My only feedback is that it would be nice if there was a top-line "sentience score" for how sentient a creature is. It's hard to compare the sentience of cows to bees in a digestible way. Especially as an advocate and lay-person. For anyone though, I think this might make the information more digestible.
It might work something like this: a being would be granted the score of 1.0 if we categorize all taxa as a likely yes. They get "docked" points as the answers lean towards no. It would be great if this was weighted for how important a taxa is in determining sentience (if that's even something we could do).
Can we even weigh out how "important" some taxa are? Might this sort of system be helpful to others? Please excuse my ignorance and if this wasn't the best forum to voice this feedback!
Hi Sammy. I’m one of the researchers on Rethink Priorities’ invertebrate sentience team. Thanks for your comment. This is an issue our team has thought a lot about and plans to address explicitly in forthcoming work. I agree that our research would be more digestible if we provided an overall probability of sentience for each taxon. Unfortunately, assigning a “sentience score” is 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 her/his hands) to even justifiably begin such a scoring project. And because subjective experience is, well, subjective, strict calibration in this domain is necessarily impossible.
Despite the above difficulties, Rethink Priorities is considering reporting our best guesses about the probability of sentience for our studied taxa. We are still figuring out the best way to present these preliminary estimates. We want the estimates 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. One concern is that interested parties might skip straight to our (uncalibrated, somewhat unjustified, extremely speculative) numerical estimates without taking the time to understand the nuance and intricacy of the issue. Personally, I worry that assigning sentience scores sacrifices too much in the name of digestibility.
Nonetheless, it’s not as if our currently published findings are completely silent on the matter. Clearly, there is better evidence for sentience for cephalopods and arthropods than there is for annelids and nematodes. Stay tuned for our invertebrate welfare cause profile (slated to go up in late July) for more on the implications of our research.
Agree with the risks of presenting such a score. Agree that scores would be very speculative and your credibility intervals would be very wide. But I'd also guess that without this sort of score/summary measure, then it's very hard to use this research in practical applications?
Perhaps a compromise is to compile these sorts of summary scores, but then to only share them with advocates or researchers that have specific purposes in mind for those figures? This way, if someone wanted to use the summary score to inform an estimate, or make some decision based off of your research here, they could do so and it would only take them a few minutes to send you an email, rather than several hours trying to come up with an equivalent estimate, using your research as a starting point.
It's still possible that the figures would become more widely known if people find the numbers indirectly, e.g. in citations. But this seems unlikely to affect many people (unfortunately, I don't imagine this research going viral).
PS thanks very much for this very thorough seeming research on this difficult topic!
(Quick sidenote: "if we categorize all taxa as a likely yes"... it sounds like you're saying "taxa" are the features/rows, but "taxa" refers to groupings of animals. Sorry that the term is a bit unfamiliar.)
I mentioned Braitenberg vehicles in a reply to one of Jason's other posts and then realized I hadn't seen these mentioned elsewhere in relation to invertebrate sentience (or in EA really), so I thought it would be worth mentioning them here as the concept may provide some interesting perspectives. Essentially, the vehicles are a thought experiment by Braitenberg (a neuroscientist) on intelligence based on building up from something simple that moves faster when it doesn't like where it is (vehicle 1) to a vehicle that is practically human (vehicle 14). Essentially, the book explores at what point can we agree that the vehicle is intelligent, even if the mystery of biological intelligence isn't present because we built it (actually, this is almost exactly analogous to Mesh:Hero experiment described in the 2017 Consciousness report). Strangely, the work seems to be better known by roboticist than by neuroscientists.
I think Braitenberg vehicles could be a useful reference for this project as the vehicles were all based on biological concepts of different levels of intelligence, and the vehicles may already have been discussed by philosophers of intelligence as to what level constitutes a threshold for a intelligent (probably analogous to conscious) entity. Indeed, the vehicles could also provide inspiration for something analogous to the sentience score requested by Sammy, as each vehicle was intended to represent something of a 'step up' in intelligence. So one could take the max or average level that a taxa reaches on such a scale as its score.
Thanks for another fascinating comment. Although we haven’t been framing the subject in this way (the Braitenberg reference is new to me), we’ve been thinking about similar issues for a long time. At an early stage of the project we had a spreadsheet that attempted to judge the extent to which a handful of robots and AI programs exhibited the 53 features we investigated for invertebrates. We de-prioritized the spreadsheet because filling it in required too many subjective judgment calls and we worried that the methodology we used to investigate invertebrate sentience wouldn’t be applicable to non-biological organisms. Ultimately, this is a question we hope to return to. There is ample material to explore: functionalism (and its denial) in philosophy of mind, graded states of consciousness, “evolution” in artificial reinforcement learning, the analogy between nonhuman animals and robots, and many others.
Thanks for your enriching comment, Gavin. Just wanted to add to Jason's response that, unfortunately, there is no consensus on whether various features potentially indicative of consciousness would be adaptive for any conscious individual, regardless of a species' evolutionary history and its adaptive needs.
Complicating things even further, we do not even have such thing as a 'universal' intelligence measuring instrument for humans–cultural differences in intelligence determine results country by country. The above points out that we need more research that tells us both criteria for understanding which features might be more robust for detecting consciousness, and forms of measurement that are sensitive to relevant differences between different groups of individuals.
Some ideas for the presentation of the table to make it more digestible:
1. Is the table downloadable? Can it be made downloadable?
2. Can the table cell/font sizes and table height be made adjustable? It would be nice to be able to fit more of it (ideally all of it) on my screen at once. Just zooming out in my browser doesn't work, since the table shrinks, too, and the same cells are displayed.
3. What about description boxes that pop up when you click on (or hover over) a cell (description/motivation of the feature itself, a box with the footnotes/text/sources when you click on the given cell)? Could also stick to informal recognizable names (cows, ants) where possible and put the taxon in a popup to save on space.
For now, (1) the table is not downloadable. Regarding (3), in various cases, we use the taxon or the species name to highlight that existing evidence only applies to that specific species. Thus, generalizing from a single species to the rest of the taxon or higher taxonomic ranks is potentially problematic –primarily when an invertebrate category comprises a large group of species, as is the case with ants.
Thank you for your suggestions. We will take them into account to improve our work.
+1 for the highlighting. Especially if we could toggle this on and off! This might also be a good system to use if we could highlight species as "likely sentient", "weak-evidence for sentience", "unknown", "Little evidence of any sentience", "likely not sentient."
Perhaps someone else might have better names for these classifications...
What are the plans for maintaining/expanding this database? Would you consider making a wiki or open source version and allowing contribution from others (possibly through some formal approval process)?
I imagine it could be a useful resource not just for guiding our beliefs about the consciousness of invertebrates, but also the consciousness of other forms of life (and AI in the future).
One suggestion: I think it could be useful to have a column for the age at which each feature is first observable in humans on average (or include these in the entries for humans, as applicable).
Hi MichaelStJules. Thanks again for your valuable suggestions! We will publicly share our plans to update /expand this database. Additionally, in upcoming posts we aim to outline the current state of improving invertebrate welfare as a cause area and suggest possible next steps in this regard.
Was the mirror test experiment with ants missed or was it intentionally excluded? If the latter, why? It seems the journal it was published in is not very reputable, and the results have not been replicated independently.
We were aware of that study, thanks for sharing it!
However, that paper is published by a potentially predatory journal, and no further evidence on this matter was found. Hence, we maintain our current response (‘unknown’), but will introduce a clarificatory commentary mentioning this study and its reliability limitations.
It looks like that ant-self recognition study should have been published somewhere better - it could be better presented and is missing some controls, but otherwise seems sound and was done by two established researchers.
I'm personally much more skeptical of the work. Most notably, the paper's results are suspiciously clear-cut -- all the of treated ants showed the mirror test behavior and none of the control ants did. Generally mirror self-recognition tests are not this straightforward and it is common, even with chimps, for some tested subjects to not pass. Unless ants are potentially smarter than chimps, I don't think we should expect results this clear. Instead, this suggests something in the study may have gone wrong.
Also Max Carpendale, then a contract researcher with Rethink Priorities, pointed out to me that one of the authors of the paper, Marie-Claire Caemmarts, has a history of making dubious claims outside mainstream scientific consensus. For example, she has published research that suggests cell phone radiation and Wi-Fi are damaging to human health, for which she has received criticism of bad science.
Thanks for the comments Peter and Michael. I'm not very familiar with the mirror test in general so I can't comment with confidence about how well this compares to the results with other species. But after having looked back at Table 2 in the paper reporting the mirror test results I'd argue the results aren't so clear cut - one ant never cleaned itself, whereas the other ants cleaned themselves between 1 and 9 times over the six minute trial (also, the behaviour never occured in juvenile ants). I don't think this indicates that ants are smarter than chimps, another explanation is simply that, assuming cleaning occurs because the ants visually recognised the paint spot from their reflection, that this triggered a reflexive grooming behaviour. Chimps probably have more complex motivations - if they see the paint spot some may want to remove it, but others might not be in the mood for cleaning or could enjoy having it there. If the difference is then between reflexive ant cleaning vs. voluntary chimp cleaning you could then go onto discuss the relevance of each type of behaviour for demonstrating self-recognition, but I don't think we are there yet.
Admittedly, this study was not performed rigorously. The review Michael links to (Gallup 2018) presents some important criticism. The ants also could have perceived the paint through mechanosensory hairs on their head - although the difference in grooming between brown and blue paint seems to suggest this wasn't the case. Another point is that this study doesn't really define what was considered cleaning behaviour, and it may be that the threshold for this was lower than that used in by chimp researchers, raising the likelihood of false positives for the ants. In addition to clear monitoring of the ant grooming pre-, during, and post- mirror exposure, it would also be useful to do a recovery experiment whereby the blue paint was covered with brown paint - if this prevented later grooming behaviour it would strongly suggest that the grooming was related to the ants visual perception of its reflection.
It's also a fair point that both Caemmarts do not perform very rigorous research studies and generally publish them in lowly ranked to predatory journals. Publishing in low ranked journals isn't a crime in itself, although the peer review usually isn't very rigorous and I wouldn't be surprised if these authors simply keep submitting the same manuscript to a journal until one will take it without revisions (this is poor scholarship but not uncommon). However, it does present a bit of problem if the first report on a controversial topic (invertebrate self-recognition) comes out like this - it probably didn't get much attention from other ant cognition researchers (there might not be many anyway), and a simple replication study can't be published in highly ranked journal (if the results held up and the study was done well I think this would have made it into mid to top ranked generalist journal).
In reality, this is a pretty simple experiment to replicate with ants. I know a few ant navigation researchers in Europe who could easily supervise a student to replicate it over summer (the Myrmica genus is quite common). I can put you in contact if this is a useful point for RP to confirm for the invertebrate sentience project? I'd also wager $50 that this does replicate, insofar that the ants groom paint spots based on their reflection but besides any interpretation of self-awareness based on this result.
Hey Gavin, we (at Rethink Priorities) would be interested in exploring funding for this. Would you be able to reach out to [email protected] and we can discuss next steps?
After hearing opinions about the Cammerts from another academic who knows them‚ I've unfortunately become a lot less confident that this study could replicate.
Is the report by Cammaerts and Caemmaerts (2015) positive evidence of self-recognition in ants? Our answer is an emphatic no. Too many crucial methodological details are not given. No formal period between marking the subjects and then exposing them to the mirror was included; the reader is simply asked to accept that no self-cleaning movements occurred before marked ants first saw themselves in the mirror and that marked ants without any mirror did not do so. There is no clear mention of how these data were collected. Were the ants recorded on video? Were they observed directly? In other studies of ant behavior some means of magnification are used, but Caemmerts and Cammaerts provide no information about this, and it is not even clear if any attempt to assess inter-observer reliability was made.
It also remains a possibility that responses to the mirror on the mark test were confounded by chemical cues from the ant’s antennae and chemoreceptors on the mandibles. For instance, if the blue dye was chemically different from the brown dye, chemoreception could explain why ants marked with blue dye were more likely to be attacked by other ants. It is also important to note that the ants must have sensed that they had the marks on themselves through these and other olfactory channels prior to being exposed to the mirror, which would invalidate the mark test.
Notwithstanding the absence of evidence for vision-based individual facial recognition in ants, it would be astonishing if such poorly sighted, small brained insects − especially those without any mirror experience − could immediately use their reflection to try to remove a freshly applied foreign mark that was only visible in the mirror
While invertebrates make up the majority of animal species, our knowledge about their capacity for valenced experience is overlooked compared to existing evidence about vertebrates. In particular, whether invertebrates have the capacity to have valenced experience is uncertain, and hence, it remains unclear whether these organisms have a welfare of their own we should care about.
Rethink Priorities has been systematically exploring this issue during the past months. First, we examined the philosophical difficulties inherent in the detection of instances of morally significant pain and pleasure in nonhumans. Second, given the current epistemic state about invertebrate consciousness, we have been compiling and analyzing relevant scientific evidence regarding this issue. In particular, we investigated the degree to which different features potentially indicative of phenomenal consciousness are found throughout different taxa. In a second post, we described our approach, the rationale of this project, and its limitations. The features we believe to be most relevant for assessing invertebrate sentience are explained in detail in a third, fourth, and fifth post.
In this sixth post, we present our summary of findings, both in narrative form and as a database.
Overview
The database is available here. It is an interactive table, where we summarize scientific data about 53 features potentially indicative of the capacity for valenced experience and examine the degree to which these features are found throughout 18 representative biological taxa. 12 invertebrate taxa are included: honey bees (genus Apis), cockroaches (genus Periplaneta), fruit flies (Drosophila melanogaster), ants (family Formicidae), spiders (order Araneae), the nematode Caenorhabditis elegans, sea hares (genus Aplysia), moon jellyfish (genus Aurelia), crabs (infraorder Brachyura), crayfish (family Cambaridae), and octopuses (family Octopodidae). For comparative purposes, we included three kinds of non-animal organisms –prokaryotes, protists and plants (kingdom Plantae)– and three vertebrate species –chickens (Gallus gallus domesticus), cows (Bos taurus), and humans (Homo Sapiens).
For each taxon and feature, we reviewed the existing literature and determined whether there was sufficient scientific data to make a call as to whether that taxon possesses the feature in question. Then, we evaluated the likelihood that the taxon possesses that feature. For those cases where we found direct evidence, we established four different responses corresponding to four rough probability ranges:
“LikelyNo”: representing credences of 0% - 25%
“Lean No”: 25% - 50%
“Lean Yes”: 50% - 75%
“Likely Yes”: 75% - 100%
As mentioned in a previous post, these four “credence buckets” represent our position regarding whether an animal possesses a feature; they do not necessarily represent the extent to which the animal possesses that feature. Given that current evidence about invertebrate consciousness is limited and unequal –i.e. some taxa have been much more studied than others, or specific features, unlike others, are well-studied phenomena– our credences are not necessarily well-calibrated in all cases. Further research is likely to contribute to revising and improving our confidence and reliability in our assessment of features potentially indicative of consciousness in different invertebrate taxa.
On top of that, there are several cases where no direct evidence of a specific feature for a given taxon was found. For those cases in which there have been only indirect studies and a specific feature was expected to be observed but, finally, no evidence arises, we used the category “not observed”. When no direct or indirect empirical evidence was found about a specific feature for a determined taxon, it was flagged as “unknown”. Additionally, a handful of features, such as average lifespan or brain neurons, are populated with short comments or numerical responses, as the case may be.
All the data compiled in the table is easily accessible: the taxa are presented in the columns and the features as rows. The consulted bibliography is linked to each cell, oftentimes with accompanying clarificatory quotes and/or commentary. In addition, our database offers a filter by taxon and attributes that enables users to filter and limit the data displayed within the table.
In addition to the information compiled in the database, an accompanying narrative report (part 1, part 2) summarizes our main findings about the extent to which the considered invertebrate taxa and other organisms display various features potentially indicative of consciousness.
Furthermore, we explored whether each feature can operate unconsciously in humans, since that might be a good defeater for that indicator or function when assessing consciousness in other non-human individuals. In this case, each feature was evaluated following the same procedure and nomenclature cited above. These findings are also compiled in the table, and they are further described in this narrative report.
We expect this work will enable animal welfare advocates to explore and consider relevant evidence about whether invertebrate individuals of a given species or taxon have the capacity for valenced experience. In forthcoming work, we discuss and analyze whether invertebrate welfare is a promising cause area and suggest possible next steps in this regard.
It was written by Daniela R. Waldhorn with contributions from Jason Schukraft, Peter Hurford, and Marcus A. Davis. The database was developed by Daniela R. Waldhorn, Jason Schukraft, Max Carpendale, Peter Hurford, and Marcus A. Davis. Haven King-Nobles, Brandon Perry, Gavin Taylor, Kyle York, Tom Houlden, Ronke Bankole also contributed research. Table programming was engineered by Timothy “Manu” Meixell. Ashley Francis, Gabrielle Hass, Riley Peebles, and Avi Iyer helped with copyediting. Thanks to Kieran Greig and Luke Muelhauser for important feedback. Thanks to Catherine Low for help with project coordination.
This is a crosspost from the new Animal Welfare Alignment Newsletter by Anima International. You can subscribe on Substack if you are interested in following these efforts. Audio reading also available on Substack.
The goals of this post are to:
1. Raise a question I see as crucially important to the goal of aligning AI to animal welfare...
Hello! I'm Justin Portela. I got hired by GWWC to make YouTube videos after AI in Context did such a kickass job.
My channel is using that same cinematic, high-production value beauty to talk about everything in the EA universe that isn't AI.
...
This is a linkpost for Request for Proposals: Research and Applied Work on Digital Minds.
I'm glad to announce a request for proposals for research and applied work on digital minds at Longview Ph...
Great work, thank you for doing this.
My only feedback is that it would be nice if there was a top-line "sentience score" for how sentient a creature is. It's hard to compare the sentience of cows to bees in a digestible way. Especially as an advocate and lay-person. For anyone though, I think this might make the information more digestible.
It might work something like this: a being would be granted the score of 1.0 if we categorize all taxa as a likely yes. They get "docked" points as the answers lean towards no. It would be great if this was weighted for how important a taxa is in determining sentience (if that's even something we could do).
Can we even weigh out how "important" some taxa are? Might this sort of system be helpful to others? Please excuse my ignorance and if this wasn't the best forum to voice this feedback!
Hi Sammy. I’m one of the researchers on Rethink Priorities’ invertebrate sentience team. Thanks for your comment. This is an issue our team has thought a lot about and plans to address explicitly in forthcoming work. I agree that our research would be more digestible if we provided an overall probability of sentience for each taxon. Unfortunately, assigning a “sentience score” is 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 her/his hands) to even justifiably begin such a scoring project. And because subjective experience is, well, subjective, strict calibration in this domain is necessarily impossible.
Despite the above difficulties, Rethink Priorities is considering reporting our best guesses about the probability of sentience for our studied taxa. We are still figuring out the best way to present these preliminary estimates. We want the estimates 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. One concern is that interested parties might skip straight to our (uncalibrated, somewhat unjustified, extremely speculative) numerical estimates without taking the time to understand the nuance and intricacy of the issue. Personally, I worry that assigning sentience scores sacrifices too much in the name of digestibility.
Nonetheless, it’s not as if our currently published findings are completely silent on the matter. Clearly, there is better evidence for sentience for cephalopods and arthropods than there is for annelids and nematodes. Stay tuned for our invertebrate welfare cause profile (slated to go up in late July) for more on the implications of our research.
Agree with the risks of presenting such a score. Agree that scores would be very speculative and your credibility intervals would be very wide. But I'd also guess that without this sort of score/summary measure, then it's very hard to use this research in practical applications?
Perhaps a compromise is to compile these sorts of summary scores, but then to only share them with advocates or researchers that have specific purposes in mind for those figures? This way, if someone wanted to use the summary score to inform an estimate, or make some decision based off of your research here, they could do so and it would only take them a few minutes to send you an email, rather than several hours trying to come up with an equivalent estimate, using your research as a starting point.
It's still possible that the figures would become more widely known if people find the numbers indirectly, e.g. in citations. But this seems unlikely to affect many people (unfortunately, I don't imagine this research going viral).
PS thanks very much for this very thorough seeming research on this difficult topic!
(Quick sidenote: "if we categorize all taxa as a likely yes"... it sounds like you're saying "taxa" are the features/rows, but "taxa" refers to groupings of animals. Sorry that the term is a bit unfamiliar.)