. I understand having no neurons at all would also lead to a lower probability of sentience, but I think it should directly imply a much larger decrease in the welfare range conditional on sentience.
What is your best guess for the probability of sentience of nematodes? It could be lower than 6.8 %, but still very far from Pascalianly low. I think reasonable approaches to deal with meta-normative uncertainty (about how to aggregate the recommendations of different moral theories) should not dismiss a 6.8 % or slightly lower chance of causing huge amounts of suffering.
In the book, “10 percent [weight is assined] to the equality model”, under which the welfare range conditional on sentience is 1. So the final welfare ranges conditional on sentience are at least 0.1 (= 0.1*1). Do you endorse the estimates presented in Table 8.6 of the book over RP’s mainline welfare ranges?
Hi Vasco,
When Bob was selecting the species, he was thinking of adult insects as the edge cases for the model (bees, BSF). He included juveniles to see what the model implies, not because he really thought the model should be extended to them. You’ll notice that, in the book, the species list narrows considerably partly for this reason.
On the points related to sentience-conditioned welfare ranges, e.g. “So an organism having 0 neurons only decreases its welfare range conditional on sentience, and the rate of subjective experience of humans by 1⁄9. I understand having no neurons at all would also lead to a lower probability of sentience, but I think it should directly imply a much larger decrease in the welfare range conditional on sentience.”
I think it’s a mistake to point to a hypothetical sentience-conditioned welfare range, which is an intermediate step in the calculations, for an animal that has zero neurons as indicative of an issue with the methodology overall for animals with complex brains.
Put straightforwardly, if an animal has zero neurons, it would have a welfare range of 0 overall, because I would give it a zero percent chance of being sentient, which affects all the models.
I also am not going to put a precise probability of sentience on nematodes, but I do think it’s much much closer to zero and crosses the threshold of being Pascal’s mugged.
I’m finding these discussions very draining and not productive at this point, so will not be engaging further in this debate.
Thanks, Laura.
I encourage you to disclaim in the post with RP’s mainline welfare ranges that Bob does not think the methodology used to produce them is applicable to silkworms. In practice, what does this mean? Would it be reasonable to neglect beings to which your methodology is not supposed to apply? Why is the methodology applicable to black soldier flies (BSFs), but not silkworms? I understand a methodology can be more or less applicable, but I still do not understand which concrete criteria you are using. I also think the applicability of the methodology should ideally be taken into account in the estimates such that these are more comparable.
I suggest people account for the lower applicability of your methodology to less complex organisms by using welfare ranges equal to the geometric mean between RP’s mainline welfare ranges, and the number of neurons as a fraction of that of humans. Does this seem reasonable?
I am not certain that neurons are required for an organism to have a non-constant welfare, so I think organisms without neurons have welfare ranges above 0. I guess you mean that organisms without neurons have a welfare range of roughly 0, but exactly how close to 0 matters. As I say in the post, “Rounding to 0 a probability of sentience, or welfare per animal-year close to 0 introduces an infinite amount of scope insensitivity. Regardless of the number of beings affected, the change in their welfare will be estimated to be exactly 0”.
Could you elaborate on why you seem to believe the probability of sentience of nematodes is Pascalianly low, and therefore arguably much lower than RP’s mainline estimate of 6.8 %? I feel like one can reasonably argue from this that the probability of sentience of silkworms is also Pascalianly low, and therefore not worry about improving the conditions of BSFs and mealworms, which RP estimates will be 417 billion in 2033.
Feel free to follow up later if you are finding this discussion draining, and not productive. I think it would be good for RP to write a post clarifying the extent to which the methodology used to produce RP’s mainline welfare ranges apply to the animals covered and not covered, and why.
I would say the probability of sentience of nematodes is higher than 6.8 %. From Andrews (2024):
A few responses to @Bob Fischer and @Laura Duffy
Love points one to three from Bob! Perhaps unsurprisingly once he starts disagreeing with me I have some issues.
1. I think I’ve been misrepresented somewhat. I never claimed that the moral weights project did this “Sum the number of proxies found for a species and divide by the total number of proxies to get the welfare range.”
What I said in the comment was “BOTH their sentience ranges and their behavior scores rely heavily on the presence of pain response behavior”.
And In a previous post I did comment that Median final welfare ranges are fairly well approximated by the simple formula Behavioural Proxy sum x Sentience (see graph).
So indeed headline numbers did actually turn out pretty close to the rsult your statement..… “If that were true, then the number of proxies would straightforwardly determine the maximum difference in welfare ranges” . I might be misinterpreting what you mean by this though.
(Behavioral proxy percent) x (Probability of Sentience) = Median Welfare range
2. I stand by (for the moment) my opinion that both the behavioral proxies and Sentient probabilities DO seem to guarantee pretty high final moral weight numbers, although we all will have very different opinions on what ‘high’ means.
I don’t understand how you chose the 0.5% chance of sentience for your low-end calculation? Its far lower than any number in your model The lowest number in your sentience modelling for a nematode is 6.8%, and the silkworm which was included in the MWP is 8.5%. Why pick a number for the example 13x lower than you model actually generated? The 6.8% number from your model would bring a calculation of more like 0.068x0.875x 5⁄80 which equals 0.0037, or 0.37% as a low end number. This by my lights at least isn’t a very low baseline moral weight, but I understand if some would consider that a decently low baseline.
I agree that you have individual models with a low baseline but I’m discussing your overall process. Using your original overall process I still think that high numbers are guaranteed. Also if your method decides to combine bunch of models where some of them are close to P = 1, balanced with other models which are P=0.00001, then you’re going to get something in-the-middle-ish (say 0.2-0.8) which also seems high to me.
Also as a side note (less important) I think that 5⁄80 for behavioural proxies is pretty hard to get for anything that moves around. Anything that has evolved to move is likely from an evolutionary standpoint to be attracted to things, withdraw from things and have some kind of way to remember that—otherwise they wouldn’t survive. Maybe that does mean that anything that has evolved to move has a high chance of being sentient though, its an interesting question I know has been discussed before (Can’t remember where).
I was surprised to hear “We don’t really put much stock in the probability of sentience estimates, which weren’t the focus of the project and are subject to much more uncertainty than the welfare range estimates themselves conditional on sentience”. Given that the sentience number is half the final calculation for your headline numbers, which are used freely and widely for expected value calcluations, the sentience number seems pretty important. It also does seem like you put a lot of work into estimating them. Given this statement “On reflection, I think lower numbers are more appropriate than 6.8%, and I really would not anchor on that as “RP’s own lights” I wonder whether reasonable options might be
1. Review the sentience numbers from the project and adjust them to where your thinking is now
2. Not publish a sentience-adjusted moral weight—Instead publish your unadjusted welfare ranges and let people choose their own best-guess sentience multiplier.
Thanks for the great points, Nick. Strongly upvoted.
Hi Bob.
Assuming i) expectational total hedonistic utilitarianism, ii) nematodes are sentient, and iii) other animals are not sentient, would you still “think that any theory of change for the smallest animals begins with key victories for larger animals” is principled? I only endorse i), not ii) and iii). However, I think I should act roughly as if ii) and iii) are true, as I estimate effects on nematodes account for the vast majority of the effects of interventions on animals.