Opinion: Estimating Invertebrate Sentience

by Jason Schukraft23 min read7th Nov 201933 comments

121

Sentience & ConsciousnessRethink PrioritiesWild Animal Welfare
Frontpage

Introduction

Between May 2018 and June 2019 Rethink Priorities completed a large project on the subject of invertebrate sentience.[1] We investigated the best methodology to approach the question, outlined some philosophical difficulties inherent in the project, described the features most relevant to invertebrate sentience, compiled the extant scientific literature on the topic, summarized our results, and ultimately produced an invertebrate welfare cause profile. We are currently in the process of identifying concrete interventions to improve invertebrate wellbeing, with a report on the welfare of managed honey bees due out in mid-November and a report on the welfare of farmed snails nearing completion.

One thing we did not do was publish explicit numerical estimates of the probability that various groups of invertebrates are sentient.

Our team discussed publishing such estimates many times, but these discussions generated considerable internal disagreement. Two members of the (four person) team believed that publishing explicit sentience estimates was a bad idea. The other two members felt that it was a good idea. In the end, we settled on the following compromise: several months after the completion of the invertebrate sentience project, we would publish an unofficial opinion piece in which each of us could share her/his own reasoning on the subject and, if so desired, her/his own estimates.[2]

This post fulfills that compromise. In it, the four members of Rethink Priorities’ invertebrates team—Daniela R. Waldhorn, Marcus A. Davis, Peter Hurford, and Jason Schukraft—outline their views on the value, feasibility, and danger of quantitative estimates of invertebrate sentience. Marcus and Peter provide numerical estimates of sentience for each of the taxa we investigated for our invertebrate sentience project, Daniela offers a qualitative ranking of the same taxa, and Jason argues that we are not yet in a position to deliver estimates that are actionable and robust enough to outweigh the (slight but non-negligible) harm that publishing such estimates prematurely might engender.

What follows are the personal opinions of individual researchers. Officially, Rethink Priorities does not have a position on the explicit probability that various invertebrates are sentient.

Daniela R. Waldhorn

1. Vertebrates

There is an ample and detailed body of empirical data which justifies believing that non-human vertebrates are sentient. In particular, there is solid neuro-anatomical, physiological and behavioral evidence that vertebrates like cows and chickens are conscious. There is also a growing trend to recognize that these animals do not only experience physical suffering (and pleasure) but also have emotional lives (see e.g. Marino, 2017; Proctor et al., 2013).

Based on existing evidence and the generalized acceptance of the Cambridge Declaration on Consciousness (2012)[3], my overall conclusions regarding the probabilities of consciousness for these animals are presented as follows:

2. Invertebrates

When we consider invertebrates, the debate about whether they are conscious becomes much more complex. First, it must be conceded that the numerous invertebrate species and their diversity impose severe constraints to justifiable generalizations about the presence of consciousness in this group of animals. Second, the scientific literature about sentience in invertebrates is not only scarce but fragmentary–that is to say, the extent to which invertebrates have been investigated varies. Thus, there are some particular species about which there is a comparatively great deal of knowledge (e.g., fruit flies), whereas much less research has focused on individuals of other taxa (see our Summary of findings Part 1 and Part 2).

The existing gaps in this field of research entail that we face significant constraints when assessing the probability that an invertebrate taxon is conscious. In my opinion, the current state of knowledge is not mature enough for any informative numerical estimation of consciousness among invertebrates. Furthermore, there is a risk that such estimates lead to an oversimplification of the problem and an underestimation of the need for further investigation. As it was stated in our summary of findings, the available evidence is not enough for determining with sufficient certainty whether most invertebrate taxa are conscious or not. Yet it does allow us to identify certain aspects as potentially relevant for consciousness and flag certain areas that should be further investigated if we want to arrive at more robust conclusions.

Nevertheless, our literature review allows us to conclude, for instance, that there is relatively strong evidence that octopuses are conscious. We can even assert that octopuses are more likely to be conscious than other invertebrates, like earthworms. Hence, although, as stated, this field is not mature enough to make robust estimations of sentience, we can still make some relevant comparisons. Therefore, I propose developing a framework for examining existing evidence for or against consciousness in particular invertebrate taxa.

To contribute to this discussion, here I present a preliminary and generic attempt to such a framework. It consists of five broad categories of answers to the question of whether invertebrate organisms of a particular taxon are sentient or not. These categories are “very probably yes”, “probably yes”, “possibly yes”, “possibly no”, and “probably no”. A sixth category, “very probably no” is later introduced to assess consciousness in other non-invertebrate taxa included for comparison purposes. These six categories range from highest to lowest assigned probabilities of consciousness.

Each category is established by appeal to specific criteria. Hence, existing evidence of consciousness of an invertebrate taxon is assessed in light of those criteria. The different taxa have been classified depending on the degree to which they meet them. These allows for comparisons between these taxa. Additionally, this framework can be useful for assessing new evidence that may update our views, and for studying other invertebrate taxa as well.

It must be stressed that differences between adjacent categories do not necessarily represent equal intervals in the underlying “scale” which gives rise to these groups. However, since the categories are ordered from higher to lesser probability, this framework allows us to claim, for example, that honey bees are more likely to be conscious individuals than jellyfish. This and similar judgements reflect my interpretation of the available evidence in light of the selected criteria. New evidence may change how current knowledge is understood and how organisms of a given taxon are classified. Therefore, the fact that an organism falls into a category should not be interpreted as a definitive answer to that organism’s probabilities of being conscious.

In what follows, I try —to a limited degree— to describe the criteria that define each category. I devote space to elaborate on such criteria because a practical framework for categorizing our findings requires clear and defined indicators. Otherwise, if a criterion is not correctly specified, what counts or not as evidence and what the evidence points to will be subject to the researcher's interpretation. However, please keep in mind that this is a preliminary work, and I am not an expert. Presumably, specialized knowledge, further empirical research and the development of a fundamental theory of consciousness may contribute to improve our conceptual clarity and to better weigh the evidence we possess about consciousness in non-human individuals.

Finally, the fact that several organisms fall into the same category does not necessarily entail that the existing evidence for them is similar. There are cases in which our evidence for some taxa is stronger than for others, even in the same category. To better understand why a specific taxon falls into a given category, see our ‘Invertebrate Sentience Table’ and our Summary of findings Part 2, where it is further described which potentially consciousness-indicating features are found in each taxon.

Very probably yes

I include in this category taxa that meet the following criteria:

  • Given the current state of our knowledge, there is direct evidence that individuals of these taxa exhibit features which, according to expert agreement, seem to be necessary –although not sufficient– for consciousness (Bateson, 1991; Broom, 2013; EFSA, 2005; Elwood, 2011; Fiorito, 1986; Sneddon et al., 2014; Sneddon, 2017). These features are:
    • Neuroanatomical structures and physiological functions, such as nociceptors or equivalent structures, centralized information processing, vertebrate midbrain-like function, and physiological responses to nociception or handling. Additionally, it is expected that conscious individuals have opioid-like receptors and analgesics reduce their nociceptive reflexes and avoidant behaviors;
    • Behavioral responses that are potential indicators of pain experience, such as defensive behavior or fighting back, and moving away from noxious stimuli. These reactions seem to take into account a noxious stimulus’ intensity and direction. Other observed behaviors include pain relief learning, and long-term behavior alteration to avoid a noxious stimulus.
  • In several cases, direct evidence of those features is incomplete. Still, we find an important body of information about other potentially consciousness-indicating features that, broadly, account for the organism’s ability to display complex and flexible behaviors. Additionally, other behavioral indicators suggest that these organisms may experience emotional states and have cognitive skills. This criterion includes:
    • The individual shows different reactions to noxious stimuli, depending on exogenous or endogenous changes (motivational trade-offs);

    • Proxy indicators of memory are observed, such as ‘spatial memory’ and ‘long-term behavior alteration to avoid noxious stimulus (24+ hours)’ (about the importance of memory for assessing consciousness, see Baars (2003) or Stein et al., (2016));

    • There is evidence of additional forms of learning different from habituation, sensitization and mere associative learning. That is to say, the organism can learn in forms that are more likely to require consciousness, such as operant conditioning with an unfamiliar action, observational learning and contextual learning;

    • Individuals may display certain benchmarks of cognitive sophistication, such as tool use[4];

    • As conscious pain is plausibly correlated with certain kinds of behaviors, so, too, we expect that various emotional states are correlated to an array of behavioral indicators. Hence, behaviors that suggest different positive and/or negative mood states are also observed in these organisms;

    • Various forms of navigational skills are identified.

  • In addition, these organisms also respond to several drugs in a manner similar to humans;
  • There may be other behavioral evidence of consciousness, such as goal-directed behavior, communicational behavior and forms of interaction that probably require consciousness;
  • Consistent with the above, there is some agreement among prominent scientists and philosophers that these animals may be conscious.

Given the current evidence I assign a high credence to individuals of the following taxa being sentient:

Recent research has shown that octopuses are highly intelligent and potentially capable of experiencing pain. This has led to the inclusion of octopuses in animal protection legislation in some jurisdictions (e.g. in the European Union, see Directive 2010/63/EU), and their recognition as conscious individuals in the Cambridge Declaration on Consciousness (2012).

Probably yes

In these cases, the defined criteria are:

  • Although incomplete, there is direct evidence that individuals of these taxa exhibit features which, according to expert agreement, seem to be necessary –although not sufficient– for consciousness (Bateson, 1991; Broom, 2013; EFSA, 2005; Elwood, 2011; Fiorito, 1986; Sneddon et al., 2014; Sneddon, 2017) (see the first criterion of the ‘very probably yes’ category);
  • There is an important body of information about other potentially consciousness-indicating features that, broadly, account for the organism’s ability to display complex and flexible behaviors. Additionally, other behavioral indicators suggest that these organisms may experience emotional states and have cognitive skills (see the first criterion of the ‘very probably yes’ category). However, this sort of evidence is not as abundant as in the previous case (‘very probably yes’). Or, for some features, it is discussed whether a specific behavior is a conscious reaction or an automatic response (e.g., autotomy or grooming, as forms of protective behavior). Alternatively, it may happen that available evidence is indicative of consciousness but it is still relatively recent;
  • In addition, these organisms also respond to some drugs in a manner similar to humans;
  • There may be other behavioral evidence of consciousness, such as goal-directed behavior, communicational behavior and forms of social interaction and organization that probably requires consciousness;
  • Hence, some prominent scientists and philosophers claim that these animals may be conscious. However, there does not seem to be as widespread an agreement on this as that observed for the previous category (‘very probably yes’).

Given the current evidence I considered that it is probable that individuals of the following taxa are conscious:

Honey bees display considerable learning abilities, complex social behaviours and even communicational behaviors. They are probably conscious but we have little evidence of the presence of a pain system, if any. Regarding decapod crustaceans (e.g., crabs, crayfish), recent research has shown that they are potentially capable of experiencing pain. According to the European Food Safety Authority (2005), these animals exhibit complex behaviors and should be legally protected. A similar position is held by the British Veterinary Association (2017). Fruit flies, for their part, have been widely used as biological models. New research provides compelling evidence suggesting that they not only feel pain, but also experience chronic pain that lasts long after an initial injury has healed (Khuong et al., 2019). Behaviorally, however, fruit flies do not seem to exhibit such complex reactions as those observed in honey bees or decapods.

Possibly yes

In these cases:

  • Several potentially consciousness-indicating features are found through these taxa, suggesting that these individuals may be sentient. Most of this evidence refers to behavioral observations. These behavioral indicators include:
    • Responses such as moving away, escaping and avoidance, that seem to account for noxious stimuli intensity and direction;
    • Proxy indicators of memory, such as ‘spatial memory’ and ‘long-term behavior alteration to avoid noxious stimulus (24+ hours)’ (about the importance of memory for assessing consciousness, see Baars (2003) or Stein et al., (2016));
    • There is evidence of additional forms of learning different from habituation, sensitization and mere associative learning. That is to say, the organism can learn in forms that are more likely to require consciousness, such as operant conditioning with an unfamiliar action, observational learning and contextual learning:
    • The individual shows different reactions to noxious stimuli, depending on exogenous or endogenous changes (‘motivational trade-offs’);
    • There may be other indirect behavioral evidence of consciousness, such as navigational skills or forms of social interaction and organization that probably require consciousness.
  • However, in these cases, the neuroanatomical structures and neuroanatomical functioning related to valenced experience remain unclear. In general, the physiology or structures that seem to correlate with the presence of sentience (e.g., nociceptors), if they exist, have not been identified yet;
  • Nevertheless, limited evidence of the neurophysiological bases of consciousness should not be confused with their absence. Further studies should assess some open questions about the neuroanatomical and physiological bases of existing behavioral findings.

Hence, the case for the taxa in this category being conscious is not as strongly supported as in previous taxa. I therefore assign a lower credence to their being conscious:

Social ants, and to a lesser extent, spiders (especially, jumping spiders) and cockroaches, show considerable learning abilities and complex social behaviour. The small size of the brain does not necessarily mean poor functioning, since their nerve cells are very small. However, we have little evidence of a pain system (if any). Of these three taxa, the available evidence is weaker for cockroaches.

Possibly no

  • There is limited evidence about features which, according to expert agreement, seem to be necessary –although not sufficient– for consciousness. In particular, some neuroanatomical and physiological features may be observed (e.g., nociceptors). However, there is too little evidence for us to positive conclude that these organisms possess a pain system;
  • Behaviorally, different nociceptive responses may be observed, along with expressions of associative learning. Nevertheless, the existing evidence about motivational tradeoffs, mood state behaviors or cognitive sophistication is highly limited;
  • In general, there is very limited scientific evidence related to the presence of valenced experiences in these individuals. Still, there is more positive than negative evidence about the possibility that these individuals are sentient.

Sea hares (Aplysia) are an example of a taxon falling into this category, as per the above criteria.

Aplysia are the most active marine gastropod molluscs. Much research has been carried out on the nervous system of sea hares and their relatives. Evidence of learning and flexibility of behaviour is considerable but there are also studies showing very rigid responses.

Probably no

  • There is highly restricted evidence of features that –according to expert agreement– seem to be necessary for consciousness. This reduces in a significant manner the likelihood that these animals are sentient;
  • Similarly, information about other anatomical, physiological and behavioral features is scarce. Other potentially indirect evidence of consciousness has not been not observed either;
  • We may not encounter direct negative evidence against any potentially consciousness-indicating features in these organisms, but indirect research suggests that some features may not be present or that there is a low chance of them to be found in these animals;
  • Broadly, these animals’ pain system (if they have any) has not been sufficiently studied. And in spite of certain behavioral evidence, it is hypothesized that observed reactions may be reflexes.

According to these criteria, the following taxa fall into this category:

Regarding earthworms, there is limited research about them, since they are not usually considered sentient organisms. Most of the existing evidence points out that their noxious-stimuli-related responses are highly rigid.** **However, of the three taxa listed in this category, current evidence is slightly stronger for earthworms, especially because of the relative complexity of their nervous system.

In the case of C. elegans, despite being a widely studied animal, evidence of their being conscious is weak and their behavioral reactions are simple and highly stereotyped. Finally, about moon jellyfish, although they display some noxious stimuli reactions and have a nerve net that allows them to detect various stimuli, they do not possess a central nervous system or other equivalent structure that centralizes information processing.

3. Other organisms

As explained in a previous post, we investigated plants (kingdom Plantae), prokaryotes and protists to give a sense of how often potentially consciousness-indicating features are found in organisms that are widely believed to lack consciousness. Regarding these three broad taxa**:**

  • There is no solid evidence that individuals of these taxa exhibit features which, according to expert agreement, seem to be necessary –although not sufficient– for consciousness. In particular:
    • The physiology and structures that seem to correlate invariably with the presence of sentience (e.g., nociceptors) are not observed. No equivalent structures or functioning are found either;
    • Behaviorally, the organism may display some basic noxious stimuli reactions (e.g., moving away), but these responses do not seem to account for noxious stimulus intensity and direction.
  • Similarly, information about other anatomical, physiological and behavioral features is not found. Other potentially indirect evidence of consciousness is not observed either;
  • Consistent with the above, there is a general agreement among the scientific community that these organisms are not conscious.

Following these criteria, plants, prokaryotes and protists are very probably not conscious:

4. Results

The following graph summarizes my previous estimations:

Marcus A. Davis

Estimating the Probability of Sentience

Humanity currently doesn’t understand what fundamentally generates the ability for humans to experience the world. Even were we to gain such an understanding, unless we also gather the ability to understand everything that could plausibly generate such experiences, as opposed to merely what does in humans, it will still be difficult to ascertain if other creatures who have different neuroarchitecture than humans are also experiencing the world. For these reasons and others, I’m very uncertain about the value of providing quantitative estimates for the taxa we’ve studied but still think it’s worth doing so on balance. To me, the case for doing so is based primarily on reasoning transparency as opposed to others using my estimates as placeholders in their calculations. Given the uncertainty around perceptions of invertebrate sentience, for others considering our research and actions it’s useful to see how this project affected, or did not affect, my beliefs and what those beliefs are with respect to invertebrates.

However, there are several other additional qualifications worth attaching to my estimates including the following:

  • I’m by no means an expert on consciousness nor am I an expert on any of the classes of invertebrates we are considering.
  • In general, I don’t believe my estimates are very likely to be well calibrated, and my best guess isn’t necessarily very stable in all cases. Even relative to the difficulties of generally being well-calibrated, I find my subjective estimates likely aren’t very meaningful and I’ve accordingly provided relatively wide estimates for how probable these creatures are to be sentient. Even with this caveat, the range is still more of a guideline for my subjective impression than a declaration of what all agents would estimate given their engagement with the literature.
  • For many groups of invertebrates, the amount that is currently unknown suggests that any attempt to place a particular number is largely a forecast about what future studies may reveal as opposed to reflective of what current studies have revealed.
  • Given the uncertainty, the absolute point estimate I provide of each taxa’s sentience is probably less valuable than the relative rankings of a given taxon compare to each other.
  • Extrapolating from the taxa we examined to other invertebrate taxa, while tempting, is not advisable. I set my estimates for the specific creatures under consideration, and without considering how others might try to extrapolate them to other creatures, even other organisms within the same genus or family we considered.
  • These estimates do not represent the moral weight I believe these creatures would possess provided they are in fact sentient. I’ve not attempted to estimate moral weight for these groups but were I to do so the answer could be affected by a range of features including intelligence, internal clock speed, social complexity, and affective complexity.

Finally, given these considerations, I would add, sticking my specific point estimates (or all of RP’s) directly into a cost-effectiveness analysis is probably foolhardy. Ranges would serve you better, but even then you reduce these estimates to an easy calculation used for expected value at your own hazard.

How I’ve updated since the beginning of this project

Overall, I found some evidence of invertebrate behavior relatively surprising, particularly some responses to drugs, certain tool use, and long-term changes due to learning. Though I didn’t create point estimates for each of these taxa before beginning this project, I suspect the combination of work this area caused me to, at least, update positively for ants, cockroaches, honeybees, fruit flies, and octopuses.

However, perhaps my largest surprise wasn’t an update toward or against a particular type of animal, rather it was based on the extent of conditioned learning behavior that is more or less exhibited by all taxa we considered, including single-celled organisms and animal bodies detached from brain communication, including the lower body of a mouse with a severed spine. While one could take this as weak evidence of widespread sentience, this updated me toward thinking many of these behaviors aren’t very impressive and they were thus largely disregarded in contemplating the positive case for sentience.

Further notes on specific taxa

For chickens and cows, my priors about their probability of sentience remained approximately where it was when we began this project. This isn’t a surprise given they weren’t our focus and I didn’t learn much new information about chicken or cow capacities during our work.

For the remaining non-invertebrate taxa—plants, protists, prokaryotes—they possess particularly weak evidence of sentience, and lack central nervous systems and brains. My estimates reflect this doubt in their sentience. However, one significant issue I encountered was while they vary in abilities and thus there’s a good case a precise and well-calibrated estimate of sentience would vary, generally attempting to estimate the probability that any of these taxa are sentient amounts to an attempt to estimate the probability that more or less everything I understand about sentience is incorrect. Are the odds of such an occurrence one in ten thousand? One in one million? Or, are they one in one hundred billion? I don’t have a relevant reference class to draw from to answer such a question and thus it would likely be more accurate to surmise I take the probability that any of these groups is sentient is vanishingly unlikely and, for all intents and purposes, approximately zero.

Peter Hurford

Some of my colleagues have declined to produce estimates for the likelihood of sentience for various invertebrates, instead suggesting that the interested reader go through our evidence and create their own estimates. I’m somewhat sympathetic to this view - there is a large amount of uncertainty involved and judgement calls must be made on both personal values and weighing the evidence we have. Furthermore, keep in mind that yet to be resolved questions around moral weight, Pascal’s mugging, and generalizability also complicate how these probabilities will be used in practice. Lastly, I don’t want you to overweight my opinions or think of me as some sort of expert where I am not. While I may have my personal probabilities, I respect that there may still be a wide variety of disagreement on these probabilities, implied values, and implied views on how to make decisions based on these. I invite people to come to their own conclusions.

That being said, I still wish to offer my own probabilities. For one, I think we owe the interested reader the clearest possible takeaways to do as much work for them and save them time - as long as we provide the appropriate caveats and warnings (you have been warned). Additionally, I think stating clear probabilities can be a clear way to explain and express my personal worldview and help make disagreements more precise, similar to what was done by Muehlhauser (2017), Section 4.2. I hope that the way we have structured this exercise - by presenting individual opinions that showcase disagreement among our own team and that go above and beyond to state the many caveats and uncertainties - will be enough to dissuade you from overweighting our opinions. I believe you deserve to know where I am personally coming from - more precisely and quantitatively.

Overall, I believe that many kinds of invertebrates are a lot more likely to be capable of phenomenal consciousness than I thought, given that these taxa are a lot more behaviorally sophisticated than I thought, and it becomes difficult to point to certain sophistications that cows and chickens display that honeybees do not. My general “case against” primarily rests upon what I think is a substantial chance for systematic bias in the papers we cite toward finding startling publishable conclusions, plus thinking that maybe a fundamental theory of consciousness will eventually emerge that leads me to put much more reduced epistemic rate on behavioral evidence, plus some remaining credence to my uninformed prior intuition for invertebrates are not all that sophisticated (e.g., intuitions from phylogenetic distance, intuitions that complex behavior may not actually be that complex, difficulties figuring out the complexity of behaviors).

I should also endeavor to repeat that I don’t have any particular reason to think these probabilities are well-calibrated or even all that useful, except as a more detailed expression of my current views and how they changed (in some cases, significantly). I’m definitely not an expert in thinking probabilistically and I’m unsure of the extent to which these probabilities do capture my views - though I do think they capture my views more clearly than I would express qualitatively in words, hence my reasoning for giving them.

While I would like to share my probabilities both before and after starting this research, I unfortunately did not actually record probabilities prior to starting the research. I tried my best to reconstruct them now as I would’ve thought about them then, to at least provide a rough guide as to how my views have changed.

I think my explanation for changing my views comes from two sources. First is the large amount of research that went into our table -- I found myself particularly weighing centralized information processing, contextual learning abilities, social learning, long-term behavior alteration, flexible tool use, and play behavior when attributing phenomenal consciousness to some organisms and not others. Also, while I weigh neuron count now much less as a factor than I previously did, I still think it is plausible that the capacity for phenomenal consciousness may scale to some degree with higher neuron count.

Second was just generally thinking about the role and function of consciousness for another year lead me to become more confident in certain areas. For example, while I am now more skeptical of a clear line being established between chickens and honeybees, I am more certain we can establish a clear line between honeybees and plants that excludes the plausibility of plant sentience. While looking into research on plant sentience did make me more impressed at the capabilities of plants, it also helped me understand how much different they are than the capabilities of other invertebrates and how seemingly complex behavior can likely result without consciousness.

*I added chimpanzees here because they were in Muehlhauser (2017), Section 4.2 and seem useful as a point of comparison, but they weren’t actually studied in our work.

Jason Schukraft

The Issue

What are the odds an octopus has the capacity to experience pleasure and pain? How much likelier is it that flies are sentient compared to earthworms? Given the available evidence, what is the rational credence in the proposition crayfish are phenomenally conscious?

Answers to questions like these could be extremely useful. Invertebrates outnumber vertebrates by a wide margin, but population counts alone aren’t the whole story. If invertebrates lack the capacity for valenced experience—that is, their experiential states never take on a positive or negative affect (either because they don’t have experiential states or those states are always neutral)—then invertebrates aren’t moral patients and don’t possess intrinsic moral worth.[5] If we knew the rational probability, conditional on the available evidence, that different groups of invertebrates had the capacity for valenced experience, we could multiply that probability by the (estimated) number of animals in each group. That would yield the expected number of animals for each taxon, and this figure could form the basis for evaluating how to divide scarce resources among different groups of animals.[6]

Nonetheless, I think we shouldn’t (yet) try to answer these types of questions.

More precisely, I contend that publishing specific estimates of invertebrate sentience (e.g., assigning each taxon a ‘sentience score’) would be, at this stage of investigation, at best unhelpful and probably actively counterproductive. The benefits are slim, and the risks are non-negligible. To be clear: I don’t believe it’s a bad idea to think about probabilities of sentience. In fact, anyone directly working on invertebrate sentience ought to be periodically recording their own estimates for various groups of animals so that they can see how their credences change over time. I also believe that we ought to aspire, as a field, to be able to provide robust, actionable estimates of invertebrate sentience. I just don’t think we’re there yet.

The Problem

Undoubtedly, the work of Rethink Priorities would be more digestible if we assigned a sentience score (either as a straightforward probability or as a position on an arbitrary scale so that such scores could be compared across taxa) for each of the groups of animals we investigated. Unfortunately, assigning objectively good sentience scores would be extraordinarily difficult. The 53 features we investigated are not equally important, and the context in which they are displayed often makes a substantial difference to their evidential weight. One would have to have an expert grasp on biology, philosophy, and neuroscience (as well as lots of time on their hands) to even justifiably begin such a scoring project. But without a clear methodology, it’s unclear how effectively the scores could be criticized or improved in the future. And because subjective experience is, well, subjective, strict calibration in this domain is necessarily impossible.

Of course, having studied the topic for some time now, I expect that my estimates would be better than the estimates of the average member of the EA community. If that’s true, then it’s tempting to conclude that making my estimates public would improve the community’s overall position on this topic. However, I think there are at least three reasons to be skeptical of this view.

(1) Sentience scores for specific taxa aren’t that useful

Rethink Priorities investigated the evidence of sentience for 18 groups of organisms.[7] After deciding which types of organisms we wanted to investigate, we next needed to decide the taxonomic level at which we would investigate those organisms. We tried to drill down to a fairly narrow taxon (species, genus, or family) because, in general, the higher up the taxonomic hierarchy one goes, the more diverse a taxon becomes. If a taxon becomes too large, then saying that the taxon possesses some feature ceases to be informative. If 50 different arthropods each possess one (and only one) of the features we investigated, and each species possesses a different feature, a database with the category “arthropod” would give the misleading impression that arthropods definitely have the capacity for valenced experience.

But when it comes to actual interventions, some degree of generalization is going to be inevitable. It’s extremely unlikely that a cost-effective intervention is going to specifically target Drosophila melanogaster or Caenorhabditis elegans. It’s much more likely that the intervention would target some much larger group, like insects or nematodes. So a sentience score for Drosophila melanogaster or _Caenorhabditis elegans _wouldn’t actually provide much actionable information without some idea whether and to what extent that sentience score generalizes. But developing a justified sentience score for Drosophila melanogaster or Caenorhabditis elegans is hard enough; doing so for groups as large and diverse as insects or nematodes would require far more data than we gathered.[8]

Moreover, even well-justified generalized sentience scores aren’t super useful unless we have at least a rough handle on how different groups of animals compare in moral weight. Knowing that there is a 25% chance eusocial insects are sentient and a 95% chance birds are sentient doesn’t help one judge which interventions are most effective unless one also know approximately how many insect-life-days a bird-life-day is worth.

(2) Sentience scores would inevitably be over-emphasized

Another concern is that interested parties might skip straight to our (uncalibrated, somewhat unjustified, extremely speculative) numerical sentience estimates without taking the time to understand the nuance and intricacy of the issue. Numbers are seductive because they are so easy to manipulate (e.g., just stick them into a cost-effectiveness analysis!). But caveats don’t fit into spreadsheets. Numbers, even ranges of numbers, can’t convey subtlety or complexity or nuance. Withholding explicit sentience estimates forces the reader to take a more holistic approach to our work. For an enormously complicated subject like sentience, the details really matter. I would much prefer a reader to come away from one of our posts with a renewed appreciation for the complexity and importance of the issue rather than sentience credences that exactly match mine.

It’s difficult to present explicit estimates of invertebrate sentience in a way in which those estimates don’t steal the show. It’s hard to imagine a third party summarizing our work (either to herself or to others) without mentioning lines like ‘Rethink Priorities think there is an X% chance ants have the capacity for valenced experience.’ There are very few serious estimates of invertebrate sentience available, so members of the community might really fasten onto ours. But if we did publish our estimates, I would want them to be viewed as hypotheses to be further refined (or perhaps completely abandoned) as more evidence comes in rather than hard conclusions that our work definitively supports. Personally, I worry that assigning sentience scores sacrifices too much in the name of digestibility.

(3) Sentience scores might reduce our credibility with potential collaborators

The future of the invertebrate welfare cause area will depend in large part on our ability to collaborate with biologists, ethologists, and neuroscientists. There is a great deal of empirical information, from better population appraisals to studies on the self-administration of anxiolytic drugs, that needs to be gathered to generate an informed view of how to best help various invertebrates. But science, especially peer-reviewed science, is an inherently conservative enterprise. Scientists simply don’t publish things like probabilities of sentience. For a long time, even the topic of nonhuman sentience was taboo because it was seen as unverifiable. Without a clear, empirically-validated methodology behind them, such estimates would probably not make it into a reputable journal. Intuitions, even intuitions conditioned by careful reflection, are rarely admitted in the court of scientific opinion.

Rethink Priorities is a new, non-academic organization, and it is part of a movement that is—frankly—sort of weird. To collaborate with scientists, we first need to convince them that we are a legitimate research outfit. I don’t want to make that task more challenging by publishing estimates that introduce the perception that our research isn’t rigorous. And I don’t think that perception would be entirely unwarranted. Whenever I read a post and encounter an overly precise prediction for a complex event (e.g., ‘there is a 16% chance Latin America will dominate the plant-based seafood market by 2025’), I come away with the impression that the author doesn’t sufficiently appreciate the complexity of the forces at play. There may be no single subject more complicated than consciousness. I don’t want to reduce that complexity to a number.

The Solution

The unfortunate reality is that producing useful estimates of invertebrate sentience is going to be incredibly hard. Nevertheless, the cause is too important to give up on. There are just too many invertebrates to justifiably ignore. The best and only solution is just to keep working as diligently as we can on the issue until we are in a position to deliver informed, actionable estimates. And let’s not kid ourselves: we’re not there yet.

But it’s not as if our work makes no progress on the question. We’ve argued that coleoid cephalopods (i.e., squid, cuttlefish, and octopuses) are clearly the best candidates for phenomenal consciousness in the invertebrate world. There is also very intriguing evidence of sentience for decapod crustaceans (i.e., prawns, shrimp, crayfish, crabs, and lobsters) and eusocial insects (i.e., ants, termites, bees, and wasps). The evidence of sentience is much greater for these groups than it is for jellyfish, earthworms, or roundworms. There is basically no reason to think plants, protists, or prokaryotes are sentient. It’s still possible that they are, but if so, our best philosophical and neuroscientific theories of consciousness are radically mistaken. These initial claims are well-justified and supported by the available data. They give us a foothold from which to explore other questions, and they offer at least rough guidance on where to focus further research. For now, that’s enough.

Credits

This essay is a project of Rethink Priorities. It was written by Jason Schukraft, Peter Hurford, Marcus A. Davis, and Daniela R. Waldhorn. Thanks to David Moss and Adrià Voltes for helpful feedback. If you like our work, please consider subscribing to our newsletter. You can see all our work to date here.

Notes


  1. Before November 2018 there were only about 10-15 person-hours devoted to this project per week. Beginning in November 2018, there were closer to 40-50 person-hours per week devoted to this project. ↩︎

  2. For examples of opinion pieces published in a high-quality science journal, see, inter alia, “What does AI’s success playing complex board games tell brain scientists?”, “Governing the recreational dimension of global fisheries,” and “Why science needs philosophy,” all published in PNAS. ↩︎

  3. The Cambridge Declaration on Consciousness claims that "convergent evidence indicates that non-human animals have the neuroanatomical, neurochemical, and neurophysiological substrates of conscious states along with the capacity to exhibit intentional behaviors. Consequently, the weight of evidence indicates that humans are not unique in possessing the neurological substrates that generate consciousness. Non-human animals, including all mammals and birds, and many other creatures, including octopuses, also possess these neurological substrates." ↩︎

  4. It should be noted that this criterion may, in itself, be the subject of further discussion. Which behaviors account for "cognitive sophistication" or assigning a "level of sophistication" to a given action is a complex matter that is beyond the scope of this article. Generally, it is assumed that the use of tools by octopuses is a display of cognitive sophistication. However, some scientists resist such an implication. In particular, Schnell & Clayton (2019) argue that, although octopuses do use tools (e.g., carrying coconut shells around as mobile dens), in the absence of controlled experiments, simpler explanations for this behavior cannot be ruled out. Further research is needed based on other than anecdotal observations. Hence, the conclusions here presented should be taken with caution. I thank Adrià Voltes for these points. ↩︎

  5. Strictly speaking, this argumentative move is too fast, but the details need not detain us. ↩︎

  6. Of course, final decisions would factor in lots more information, such as our best guesses about relative moral weight, the tractability of various interventions, flow-through effects of those actions, and so on. ↩︎

  7. I can’t just say ‘animals’ because we also investigated plants, protists, and prokaryotes. ↩︎

  8. Moreover, even well-justified generalized sentience scores aren’t super useful unless we have at least a rough handle on how different groups of animals compare in moral weight. Knowing that there is a 25% chance eusocial insects are sentient and a 95% chance birds are sentient doesn’t help me judge which interventions are most effective unless I also know approximately how many insect-lives a bird-life is worth. ↩︎

121

33 comments, sorted by Highlighting new comments since Today at 6:07 PM
New Comment

I've been impressed by the work being produced by Rethink Priorities over the past several months. I appreciate the thought and nuance that went into this. Great job again!

Interesting stuff, thanks!

(Maybe this comment would be more appropriate on another one of the posts of the series, perhaps this one or this one, but I've decided to leave it here since this is the most recent post about sentience across species. This is mostly speculative and suggestive of areas for further research and questions that could inform our credences in consciousness.)

I haven't been able to find the paper, unfortunately, but I thought this might be pretty good evidence for consciousness:

Another way to tap into an animal's emotions is to train them to communicate how they feel. A group of researchers from London taught pigs to give one response when they felt normal and a different response when they were anxious (in this case they were given a drug designed to induce temporary anxiety). Not only could the pigs discriminate between these two states, but later they made the same "anxious" response when exposed to novel events such as an unfamiliar pig or a new pig pen. It seems that, since pigs are smart enough to tell researchers how they feel, they could be trained to understand that although a routine husbandry procedure might be frightening, it could be over relatively quickly and painlessly.

If an animal can discriminate between their own emotions and be taught to react to them in different ways (and not just to external stimuli or representations thereof), is this a sign of some kind of higher-order consciousness? Their learning systems are taking their own emotions as input to learn about, rather than just as final signals (e.g. reinforcers) for learning, and their emotions are being passed through their learning systems and not just directly to changes outside their brains (tail wagging, growling, stress hormones in the blood, etc.).

More recently, there's the dog who was trained to press buttons that play words. At about 4:00, after a button didn't work, the dog pressed "No" and "Help".

And then cows may be able to react emotionally to their own learning, specifically, to their progress on a problem. Maybe this kind of learning is distinct from the kind of learning pigs and dogs seem to do about their own emotions.

Together (although I'm generalizing between species), this may suggest these animals can:

1. learn about their own emotional reactions, and

2. react emotionally to their own learning.

I suppose learning does not imply "knowing" or "thinking", though, which should involve consciousness. So the pigs and dog seem to be able to learn about their emotions, but do they "know" or "think about" their emotions? The cows seem to react to their own learning/progress, but do they "know" that they're learning/making progress?

On the other hand, how far is this kind of behaviour from nonhuman primates being able to communicate using sign language and how good was that as evidence for nonhuman primate consciousness? Nonhuman primates may have learned to communicate more than just their emotions, including perhaps past events (skepticism here, though), which would be evidence for episodic memory.

But maybe even rats have episodic memory and perform mental time.[1][2] It seems worth investigating whether or not these memory replays are associated with emotions similar to the original events; if that's the case, that would be a good sign that they're actually experiencing the "simulation". I'd imagine that when a dog runs in their sleep, they're simulating experiences and actually experiencing the simulation, emotions and all, although the claim that dreams are actually experienced while they're happening in humans if not lucid - rather than just at recall after waking up - is debated in philosophy. Some related reading: [3][4][5]

It also seems worth looking into intentionality here: do they ever consider alternatives (e.g. run multiple simulations) and then make a decision based on the outcomes of the simulations? Can they choose to recall or run simulations in the first place?

An excerpt on gorilla sign language for emotions and memories from this paper (bold mine):

Which nonhuman animals may be near-persons like Jenny? Reviewing the evidence, Varner argues that the category includes great apes, cetaceans, elephants, and, perhaps, corvids and parrots. To defend his claim that great apes do not have the kind of episodic memory required to have a biographical sense of one’s past, he examines the evidence provided for believing that Koko, the gorilla, has narrative and uses it to communicate deeply emotional personal memories from the distant past.
Koko was five years old in July, 1976. According to Francine “Penny” Patterson, who worked more closely with Koko than anyone, in 1976 Koko narrated an event that had happened three days prior:
(P = Patterson; K = Koko)
P: What did you do to Penny?
K: BITE.
P: You admit it? (Koko had earlier called the bite a SCRATCH.)
K: SORRY BITE SCRATCH. (Penny shows the mark on her hand; it does resemble a scratch.)
K: WRONG BITE.
P: Why bite?
K: BECAUSE MAD.
P: Why mad?
K: DON’T KNOW
(Patterson & Cohn, 1994, p.282)
Koko’s one and two word responses here, drawn from her knowledge of more than a thousand American Sign Language (ASL) signs, clearly show an understanding of concepts, words, and causal relations (What did you do to Penny? BITE). However, as Varner notes, there is no evidence here of episodic memory, in which one remembers oneself at a particular place at a particular time. Koko is using ASL which, Varner tells us, does not include tenses. Consequently, he observes, “temporal references must generally be inferred from the context, and in these studies, that context is provided by the English sentences uttered by the human trainers” (Varner, 2012, p.155). Varner has his doubts about whether Koko is here communicating a conscious memory of what happened three days ago. Rather, Koko may simply be making signs she knows will succeed in eliciting the responses Koko desires from Patterson.
But if Koko is not capable of expressing memories of events three days in the past, she is able to communicate her emotions. When asked, “How do you feel?” she will respond appropriately, for example, with FINE, or HUNGRY, or SAD. In children, internal immediate-state language reporting one’s mood emerges in the third and fourth years. We are on firm ground, then, in thinking Koko has words and concepts, social communication, rationality in the sense of cause and effect thinking, emotions, awareness, and beliefs and desires. But she does not seem to have the second-order desires, executive control, or autonomy required for a biographical sense of self.
Varner is similarly cautious about long-term memories allegedly recounted by a gorilla, Michael, who was captured by poachers as an infant. Patterson made a video of Michael allegedly recounting this memory of the incident in which Michael’s mother was killed. In the recording we see Michael’s signings rendered in the following captions provided by Patterson: “SQUASH MEAT GORILLA. MOUTH TOOTH. CRY SHARP-NOISE LOUD. BAD THINK-TROUBLE LOCK-FACE. CUT/NECK LIP(GIRL) HOLD” (The Gorilla Foundation, n.d.). Varner, noting the ambiguity of the string of words, observes that “even Patterson’s sympathetic co-author Eugene Linden doubts her claim that Michael was telling the story about his mother’s death” (Ibid, pp.155–156). Varner concludes that in spite of such anecdotes and Patterson’s claim that Michael told her this story on several occasions, there is “no good evidence that apes understand or use language to express thoughts about the non-immediate past” (Ibid, pp.156). If Varner is wrong and Michael is recounting an episodic memory, Michael has an important claim to personhood. If Varner is right, perhaps Michael is just making signs he thinks Patterson is subconsciously nudging him to make, perhaps in Clever Hans fashion. In that event, Michael may not have episodic memories of the traumatic events. Rather, he may only be signing in sequences he has learned satisfy Patterson’s promptings.

Also, for another comparison, children only really start passing the mirror test at around 2 years old.

EDIT: On further consideration, "In children, internal immediate-state language reporting one’s mood emerges in the third and fourth years" seems suspiciously late to me.

Also interesting here to reflect on what it'd mean for consciousness & intentionality to be (mostly) orthogonal, though that heads down a different road.

Also, apparently the theories of higher-order consciousness are pretty broad (pdf, see especially table 1 there). Maybe most animals we suspect are conscious in the RP report would meet at least one definition of HOT.

Hi Michael,

Thanks for the interesting comment and many useful references. Speaking for myself and not for the rest of the team, I am very confident that dogs, pigs, cows, rats, and apes are all sentient and have the capacity for valenced experience. (That is, there is something it is like to be these animals and that experience includes pleasures and pains.) Whether or not these creatures are capable of higher-order conscious thought (that is, reflecting on their own first-order beliefs, desires, or emotional states) is debatable. I don't think higher-order conscious thought is a necessary condition for sentience, but I do think it may be relevant to moral status. In fact, many of the features mentioned in your comment (e.g., episodic memory, emotional complexity and awareness, social communication, cause-and-effect thinking, executive control, autonomy, biographical sense of self) plausibly help determine a creature's moral status. (Even if you are suspicious of degrees of moral status, you might think that these features contribute to the range and types of experiential states a creature can undergo and thus are important for determining a creature's welfare.) So I think it would be good for the animal welfare movement to have a decent grasp of which of these features (as well as the other features that plausibly affect moral status) are exhibited by which animals. In the coming months, Rethink Priorities might have something more concrete to say about the topic.

Agreed!

I do agree that higher-order consciousness is unlikely to be necessary for sentience, but some disagree (not any of the authors here, AFAIK), and showing them that there's a good chance many animals are capable of higher-order consciousness could lead to further consideration for these animals' interests. Unfortunately, this may neglect many animals whom we think aren't capable of higher-order consciousness or even entrench the view that higher-order consciousness is necessary.

Also, pigeons and rats seem to be able to answer unexpected questions about what they've just done. See the section "The unexpected question" here. This is, of course, different from reporting an internal state, though.

You could also consider providing your probabilities in the form of distributions, for example by answering a question like "What probability would you assign to tribbles [1] being sentient after 1000 more hours of research?" This would perhaps solve part of the problem of communicating the uncertainty which you want to communicate.

Some example answers might be:

In the first one your uncertainty is significant. You have a probability of ~45%, but you consider it likely that you will update a lot. You just don't know in which direction: You consider it equally likely that you will end up at 30% or at 70%.

In the second one, research has mostly been done and you've already mostly made up your mind. You have a probability of ~45%, and you believe that further research is most likely to move you to 42%, or to 47%, but not that much further. You'd be shocked if you ended up with a probability of more than 60%.

The third one is the distributional equivalent of you making a shrug. Your probability is 45%, but really, tribbles being so underexplored means that your distribution looks pretty much uniform.

Note that it's even possible that all three distributions have the same mean (~45%). This would mean that in all three cases you'd think that a bet of 45:55 would be a fair bet (that is, that it has an expected value of 0).

You could also predict the standard deviation of your distribution (how broad it is) after 1000h of research, 2000h, 5000h, etc., aggregate the distributions of all your researchers, and do other nice things.

[1] Fictional species.

Thanks! That's helpful way of thinking about the topic and a useful strategy for addressing the problem.

"However, perhaps my largest surprise wasn’t an update toward or against a particular type of animal, rather it was based on the extent of conditioned learning behavior that is more or less exhibited by all taxa we considered, including single-celled organisms and animal bodies detached from brain communication, including the lower body of a mouse with a severed spine. While one could take this as weak evidence of widespread sentience, this updated me toward thinking many of these behaviors aren’t very impressive and they were thus largely disregarded in contemplating the positive case for sentience. "

Marcus, is there any chance you could elaborate on why you leaned one way on this vs the other? I don't have a clear sense of what I should take away from that, so I'd be curious what your reasoning was.

...

I'd also be interested in all of your thoughts on what exactly a percentage probability of valenced experience (or whatever the morally relevant mind-stuff should be called) is - obviously, they aren't that close to the fact of whether or not these organisms have valenced experience (which, unless the world is very strange, should be 1 or 0 for all things)

It seems more like they are statements about how you'd make a bet, or something like "confidence in the approach * results from the approach", or something else about the approach and prioritization. I'm curious how you were defining these probabilities to yourselves, and how definitions would impact their usefulness in cost-effectiveness analyses? i.e. if we were doing a cost-effectiveness estimate, and treating these as confidence * results, I might weight my confidence in this method higher than using my intuitions, but still include other approaches like intuition in my estimate because it theoretically gives me a more accurate model of my current knowledge. But, with a different definition I might just use these numbers.

I'd also be interest in all of your thoughts on what exactly a percentage probability of valenced experience (or whatever the morally relevant mind-stuff should be called) is - obviously, they aren't the close to the true probabilities these organisms have valenced experience (which, unless the world is very strange, should be 1 or 0 for all things)

I may be an odd person to answer this question, as I chose not to offer probability estimates, but I'll respond anyway.

I agree that sentience, at least as we've defined it, is an all-or-nothing phenomenon (which is a common view in philosophy but not as common in neuroscience). As I understand them, the probabilities we discuss are credences, sometimes called subjective probabilities or degrees of belief, in the proposition "x is sentient." Credence 1 (or 100%) represents certainty that the proposition is true and credence 0 (or 0%) represents certainty that the proposition is false. Since there are very few propositions one should be absolutely certain about, the appropriate credences will fall between 0 and 1. The betting analysis of credence is common, though there are some well known problems.

Thinking of these probabilities as credences is neutral on the question of the best way to develop and refine these credences. Someone might base her/his credences entirely on intuition; another person might completely disregard her/his intuitions. This post details what we take to be the best available methodology to investigate invertebrate sentience.

I agree that sentience, at least as we've defined it, is an all-or-nothing phenomenon (which is a common view in philosophy but not as common in neuroscience).

What do you think of the argument that there may be cases where it's unclear if the term is appropriate or not. So there would be a grey area where there is a "sort of" sentience. I've talked to some people who think that this grey area might be taxonomically large, including most invertebrates.


Hey Max, good question. I think we need to clearly separate our metaphysics from our epistemology in this area. If an entity is sentient if and only if there is something it is like to be that entity, then it's hard to see how sentience could come in degrees. (There are closely related phenomena that might come in degrees--like the intensity of experience or the grain of sensory input--but those phenomena are distinct from sentience.) There are certainly going to be cases where it's difficult to know if an entity is sentient, but our uncertainty doesn't imply that the entity is only partially sentient. I think it's plausible that this area of epistemic indeterminacy could remain quite large even with all the empirical facts in hand.

However, there are some theories of mind on which it looks like there could be cases of metaphysical indeterminacy. If a certain type of reductive physicalism is true, and sentience doesn't reduce to any one feature of the brain but is instead a cluster concept, and the features that constitute the concept aren't coextensive, then there could be cases in which we don't know if an entity is sentient even with all the empirical and the philosophical facts in hand. (Technically, the fact that it can be metaphysically indeterminate that an entity possesses a property doesn't entail that the property comes in degrees, but it's a natural extension.)

Thanks Jason!

That makes sense - I understood that you all were expressing credences. I think my comment wasn't written very clearly. I'm interested in what process you all took to reach these credences, and what you think the appropriate use of them would be. Would these numbers be the numbers you'd use in a cost-effectiveness analysis, etc.? Or a starting point to decide how to weigh further evidence, etc? I know credences are a bit fuzzy as a general concept, but I guess I'd love thoughts on the appropriate use of these numbers (outside your response that we shouldn't use them or should only use them very carefully).

I meant to highlight a case where I downgraded my belief in a scenario in which there multiple ways to update on a piece of evidence.

To take an extreme example for purposes of clarification, suppose you begin with a theory of sentience (or a set of theories) which suggests behavior X is possibly indicative of sentience. Then, you discover behavior X is possessed by entities you believe are not sentient, say, rocks. There are multiple options here as to how to reconcile these beliefs. You could update towards thinking rocks are sentient, or you could downgrade your belief that behavior X is possibly indicative of sentience.* In the instance I outlined, I took the latter version of the fork here.

As to the details of what I learned, the vast bulk of it is in the table itself, in the notes for various learning attributes across taxa. The specific examples I mentioned, along with similar learning behaviors being possible in certain plants and protists, are what made me update negatively towards the importance of these learning behaviours as indicative to sentience. For example, it seems classical conditioning, sensitization, and habituation are possible in protists and/or plants.

*Of course, these are not strictly the only options in this type of scenario. It could be, for example, that behavior X is a necessary precondition of behavior Y which you strongly (perhaps independently but perhaps not) think is indicative of sentience. So, you might think, the absence of behavior X would really be evidence against sentience, while its presence alone in a creature might not be relevant to determining sentience.

So, you might think, the absence of behavior X would really be evidence against sentience, while its presence alone in a creature might not be relevant to determining sentience.

I might be wrong about this or might be misunderstanding you, but I believe that, in any case where the absence of X is evidence against Y, the presence of X has to be evidence for Y. (Equivalently, whenever the presence of X is evidence for Y, the absence of X has to be evidence against Y.)

This does go against the common statement that "Absence of evidence is not evidence of absence." But we can understand that statement as having a very large kernel of truth, in that it is often the case that absence of evidence is only extremely weak evidence of absence. It depends on how likely it would be that we'd see the evidence if the hypothesis was true.

For an extreme example, let's say that an entity not being made of molecules would count as very strong evidence against that entity being sentient. But we also expect a huge number of other entities to be made of molecules without being sentient, and thus the fact that a given entity is made of molecules is extraordinarily weak evidence - arguably negligible for many purposes - that the entity is sentient. But it's still some evidence. If we were trying to bet on whether entity A (made of molecules) or entity B (may or may not be molecules; might be just a single atom or quark or whatever) is more likely to be sentient, we have reason to go with entity A.

This seems to sort-of mirror the possibility you describe (though here we're not talking behaviours), because being made of molecules is a necessary precondition for a huge number of what we'd take to be "indicators of sentience", but by itself is far from enough. Which does mean evidence of X is extremely weak evidence of sentience, but it's still some evidence, relative to a state in which we don't know whether X is true or not.

(I'm aware this is a bit of a tangent, and one that's coming fairly late. The post as a whole was very interesting, by the way - thanks to everyone who contributed to it.)

Very interesting, thank you!

Potential typo: In the following paragraph, I think "likely yes" should probably read "very probably yes".

Although incomplete, there is direct evidence that individuals of these taxa exhibit features which, according to expert agreement, seem to be necessary –although not sufficient– for consciousness (Bateson, 1991; Broom, 2013; EFSA, 2005; Elwood, 2011; Fiorito, 1986; Sneddon et al., 2014; Sneddon, 2017) (see the first criterion of the ‘likely yes’ category);

(Super minor: It also seems slightly inconsistent that the section title "Very Probably Yes" is capitalized, while others aren't.)

You're right! Thanks for your feedback, Max Daniel! We'll correct that shortly.

Some neuroscience-based arguments against pain in

It might be worth adding more metacognition features to the table. This could be a good place to start:

http://animalbehaviorandcognition.org/issue.php?id=25

Some more evidence I find pretty compelling and might be surprising:

1. Fishes have friends.

2. Fishes get depressed.

3. Wild birds get PTSD.

Generally I find social behaviour and mental disorder pretty compelling.

I'm a little surprised that the estimates for chickens and cows aren't higher. Personally, I find evidence of complex and varied emotions to be very compelling, especially social emotions, e.g. play behaviour, emotional empathy/contagion, affection and social attachments to particular individuals (companionship), helping behaviour (altruism), parenting generally, separation anxiety and perhaps even something like grief. Also, possible emotional reactions of cattle to learning. :P

I would be comfortable using the word 'love' to describe the attachments chickens and cows often have towards others, although it may of course be quite different from an adult human's experience of love, but perhaps not that different from an infant's or toddler's. It's hard for me to imagine an individual capable of love like this not being sentient.

I suppose I also give weight to anecdotes and videos of individual animals, though.

I was one of the authors who gave a "low" (~80%) probability to chicken and cow consciousness. I agree the evidence here appears compelling, but I personally put some probability weight on the idea that this overall "inference to the best explanation" is fundamentally wrongheaded somehow and that there may be a good explanation for seemingly very complex behavior not arising out of a state of consciousness that we just don't know of yet. Seemingly complex behavior can arise from simple systems For example, OpenAI's (presumably) non-conscious computer programs exhibiting emergent tool use). Muehlhauser (2017) explains this line of argument in depth, especially Section 3.2.3, Section 3.4, and Section 4.3.

Very late follow-up, but what specific features do chimpanzees have that account for most of the gap between your credence in their consciousness and that of chickens and cows?

Chimps have much more neuroanatomical similarity with humans, more apparent cognitive sophistication (such as learning ability, language ability), and more complex social behavior, relative to chickens and cows.

How should we interpret ranges of probabilities here?

We can talk about confidence (credence) intervals for frequencies for the population we're sampling from for polls and surveys. For species (or individuals) with characteristics of interest (possibly a feature or its absence) , we could describe our probability distribution over the fraction of them that are sentient.

Another approach might be to try to quantify the sensitivity to new information, e.g. if we also observed another given capacity (or its absence), how much would our estimate change? If we model the probability that a species (or individual) will have a set of characteristics of interest given a fixed set of observed characteristics, we could compute a credence interval for our posterior probability of sentience with , over the distribution of conditional on observed characteristics.

Are either of these what some of you had in mind, roughly (even if you didn't actually calculate anthing)? Or something else?

My ranges represent what I think is a reasonable position is on the probability of each creatures sentience given all current input and expected future input. Still, as I said:

...the range is still more of a guideline for my subjective impression than a declaration of what all agents would estimate given their engagement with the literature

I could have made a 90% subjective confidence interval, but I wasn't confident enough that such an explicit goal in creating or distributing my understanding would be helpful.

Another one for bees: information integration across or generalization between senses.

"Bumble bees display cross-modal object recognition between visual and tactile senses" by Cwyn Solvi, Selene Gutierrez Al-Khudhairy and Lars Chittka.

Humans excel at mental imagery, and we can transfer those images across senses. For example, an object out of view, but for which we have a mental image, can still be recognized by touch. Such cross-modal recognition is highly adaptive and has been recently identified in other mammals, but whether it is widespread has been debated. Solvi et al. tested for this behavior in bumble bees, which are increasingly recognized as having some relatively advanced cognitive skills (see the Perspective by von der Emde and Burt de Perera). They found that the bees could identify objects by shape in the dark if they had seen, but not touched, them in the light, and vice versa, demonstrating a clear ability to transmit recognition across senses.

Many animals can associate object shapes with incentives. However, such behavior is possible without storing images of shapes in memory that are accessible to more than one sensory modality. One way to explore whether there are modality-independent internal representations of object shapes is to investigate cross-modal recognition—experiencing an object in one sensory modality and later recognizing it in another. We show that bumble bees trained to discriminate two differently shaped objects (cubes and spheres) using only touch (in darkness) or vision (in light, but barred from touching the objects) could subsequently discriminate those same objects using only the other sensory information. Our experiments demonstrate that bumble bees possess the ability to integrate sensory information in a way that requires modality-independent internal representations.

>In the case of C. elegans, despite being a widely studied animal, evidence of their being conscious is weak and their behavioral reactions are simple and highly stereotyped.


What do you mean by simple? Having nociceptors is a good indicator of pain, right? Surely more than a 1% chance, that just seems like crazy overconfidence.

Hi!

Although the possession of nociceptors is perhaps some evidence of a capacity to feel pain, it is certainly not by itself "a good indicator" of that capacity. If nociceptors are not connected to centralized information-processing structures, these neurons could trigger reflexive reactions (i.e., similar to spinally mediated responses in mammals), but that would not imply that the nociceptive input is consciously perceived (in humans, see Becker et al., 2012; Dubin & Patapoutian, 2010). If we understand consciousness as suitably integrated information (Oizumi et al., 2014), the projections of nociceptors to integrative information-processing structures is a crucial aspect to examine when judging the probability that a nonhuman individual is conscious.

In the case of C. elegans, unlike other invertebrates, they do not seem to have a specific neural region for the processing of spatial information and organization of movement. In other words, movement and stimuli discrimination do not appear to be integrated in a manner sufficiently similar to the vertebrate midbrain (see Altun & Hall, 2011; Kato et al., 2015).

However, it should be noted that some noxious stimuli reactions have been identified in C. elegans, specifically, physiological responses to nociception and moving away from a noxious stimulus. However, heat-evoked escape responses in these animals, for example, are considered highly stereotypical, and a reflexive reaction (Leung et al., 2016).

Finally, when I used the term "simple" [nociceptive behaviors] here I specifically meant: (i) nociceptive responses can be identified, but they do not necessarily account for noxious stimulus intensity and direction, (ii) absence or insufficient indicators of 'long-term' learning and memory, and (iii) absence or insufficient indicators of motivational tradeoffs. Given our findings (summarized here), C. elegans seem to mostly display simple nociceptive responses.