Key Takeaways

• Our objective: estimate the welfare ranges of 11 farmed species.
• Given hedonism, an individual’s welfare range is the difference between the welfare level associated with the most intense positively valenced state that the individual can realize and the welfare level associated with the most intense negatively valenced state that the individual can realize.
• Given some prominent theories about the functions of valenced states, we identified over 90 empirical proxies that might provide evidence of variation in the potential intensities of those states. 
• There are many unknowns across many species.
• It’s rare to have evidence that animals lack a given trait.
• We know less about the presence or absence of traits as we move from terrestrial vertebrates to most invertebrates.
• Many of the traits about which we know the least are affective traits.
• We do have information about some significant traits for many animals.

Introduction

This is the second post in the Moral Weight Project Sequence. The aim of the sequence is to provide an overview of the research that Rethink Priorities conducted between May 2021 and October 2022 on interspecific cause prioritization—i.e., making resource allocation decisions across species. The aim of this post is to provide an overview of the Welfare Range Table, which records the results of a literature review covering over 90 empirical traits across 11 farmed species.
 

Motivations

If we want to do as much good as possible, we have to compare all the ways of doing good—including ways that involve helping members of different species. The Moral Weight Project’s assumptions entail that everyone’s welfare counts the same and that all welfare improvements count equally. Still, some may be able to realize more welfare than others. We’re particularly interested in how much welfare different individuals can realize at a time—that is, their respective welfare ranges. An individual’s welfare range is the difference between the best and worst welfare states the individual can realize at a time. We assume hedonism, according to which all and only positively valenced states increase welfare and all and only negatively valenced states decrease welfare. Given as much, an individual’s welfare range is the difference between the welfare level associated with the most intense positively valenced state that the individual can realize and the welfare level associated with the most intense negatively valenced state that the individual can realize. In the case of pigs, for instance, that might be the difference between the welfare level we associate with being fully healthy on a farm sanctuary, on the one hand, and a botched slaughter, on the other.

If there’s variation in welfare ranges across taxa, then there’s variation in the capacities that generate the determinants of welfare. So, if there’s such variation and hedonism is true, then there’s variation in the capacities that generate positively and negatively valenced experiences.

As Jason Schukraft argues, we don’t have any good direct measures of the intensity of valenced states that let us make interspecific comparisons. Indeed, we rely on indirect measures even in humans: behavior, physiological changes, and verbal reports. We can observe behavior and physiological changes in nonhumans, but most of them aren’t verbal. So, we have to rely on other indirect proxies, piecing together an understanding from animals’ cognitive and affective traits or capabilities. The Welfare Range Table includes over 90 such traits: some behavioral, others physiological; some more cognitive, others more affective. Then, it indicates whether the empirical literature provides reason to think that members of 11 farmed species possess those traits. 

Methodology

Our team was composed of three philosophers, two comparative psychologists (one with expertise in birds; another with expertise in cephalopods), two fish welfare researchers, two entomologists, an animal welfare scientist, and a veterinarian. To select proxies for variation in the capacities that generate positively and negatively valenced experiences, we needed to rely on general theoretical considerations. In this case, we relied on theories about the function of valenced experiences—that is, their adaptive value for organisms who have them. 

There are three main theories about the function of valenced experiences. The first is that they allow organisms to represent fitness-relevant information (“That’s good” / “That’s really good” / “That’s bad” / “That’s really bad” / etc.; Cutter & Tye 2011). The second is that they provide a common currency for decision-making (“A is better than B”; Ginsburg & Jablonka 2019). The third is that they facilitate learning (“If X, then A” / “If Y, then B”; Damasio & Carvalho 2013). So, we looked for empirical traits that arguably bear some relation to information representation, decision-making, and learning via valenced experiences. You can find an annotated list of those traits here.

For each taxon / trait pair, we began with searches via Google Scholar. Then, we looked at the references in recent reviews. If none of those sources provides positive or negative evidence, we searched via Web of Science. Then, we broadened the search. If, for instance, the target trait hadn’t been studied in the relevant species, we looked at the family; if the trait hadn’t been studied in the relevant family, we looked at the order. If it seemed plausible that the results of the study generalize to the target species—for example, because of other shared traits or similar ecological niches—then we included those results as well.

Based on the evidence available for each taxon and trait, we made one of five assessments: 

• Likely No (Credence: 0% – 25%)
• Lean No (Credence: >25% – <50%)
• Unknown (Credence: 50%)
• Lean Yes (Credence: >50% – 75%)
• Likely Yes (Credence: >75% – 100%)

These broad credence ranges are about trait possession (do members of the taxon have the trait or not?), not degree of possession (relative to humans, to what degree do the members of this taxon possess this trait?). The assessments are based on the existing published scholarly literature (as of August 2022), with “Unknown” being the default assessment in cases where there was no literature that spoke fairly directly to a given taxon / trait pair.

After completing the literature reviews, we performed two quality checks. First, we checked for consistency. In our initial review, for instance, honey bees got a “lean yes” for transitive inference, whereas pigs got a “lean no.” A closer look revealed that researchers were holding these animals to different standards. (In this particular case, we opted to flip pigs to “lean yes,” as the lower standard seemed reasonable for our purposes.) Second, we commissioned spot checks of 15% of the traits per taxon from someone who had not been involved with the initial literature review for that taxon. Across all the taxon / trait combinations, the spot checkers didn’t flip any positive responses (“likely yes”; “lean yes”) to negative responses (“likely no”; “lean no”) or vice versa. However, they did flip a handful of “unknown” responses to “lean yes.”

Results

The Welfare Range Table is available here. However, the following diagrams provide a more digestible visual representation of most of the information we gathered (excluding physiological proxies):

We should highlight a few key results. First, as these diagrams suggest, there are many unknowns. This shouldn’t be surprising: in many cases, no one has been interested in investigating these traits; in many other cases, no one knows how to investigate them. Moreover, insofar as it’s easier to publish positive results than negative results, academics have an incentive to study traits that they already expect to find. So, the literature is probably biased toward relatively “sure things,” in the interest of guaranteeing that people avoid walking away from their research empty-handed (i.e., without publications). This point is relevant to—and partially explains—the second key result, which is that there are very few negative results.

The table below indicates the percentage of proxies for which we were able to find some information, whether positive or negative. As the table indicates, information of any kind generally declines from terrestrial vertebrates to invertebrates.

As the next table makes clear, many of the traits about which we know the least are affective. 

On one level, this shouldn’t be surprising either, given scientists’ determination not to anthropomorphize animals—i.e., attribute human characteristics to nonhumans. (Unfortunately, scientists seem to worry less about what Franz de Waal calls “anthropodenial”—i.e., “the a priori rejection of shared characteristics between humans and animals.”) Still, these results are concerning insofar as we think that the welfare impacts of valenced states are additive, where the welfare impacts of experiences “stack” on top of one another. On such a view, experiencing maximal pain and shame (for instance) is worse than experiencing maximal pain alone. So, if we don’t know whether animals can experience shame, we don’t know something that bears directly on their welfare ranges.

However, it might not be the case that valenced states are additive: it could be that when you’re experiencing maximal pain, you’re unable to experience anything else, as pain consumes all your attention. If that’s right, then while information about many valenced states might be highly relevant to assessing welfare impacts (how animals are actually faring), it isn’t necessarily important for assessing welfare ranges (how animals could fare), at least if pain states are the most intense negatively valenced states that animals can experience.

That said, it isn’t completely obvious that pain states are the most intense negatively valenced states that animals can experience. So, that provides some motivation for attending to other kinds of negatively valenced states. Moreover, while we’re skeptical that negative experiences are additive, we aren’t certain, so we assign some credence to the additive view. In expectation, then, we take information about the breadth of animals’ emotional lives to be relevant to their welfare ranges.

In any case, one final point to note is that we do have information about some significant traits for many animals, as the following table indicates:

Parental care, for instance, provides some evidence of the range of affective states available to an organism, as does anxiety-like behavior. Multimodal integration is often touted as a key piece of evidence for sentience, as is the capacity to make trade-offs. But both traits are also relevant to welfare ranges insofar as they come in degrees (as suggested by the presence or absence of other traits). 

A clarification and some limitations

To be clear: the Welfare Range Table does not automatically supply welfare range estimates. It’s one thing to collect data that are relevant to welfare ranges; it’s another thing to score and aggregate them. In this post, we’re only discussing the data. We’ll defer the discussion of scoring and aggregation until later in the Moral Weight Project Sequence.

The Welfare Range Table is the first of its kind. Inevitably, then, it has several limits. Among them: 

1. Our searches in related taxa were limited. In the interest of completing the project in a reasonable timeframe, we tried to limit the search for each taxon / trait pair to one hour. In fact, we often spent two or three hours per taxon / trait pair. Still, spending more time would have made it possible to investigate a wider range of related taxa for each taxon / trait pair.
2. The proxy list could be expanded. We opted for an inclusive approach to the proxies. This made the project enormous. Still, there are many other traits that could have been included—and, in some cases, perhaps ought to have been included in a list of this length. 
3. The proxy list could be refined. There are two senses in which the proxy list could be refined. First, some proxies may not be sufficiently informative about animals’ welfare ranges—though, of course, this is hardly obvious, as we don’t have an independent way to assess how informative they are. Second, many of the proxies are relatively coarse-grained. Consider a trait like reversal learning: namely, the ability to suppress a reward-related response, which involves stopping one behavior and switching to another. This trait comes in degrees: some animals can learn to suppress a reward-related response in fewer trials; and, having learned to suppress a reward-related response at all, some can suppress their response more quickly. A more sophisticated version of the table would account for this variation. 
4. Academics are biased toward “Unknown.” We thought it was important to have domain experts review the literature whenever possible. However, domain experts are academics. Academics are socialized into a community where it’s inappropriate to make some positive claim (“Pigs have this trait” or “pigs lack that trait”) without being able to establish that claim to the satisfaction of your peers. There are good reasons to value this socialization in the present case. For instance, it’s difficult to predict which traits an organism will have based on its other traits. Moreover, it’s difficult to predict whether one kind of organism will have a trait because a related kind of organism does. Still, even though the probability ranges we mentioned earlier establish a very low bar for “lean yes” and “lean no” (above and below 50%, respectively), we defaulted to “unknown” when we couldn’t find any relevant literature. Even if our approach is defensible, other reasonable literature reviewers may have had more “lean yes” and “lean no” assessments than we did. 

We could address these issues in the future. With enough time, for example, we could make inferences about trait possession from more distant taxa. Or, we could make efforts to include any neglected high-value proxies, eliminate relatively low-value proxies, and refine the proxies insofar as that’s possible. Finally, we could try using a missing data method to account for systematic trends in the “Unknowns.” 

Conclusion 

The Welfare Range Table is the first attempt to assess a wide range of welfare-relevant traits in such a diverse array of taxa. It represents an initial but significant step toward estimating the differences in animals’ welfare ranges. If it proves to be valuable, then it ought to be extended, as we didn’t examine many important species. While we prioritized many of the most-farmed taxa, there are many others that humans farm in extraordinary numbers. In particular, it would be good to investigate more fishes and invertebrates.

Acknowledgments

This research is a project of Rethink Priorities. It was written by Bob Fischer. Thanks to Jason Schukraft and Adam Shriver for much of the conceptual work behind the project. Thanks to Rachel Norman, Martina Schiestl, Alex Schnell, and Anna Trevarthen for helpful feedback on earlier versions of this post. Thanks to Meghan Barrett, Leigh Gaffney, Michelle Lavery, Rachael Miller, Martina Schiestl, Alex Schnell, and Anna Trevarthen for their extensive literature reviews. Thanks to Jamie Elsey for the visuals. If you’re interested in RP’s work, you can learn more by visiting our research database. For regular updates, please consider subscribing to our newsletter.