Invertebrate Sentience: A Useful Empirical Resource

Ex­ec­u­tive Summary

Re­think Pri­ori­ties re­viewed the sci­en­tific liter­a­ture rele­vant to in­ver­te­brate sen­tience. We se­lected 53 fea­tures po­ten­tially in­dica­tive of the ca­pac­ity for valenced ex­pe­rience and ex­am­ined the de­gree to which these fea­tures are found through­out 18 rep­re­sen­ta­tive biolog­i­cal taxa. Th­ese data have been com­piled into an eas­ily sortable database that will en­able an­i­mal welfare or­ga­ni­za­tions to bet­ter gauge the prob­a­bil­ity that (var­i­ous species of) in­ver­te­brates have the ca­pac­ity for valenced ex­pe­rience. This es­say de­tails what we’ve done, why, and the strengths and weak­nesses of our ap­proach.

Pro­ject Overview

This post is the sec­ond in our se­ries on in­ver­te­brate[1] welfare. In the first post we ex­am­ine some philo­soph­i­cal difficul­ties in­her­ent in the de­tec­tion of morally sig­nifi­cant pain and plea­sure in non­hu­mans. In the third, fourth, and fifth posts we ex­plain in de­tail the fea­tures we be­lieve to be most rele­vant for as­sess­ing in­ver­te­brate sen­tience. In the sixth, sev­enth, and eighth posts, we pre­sent our sum­mary of find­ings, both in nar­ra­tive form and as an in­ter­ac­tive database. In forth­com­ing work (to be pub­lished late July), we an­a­lyze the ex­tent to which in­ver­te­brate welfare is a promis­ing cause area.

We fo­cus on in­ver­te­brates for two rea­sons: (1) We are already rea­son­ably con­fi­dent that mam­mals, birds, rep­tiles, am­phibi­ans, and most fish[2] feel morally sig­nifi­cant pain and plea­sure, and hence must be in­cluded in our moral calcu­la­tions, but we are un­sure if more dis­tantly re­lated an­i­mals war­rant similar con­cern, and (2) The sub­ject of in­ver­te­brate welfare, though re­cently gain­ing trac­tion both in the sci­en­tific liter­a­ture and the effec­tive al­tru­ism com­mu­nity, ap­pears ne­glected rel­a­tive to the sheer num­ber[3] of po­ten­tially suffer­ing in­ver­te­brates.

To de­velop ac­cu­rate cost-benefit mod­els that can be used to al­lo­cate re­sources across the an­i­mal welfare move­ment, we need to take the pos­si­bil­ity of in­ver­te­brate pain and plea­sure se­ri­ously. But de­ter­min­ing whether in­ver­te­brates have the ca­pac­ity to ex­pe­rience pain and plea­sure in a morally sig­nifi­cant way is an ex­traor­di­nar­ily com­plex and difficult un­der­tak­ing. There is tremen­dous un­cer­tainty at vir­tu­ally ev­ery level at which one might in­ves­ti­gate the mat­ter. We don’t ex­pect to con­clude with high con­fi­dence that in­ver­te­brates do or do not ex­pe­rience morally sig­nifi­cant pain and plea­sure. Such an out­come is too am­bi­tious. Rather, our goal is to clearly map out the prob­lem so that we can be­gin to sys­tem­at­i­cally re­duce key un­cer­tain­ties in a cost-effec­tive man­ner.

One tractable way to im­prove our cre­dences with re­spect to in­ver­te­brate welfare is to cre­ate a com­pre­hen­sive col­lec­tion and anal­y­sis of the ex­tant sci­en­tific stud­ies rele­vant to the sub­ject. While some sci­en­tific stud­ies di­rectly ad­dress the is­sue of an­i­mal pain, the vast ma­jor­ity of stud­ies that are rele­vant at all to in­ver­te­brate welfare are rele­vant only tan­gen­tially. For ex­am­ple, stud­ies on cock­roach nav­i­ga­tion and place mem­ory don’t di­rectly ad­dress the is­sue of in­ver­te­brate welfare. Nonethe­less, the abil­ity to ac­com­plish cer­tain nav­i­ga­tional feats might be de­cent ev­i­dence that crea­tures with that abil­ity are con­scious, which is it­self a nec­es­sary con­di­tion on ex­pe­rienc­ing pain and plea­sure. How­ever, a search for “in­ver­te­brate welfare” on Google Scholar doesn’t de­liver any stud­ies on cock­roach nav­i­ga­tion. Gather­ing all the rele­vant sci­en­tific liter­a­ture in one place is thus a non­triv­ial and, to-date, un­ac­com­plished task.

Re­think Pri­ori­ties has spent the last ten months com­plet­ing just such a task. We have an­a­lyzed the de­gree to which more than 50 fea­tures po­ten­tially in­dica­tive of phe­nom­e­nal con­scious­ness[4] are found through­out 18 rep­re­sen­ta­tive biolog­i­cal taxa. For each pair of taxa (ants, say) and fea­tures (pro­tec­tive be­hav­ior, say), we first de­ter­mined whether there was suffi­cient sci­en­tific data to make a call as to whether the taxon in ques­tion pos­sesses the fea­ture. If so, we eval­u­ated the like­li­hood that the taxon pos­sesses the fea­ture. The stud­ies we con­sulted are linked to each cell, with ac­com­pa­ny­ing clar­ifi­ca­tory quotes and/​or com­men­tary. Th­ese data have been com­piled into a sortable, user-friendly database.

Pro­ject Rationale

In the re­main­der of this post, we aim to ex­plain four things: (a) why we built the database, (b) how we chose the fea­tures that we chose, (c) how we chose the taxa that we chose, and (d) what the limi­ta­tions of the database are.

The pro­ject was in­spired in part by the work of Luke Muehlhauser at the Open Philan­thropy Pro­ject. In his com­pre­hen­sive 2017 Re­port on Con­scious­ness and Mo­ral Pa­tient­hood, Muehlhauser briefly de­scribes the po­ten­tial use­ful­ness of a large refer­ence work that col­lects sci­en­tific data per­tain­ing to con­scious­ness from dis­parate fields into one place. This sug­ges­tion was the causally prox­i­mate im­pe­tus for the cur­rent pro­ject.

The pro­ject is im­por­tant be­cause it will al­low grant­mak­ers, char­ity en­trepreneurs, an­i­mal ac­tivists, and leg­is­la­tors to make bet­ter judg­ments about which an­i­mals have the ca­pac­ity for valenced ex­pe­rience.[5] Although other re­searchers have oc­ca­sion­ally tried to gauge the dis­tri­bu­tion of con­scious­ness through­out the an­i­mal king­dom, the pro­ject is unique in both breadth of taxa ex­am­ined and the num­ber of fea­tures in­ves­ti­gated. Ad­di­tion­ally, the pro­ject is open to con­tin­u­ous im­prove­ment. It is de­signed to al­low both for the in­cor­po­ra­tion of new sci­en­tific ev­i­dence as that ev­i­dence be­comes available and for the in­clu­sion of ad­di­tional biolog­i­cal taxa (or even non-biolog­i­cal en­tities) should such ad­di­tions be deemed use­ful.

Ar­gu­ment by Anal­ogy and In­fer­ence to the Best Explanation

The pro­ject can be used to help fa­cil­i­tate two im­por­tant ar­gu­ments of­ten de­ployed to in­ves­ti­gate con­scious­ness in non­hu­man an­i­mals. The first is ar­gu­ment by anal­ogy. The ba­sic struc­ture of an analog­i­cal ar­gu­ment is as fol­lows (where E1 is the source do­main and E2 is the tar­get do­main):

(1) En­tity E1 has some prop­er­ties P1 … Pn
(2) En­tity E2 has the same prop­er­ties P1 … Pn
(3) En­tity E1 has some fur­ther prop­erty Pn+1

(4) There­fore, en­tity E2 likely has the same prop­erty Pn+1

Our database can aid in the con­struc­tion of analog­i­cal ar­gu­ments. First choose some taxon from the database to serve as your source do­main. This should be a crea­ture you are rea­son­ably con­fi­dent is con­scious (cows, say). Then se­lect some fea­tures that the source do­main taxon pos­sesses that you think are rele­vant to con­scious­ness. Next se­lect some taxon to serve as your tar­get do­main. This should be a crea­ture for which you are rea­son­ably un­cer­tain whether it is con­scious (oc­to­puses, say). Fi­nally, check to see how many of the source do­main taxon fea­tures are shared by the tar­get do­main taxon. This sort of analog­i­cal rea­son­ing should also be run in re­verse. Start with some taxon you are rea­son­ably cer­tain is not con­scious (plants, say), then com­pare the fea­ture pro­file of this taxon with your origi­nal tar­get do­main taxon (in this ex­am­ple, oc­to­puses).

The other main type of ar­gu­ment for which our database could be use­ful is called in­fer­ence to the best ex­pla­na­tion. In­fer­ence to the best ex­pla­na­tion is re­lated to ar­gu­ment by anal­ogy, but for com­pli­cated the­o­ret­i­cal rea­sons, it is of­ten more pow­er­ful.[6] To see how in­fer­ence to the best ex­pla­na­tion works, con­sider how we know our fel­low hu­mans are con­scious. When I cut my hand, I cry out, I move my hand away from the sharp ob­ject, and I later treat the wound with a clean bandage. When other hu­mans cut their hands, they also cry out and at­tend to the sub­se­quent wounds in similar ways. There are a va­ri­ety of hy­pothe­ses which, if true, could ex­plain this be­hav­ior. Per­haps they are so­phis­ti­cated robots pro­grammed to be­have as I do. But the sim­plest and best ex­pla­na­tion of the be­hav­ior of other hu­mans is that they feel pain like I do.[7]

Of course, this ex­pla­na­tion might be mis­taken, and we might come to know it is mis­taken. If I ex­am­ined the heads of many fel­low hu­mans and in each case found not a brain but a crude ar­tifi­cial de­vice re­ceiv­ing sig­nals from a robotics fac­tory, that would con­sti­tute a defeater for my prior ex­pla­na­tion. I would then no longer be able to ra­tio­nally en­dorse the view that other hu­mans have men­tal states like I do. In­fer­ence to the best ex­pla­na­tion tells us that, in the ab­sence of defeaters, we are li­censed to pre­fer the sim­plest ex­pla­na­tion of a phe­nomenon.[8]

In­fer­ence to the best ex­pla­na­tion can also be ap­plied to non­hu­man an­i­mals. First one ought to de­ter­mine which com­bi­na­tion of fea­tures cor­re­lates to the be­hav­ioral pat­terns that, in con­scious crea­tures like hu­mans, are typ­i­cally caused by con­scious men­tal states like pain and plea­sure. Then see which taxa pos­sess this com­bi­na­tion of fea­tures (tak­ing care, of course, to al­low for the be­hav­ioral pat­terns to be ex­pressed in differ­ent ways by differ­ent taxa). Next, check for defeaters. Defeaters might come in any num­ber of guises. For ex­am­ple, one might think that or­ganisms be­low a cer­tain neu­rolog­i­cal com­plex­ity (mea­sured by brain size or neu­ron count) are in­ca­pable of con­scious states. One might think that failure to al­ter be­hav­ior in or­der to avoid in­ju­ri­ous stim­uli is a tel­l­tale sign that the or­ganism in ques­tion doesn’t feel pain. One might think that no­ci­cep­tors (spe­cial­ized cells used to de­tect po­ten­tially harm­ful events) are nec­es­sary for pain sen­sa­tion, in which case or­ganisms that lack no­ci­cep­tors are un­able to ex­pe­rience pain. No mat­ter one’s views, hav­ing all the rele­vant data eas­ily ma­nipu­la­ble and ac­cessible in one place will make ad­just­ing one’s cre­dences about the dis­tri­bu­tion of con­scious­ness eas­ier and bet­ter in­formed.

Which Fea­tures?

Our goal at the out­set was to cre­ate a database with the most com­pre­hen­sive set of fea­tures po­ten­tially rele­vant to valenced ex­pe­rience yet pub­lished. Even so, we had to make some hard choices to keep the list in the man­age­able range of ap­prox­i­mately 50-60 fea­tures. In this sec­tion we at­tempt to jus­tify the com­po­si­tion of the list. To sim­plify the mat­ter, we dis­cuss the list in terms of eight broad cat­e­gories: anatom­i­cal fea­tures, nox­ious stim­uli re­ac­tions, learn­ing in­di­ca­tors, cog­ni­tive so­phis­ti­ca­tion, mo­ti­va­tional trade­offs, mood state be­hav­iors, drug re­sponses, and nav­i­ga­tional skills. Be­fore con­sid­er­ing these cat­e­gories in de­tail, it’s helpful to first look at the cur­rent state of the liter­a­ture to see which fea­tures other similar pro­jects have in­ves­ti­gated.

The State of the Literature

The most in­fluen­tial re­cent dis­cus­sion of non­hu­man pain in the liter­a­ture is Sned­don et al.’s 2014 “Defin­ing and Assess­ing An­i­mal Pain.”[9] The au­thors pre­sent 17 fea­tures po­ten­tially rele­vant to pain per­cep­tion, then dis­cuss the ex­tent to which these fea­tures are found across 7 very broad taxa (see Table 1 be­low.)

Table 1, recre­ated from Sned­don et al. 2014

To our knowl­edge, this is the most com­pre­hen­sive, au­thor­i­ta­tive table in the liter­a­ture. Nonethe­less, the Sned­don table is limited in sev­eral ways. One way is­sue is the breadth of its seven taxa: mam­malia, aves, am­phibia/​rep­tilia, ag­natha/​os­te­ichthyes, cephalopoda, de­capods, and in­secta. Ac­cord­ing to the pa­per’s method­ol­ogy, so long as at least one species in the taxon ex­hibits the fea­ture in ques­tion, the rele­vant cell re­ceives a check­mark. This method­ol­ogy is prob­le­matic for two rea­sons. First, the taxa con­sid­ered are enor­mous. The last taxon, in­secta, con­tains more than six mil­lion species. A sin­gle species isn’t pos­si­bly rep­re­sen­ta­tive of the whole taxon. More im­por­tantly, this method­ol­ogy can pro­duce a mis­lead­ing table. By Sned­don et al.’s own lights, we should only think a species ex­pe­riences pain if they ex­hibit all or most of the fea­tures. But if 17 differ­ent species within a given taxon each ex­hibit a sin­gle differ­ent fea­ture, the column would be full of check­marks, giv­ing the mis­lead­ing im­pres­sion that the mem­bers of the taxon gen­er­ally satisfy the pre­sented defi­ni­tion of an­i­mal pain.

More broadly, we find that 17 fea­tures is still not enough data points to give a rel­a­tively the­ory-neu­tral pic­ture of an­i­mal con­scious­ness. Every in­ves­ti­ga­tion, in­clud­ing our own, has to make difficult de­ci­sions about which fea­tures to ex­am­ine, and these de­ci­sions will always be in­formed, to a cer­tain ex­tent, by the­o­ret­i­cal con­sid­er­a­tions about the na­ture of con­scious­ness. (No­body is go­ing to in­clude num­ber of ap­pendages as a fea­ture to be ex­am­ined be­cause no plau­si­ble the­ory of con­scious­ness holds that ap­pendage num­ber is rele­vant to con­scious­ness.) In gen­eral, though, the fewer fea­tures in­cluded in an in­ves­ti­ga­tion, the more the­ory that has to go into de­cid­ing which fea­tures are im­por­tant. The more fea­tures an in­ves­ti­ga­tion in­cludes, the fewer the­o­ret­i­cal de­ci­sions have to be made in ad­vance. We aimed to outdo the Sned­don table both in speci­fic­ity of taxa and num­ber of fea­tures to be in­ves­ti­gated.

We turn, then, to the fea­tures we in­cluded in our own in­ves­ti­ga­tion and their (broad) jus­tifi­ca­tion.

Anatom­i­cal Features

There are sev­eral anatom­i­cal fea­tures which may be rele­vant to the ca­pac­ity for valenced ex­pe­rience. No­ci­cep­tors are spe­cial­ized periph­eral sen­sory cells used by the body to de­tect po­ten­tially harm­ful stim­uli. We in­ves­ti­gated which of our tar­geted taxa pos­sess no­ci­cep­tors, in­clud­ing both a strict defi­ni­tion of no­ci­cep­tion and a loose defi­ni­tion of no­ci­cep­tion. Other anatom­i­cal fea­tures are neu­rolog­i­cal. We know that, in hu­mans, pain and plea­sure are pro­cessed in the brain, so we in­cluded brain size and brain neu­rons as fea­tures to be in­ves­ti­gated. The pres­ence of opi­oid-like re­cep­tors is po­ten­tially im­por­tant be­cause opi­oids are anal­gesics (painkil­lers). Opi­oids are es­pe­cially sig­nifi­cant be­cause, un­like other anal­gesics, opi­oids can re­duce the af­fec­tive (valenced) com­po­nent of pain with­out a cor­re­spond­ing re­duc­tion in the sen­sory as­pect (e.g., the burn­ing/​throb­bing/​cut­ting/​sting­ing/​aching qual­ity) of pain.[10] Although opi­oid re­cep­tors play many roles, their pres­ence is at least mild ev­i­dence that the crea­tures which pos­sess them have the ca­pac­ity for valenced ex­pe­rience. We in­cluded av­er­age lifes­pan for each taxon be­cause shorter-lived crea­tures in gen­eral en­counter fewer novel situ­a­tions and thus may be able to sur­vive and re­pro­duce uti­liz­ing only in­nate, pre-pro­grammed be­hav­iors.

Nox­ious Stim­uli Reactions

One ob­vi­ous strat­egy for de­tect­ing valenced ex­pe­rience in non­hu­man an­i­mals is to see how they re­spond to var­i­ous po­ten­tially pos­i­tive and nega­tive stim­uli.[11] There are far more sci­en­tific data rele­vant to po­ten­tial an­i­mal pain than to plea­sure, so for this broad cat­e­gory of fea­tures we fo­cused on re­ac­tions to nega­tive (here­after “nox­ious,” in keep­ing with the liter­a­ture) stim­uli. If a crea­ture doesn’t seem to re­act at all to a nox­ious event (such as get­ting eaten by an­other crea­ture), that is ev­i­dence that the crea­ture in ques­tion doesn’t ex­pe­rience the event as painful. How­ever, as we’ll see, just be­cause a crea­ture does re­act (by squirm­ing away, say) doesn’t mean that the event is nec­es­sar­ily ex­pe­rienced as painful.

One evolu­tion­ary func­tion of con­scious pain ex­pe­riences—per­haps the evolu­tion­ary func­tion of con­scious pain ex­pe­riences—is to pro­mote long-term bod­ily in­tegrity. Although not all dam­ag­ing in­juries are painful and not all pain ex­pe­riences re­sult from dam­ag­ing in­juries, there is nonethe­less a fairly tight cor­re­la­tion be­tween tis­sue dam­age and pain sen­sa­tion. When I prick my hand on a rose bush, it hurts. I don’t like the feel­ing, so I learn to be more care­ful around rose bushes in the fu­ture. In this way I pro­tect the in­tegrity of my hand.

How­ever, there is a differ­ent sys­tem the body also uti­lizes to avoid tis­sue dam­age, a sys­tem that op­er­ates be­low con­scious aware­ness: no­ci­cep­tion. No­ci­cep­tion is widely con­served through­out the an­i­mal king­dom. Even rel­a­tively sim­ple crea­tures, like the round­worm C. el­e­gans, pos­sess no­ci­cep­tors. No­ci­cep­tion en­ables the de­tec­tion and avoidance of po­ten­tially nox­ious stim­uli, and it does not have an at­ten­dant phe­nomenol­ogy with what­ever evolu­tion­ary cost such a phe­nomenol­ogy in­curs. So why would a crea­ture opt for both a no­ci­cep­tive sys­tem and con­scious pain ex­pe­riences?

The an­swer may be long-term mem­ory. No­ci­cep­tion is fast and re­flex­ive, but it is not nor­mally as­so­ci­ated with a long-last­ing mem­ory. On the other hand, pain ex­pe­riences, with their at­ten­dant felt bad­ness, tend to leave a memo­rial im­print. Be­cause pain ex­pe­riences are of­ten stored in long-term mem­ory, pain of­ten in­duces long-term be­hav­ioral and mo­ti­va­tional changes. For long-lived an­i­mals in com­plex en­vi­ron­ments, pain is thus po­ten­tially more effec­tive at pro­tect­ing the an­i­mal from dam­age than mere no­ci­cep­tion.

What pain and no­ci­cep­tion have in com­mon is that they are trig­gered, in the nor­mal case, by nox­ious stim­uli. We in­cluded sev­eral fea­tures in our database which give ev­i­dence that the crea­tures who pos­sess them are aware of nox­ious events. Move­ment away from nox­ious stim­uli is a very ba­sic fea­ture, found, we dis­cov­ered, even among some plants. Pro­tec­tive be­hav­ior (e.g., limp­ing, wound guard­ing, wound rub­bing, wound lick­ing) is ev­i­dence that an an­i­mal knows it has been in­jured. Nox­ious stim­uli re­lated vo­cal­iza­tion (e.g., scream­ing, cry­ing, groan­ing) is some­times taken to be a good met­ric of farmed an­i­mal welfare, and so we in­ves­ti­gated the ex­tent to which it is pre­sent el­se­where in the biolog­i­cal world. But the best ev­i­dence that no­ci­cep­tion leads to con­scious pain ex­pe­rience is not that an an­i­mal re­acts to nox­ious stim­uli but rather that these re­ac­tions lead to be­hav­ior al­ter­a­tion—that is, the crea­ture learns from these en­coun­ters.

Learn­ing Indicators

Some short-lived an­i­mals are so well-adapted to their evolu­tion­ary and ecolog­i­cal niche that no­ci­cep­tive re­sponses are suffi­cient to keep the an­i­mal safe from harm. An­i­mals in more com­plex situ­a­tions (ei­ther ecolog­i­cally or so­cially) need more plas­tic re­sponses, which re­quires learn­ing from the en­vi­ron­ment and/​or other an­i­mals. Ac­cord­ing to some re­searchers, be­hav­ioral plas­tic­ity is a hal­l­mark of con­scious­ness. For ex­am­ple, Paula Droege, a philoso­pher at Penn State, ar­gues that “an­i­mals ca­pa­ble of flex­ible re­sponses are con­scious an­i­mals, an­i­mals that feel con­scious pain.”[12] For this rea­son we looked at a num­ber of learn­ing in­di­ca­tors, in­clud­ing clas­si­cal con­di­tion­ing, op­er­ant con­di­tion­ing, and con­tex­tual as­so­ci­a­tion. Most im­por­tantly, we looked to see if ex­po­sure to po­ten­tially painful ex­pe­riences led to long-term be­hav­ior al­ter­a­tion.

Learn­ing in­di­ca­tors are po­ten­tially rele­vant in an­other way. On one com­mon un­der­stand­ing of con­scious­ness, a con­scious men­tal state is just a men­tal state one is aware of be­ing in. Be­ing aware of one’s own men­tal states would plau­si­bly help a crea­ture un­der­stand the men­tal states of oth­ers. Un­der­stand­ing the men­tal states of oth­ers plau­si­bly aids in the pro­ject of so­cial co­or­di­na­tion. Thus, the evolu­tion­ary role of con­scious­ness might be to help crea­tures in­ter­act, co­op­er­ate, and com­mu­ni­cate in more ad­vanced and adap­tive ways. For that rea­son we also in­ves­ti­gated the ex­tent to which differ­ent an­i­mals en­gaged in ob­ser­va­tional or so­cial learn­ing. Such be­hav­ior is also ev­i­dence of cog­ni­tive so­phis­ti­ca­tion.

Cog­ni­tive Sophistication

There are a num­ber of ways in which ev­i­dence of cog­ni­tive so­phis­ti­ca­tion might also be ev­i­dence for the ca­pac­ity for valenced ex­pe­rience. At the most gen­eral level, there ap­pears to be a cor­re­la­tion be­tween the ca­pac­ity for valenced ex­pe­rience and cog­ni­tive so­phis­ti­ca­tion. Hu­mans have the ca­pac­ity for valenced ex­pe­rience. Hu­mans are the most cog­ni­tively so­phis­ti­cated an­i­mals on the planet. Other cog­ni­tively so­phis­ti­cated an­i­mals, like apes and cetaceans, are also of­ten thought to pos­sess the ca­pac­ity for valenced ex­pe­rience.

Ad­di­tion­ally, some the­o­ries of con­scious­ness put pre­req­ui­sites on con­scious ex­pe­rience which seem able to be satis­fied only by rel­a­tively so­phis­ti­cated crea­tures. Ac­cord­ing to higher-or­der the­o­ries of con­scious­ness, con­scious ex­pe­rience re­quires meta-men­tal self-aware­ness. To have a con­scious de­sire for pineap­ple, say, it is not enough that one de­sires pineap­ple. (Some de­sires, as Freud taught us, are un­con­scious.) One must also pos­sess a higher-or­der men­tal state that rep­re­sents, in some form, one’s de­sire for pineap­ple. If con­scious­ness in­deed func­tions in this way, then cog­ni­tive so­phis­ti­ca­tion of a cer­tain de­gree may be a nec­es­sary con­di­tion on con­scious ex­pe­rience.[13]

Ele­ments of cog­ni­tive so­phis­ti­ca­tion can also be more di­rect ev­i­dence for con­scious ex­pe­rience. Gen­uine de­cep­tion, for in­stance, is a com­plex cog­ni­tive and so­cial skill.[14] To de­ceive, one must be ca­pa­ble of ap­pre­ci­at­ing the dis­tinc­tion be­tween pre­tense and re­al­ity. One must also be ca­pa­ble of rep­re­sent­ing, to some ex­tent, the men­tal states of the de­cep­tion tar­get. To de­ceive one must have a cog­ni­tive grasp on how things ap­pear, both to one­self and to one’s tar­get. To be aware of ap­pear­ances, one must ex­pe­rience them. A com­plex form of be­hav­ior in which an an­i­mal uses ap­pear­ances to de­ceive other an­i­mals is thus po­ten­tially fairly di­rect ev­i­dence that the an­i­mal is con­scious.[15]

Another ex­am­ple: tool use is of­ten con­sid­ered a bench­mark of cog­ni­tive so­phis­ti­ca­tion. Gen­uine tool use re­quires the ac­qui­si­tion of a for­eign ob­ject to be uti­lized at some later time.[16] In most cases of tool use, the ac­qui­si­tion or trans­porta­tion of the tool in­flicts some cost on the an­i­mal, for which the an­i­mal is com­pen­sated by some benefit when the tool is later de­ployed. Tool use there­fore re­quires a cer­tain de­gree of fore­sight and plan­ning, two el­e­ments of cog­ni­tive so­phis­ti­ca­tion.[17] On many un­der­stand­ings of con­scious­ness, such abil­ities are difficult to imag­ine with­out a cen­tral, unified men­tal space from which to eval­u­ate com­pet­ing in­ter­ests and de­mands. Such con­sid­er­a­tions bring us to mo­ti­va­tional trade­offs.

Mo­ti­va­tional Tradeoffs

This cat­e­gory of fea­tures is char­ac­ter­ized by be­hav­ior that in­di­cates the crea­ture in ques­tion is ca­pa­ble of weigh­ing com­pet­ing in­ter­ests and de­mands. For in­stance, when de­cid­ing whether to en­ter a par­tic­u­lar for­ag­ing site, an an­i­mal might weigh the threat of pre­da­tion against the need to find food to sur­vive. Similarly, when con­fronted with a com­peti­tor, an an­i­mal might weigh the benefits of main­tain­ing pos­ses­sion over a par­tic­u­lar ter­ri­tory against the dan­ger of suffer­ing phys­i­cal harm in a fight. Th­ese types of trade­offs in­di­cate the crea­ture has some sort of un­der­ly­ing unified util­ity func­tion ca­pa­ble of in­cor­po­rat­ing differ­ent streams of in­for­ma­tion. Such in­te­gra­tion is of­ten con­sid­ered a dis­tinc­tive at­tribute of con­scious­ness.[18]

Such trade­offs can be repli­cated, con­trol­led, and mon­i­tored in the lab. Many an­i­mals, in­clud­ing at least one species of in­ver­te­brate, are will­ing to pay a cost to re­ceive a re­ward. For ex­am­ple, fruit flies will en­dure elec­tric shock in or­der to at­tain the cue as­so­ci­ated with ethanol, in­di­cat­ing that they are pre­pared to tol­er­ate pun­ish­ment to ob­tain the drug. Con­versely, many an­i­mals, even some in­ver­te­brates, are will­ing to pay a cost to avoid a nox­ious stim­u­lus. Shore crabs gen­er­ally avoid well-lit ar­eas, prefer­ring to hide in dark en­vi­ron­ments such as those found un­der rocks. Given the choice be­tween two cham­bers in a lab­o­ra­tory set­ting, one brightly lit and the other dark, the crabs will uni­ver­sally choose the dark cham­ber. How­ever, if the dark cham­ber is rigged to de­liver a mild shock, the crabs will be­gin to opt for the nor­mally avoided well-lit cham­ber. The crabs do so in in­creas­ing num­bers (and in­creas­ingly quickly) as the num­ber of tri­als in­creases. A com­pel­ling ex­pla­na­tion of this be­hav­ior is that the crabs feel pain, then learn to avoid the pain by choos­ing the op­po­site, oth­er­wise un­de­sir­able cham­ber.

Some mo­ti­va­tional trade­offs take the form of **preda­tor avoidance **strate­gies. For ex­am­ple, in a lab­o­ra­tory set­ting, her­mit crabs will aban­don their shells if they are sub­jected to a mild shock. Ini­tially, such be­hav­ior was thought to be purely re­flex­ive. How­ever, re­cent ex­per­i­ments show the crabs are sig­nifi­cantly less likely to aban­don their shells af­ter shock if the odor of a preda­tor is pre­sent. The fact that the crabs re­main in their shells when the odor of a preda­tor is pre­sent sug­gests that the be­hav­ior is not re­flex­ive. A nat­u­ral ex­pla­na­tion is that the crabs weigh the pain of the shock against the fear of a preda­tor. But to truly un­der­stand whether in­ver­te­brates can ex­pe­rience fear (and other nega­tively valenced emo­tional states), we in­ves­ti­gated a va­ri­ety of mood state be­hav­iors.

Mood State Behaviors

Just as con­scious pain is plau­si­bly cor­re­lated with a cer­tain be­hav­ioral pro­file, so, too, should we ex­pect var­i­ous emo­tional states to gen­er­ate be­hav­ioral in­di­ca­tors. We ex­am­ined the ex­tent to which stereo­typic be­hav­ior and dis­place­ment be­hav­ior, two com­mon in­di­ca­tors of welfare prob­lems among cap­tive ver­te­brates, can be found in in­ver­te­brates. Stereo­typic be­hav­ior is a type of repet­i­tive ac­tivity that is un­vary­ing and ap­par­ently pur­pose­less. Ex­am­ples in­clude ex­ces­sive groom­ing lead­ing to self-mu­tila­tion, swim­ming in cir­cles, pac­ing, mouthing cage bars, and chew­ing with­out any­thing in the mouth (so-called “sham-chew­ing”). Stereo­typic be­hav­ior is prob­a­bly caused by im­pov­er­ished states, es­pe­cially situ­a­tions where the nor­mal func­tional re­sponse of an an­i­mal is blocked. Dis­place­ment be­hav­ior is an­other re­sponse to stress. Com­mon ex­am­ples of dis­place­ment be­hav­ior in­clude pac­ing, fid­get­ing, groom­ing, scratch­ing, and ob­ject ma­nipu­la­tion. Dis­place­ment ac­tivi­ties are thought to oc­cur when an or­ganism is caught be­tween con­flict­ing mo­ti­va­tions. (For in­stance, a bird un­able to de­cide whether it ought to flee or at­tack an op­po­nent might in­stead quickly preen or peck the ground.) Notably, dis­place­ment be­hav­ior is in­creased in a dose-de­pen­dent way when an or­ganism is given anx­iety-pro­vok­ing drugs and de­creased in a dose-de­pen­dent way when an or­ganism is given anx­iety-re­liev­ing drugs.[19]

We also looked at fear-like and anx­iety-like be­hav­ior in in­ver­te­brates. One ex­am­ple: stress-in­duced avoidance be­hav­ior in crayfish bears a strik­ing re­sem­blance to mam­malian anx­iety. A 2014 study demon­strated that shocked crayfish de­velop an ex­tended, con­text-in­de­pen­dent aver­sion to light.[20] (The shocks were not as­so­ci­ated with lev­els of illu­mi­na­tion.) In con­trast, un­shocked crayfish, though prefer­ring the dark, were happy to ex­plore both illu­mi­nated and unillu­mi­nated ar­eas of their en­vi­ron­ment. In hu­mans, anx­iety is of­ten as­so­ci­ated with dan­ger that is per­ceived to be un­avoid­able or situ­a­tions in which the threat is am­bigu­ous or un­known. The elec­tric shocks ap­plied to the crayfish fit this de­scrip­tion. In hu­mans, anx­iety is as­so­ci­ated with gen­er­al­ized fear, that is, in­creased fear of un­re­lated stim­uli. The shocked crayfish ap­peared to ex­hibit in­creased fear of light that is un­re­lated to the source of stress. Most im­por­tantly, in­ject­ing the crayfish with the anx­iolytic drug chlor­diazepox­ide (used to treat anx­iety in hu­mans) elimi­nated the aver­sion to light.[21] The fact that anx­iety-like be­hav­ior in crayfish is re­duced by a drug used to treat anx­iety in hu­mans sig­nifi­cantly bolsters the case that crayfish ex­pe­rience nega­tively valenced emo­tional states. Thus, look­ing at drug re­sponses may help us de­ter­mine which an­i­mals are ca­pa­ble of valenced ex­pe­rience.

Drug Responses

Some drugs al­ter the valence of ex­pe­rience in hu­mans. Anal­gesics and anx­iolyt­ics, for ex­am­ple, are used to treat nega­tively valenced ex­pe­riences, like pain and anx­iety. Many recre­ational drugs, in­clud­ing al­co­hol, nor­mally in­duce pos­i­tively valenced ex­pe­riences. If when ex­posed to these drugs non­hu­man an­i­mals ex­hibit be­hav­ior rele­vantly similar to the be­hav­ior of hu­mans ex­posed to the same drugs, then, in the ab­sence of defeaters, we have some ev­i­dence that those non­hu­man an­i­mals un­dergo similarly valenced ex­pe­riences. Thus, we in­ves­ti­gated which or­ganisms are af­fected by anal­gesics in a man­ner similar to hu­mans. We were par­tic­u­larly in­ter­ested in stud­ies that tested whether an an­i­mal will self-ad­minister anal­gesics be­causes these stud­ies can help us dis­t­in­guish cases in which an anal­gesic gen­uinely re­duces pain sen­sa­tion from cases in which an anal­gesic merely diminishes re­sponse to stim­uli in gen­eral.

We looked at stud­ies which tested var­i­ous recre­ational drugs and an­tide­pres­sants on non­hu­man an­i­mals and tried to dis­cern which an­i­mals ex­hibited rele­vantly similar be­hav­ior (in­clud­ing self-ad­minis­tra­tion). A wide va­ri­ety of recre­ational drugs have been tested on in­ver­te­brates, in­clud­ing am­phetamine in crayfish, mar­ijuana and caf­feine in spi­ders, and MDMA in oc­to­puses. One study on the effects of co­caine in honey bees found that bees quickly be­come ad­dicted to the drug de­spite the fact that it severely dam­ages their co­or­di­na­tion, lo­co­mo­tion, and mo­tor sys­tems. This is a sig­nifi­cant find­ing, given the im­por­tance of nav­i­ga­tional abil­ities to for­ag­ing honey bees. In­deed, the im­pres­sive nav­i­ga­tional skills of honey bees is one of the main rea­sons to think they might be con­scious.

Nav­i­ga­tional Skills

To sur­vive, an­i­mals need to avoid en­vi­ron­men­tal haz­ards and preda­tors. They need to lo­cate food, wa­ter, and mates. Many an­i­mals need to re­turn to fixed sites (such as dens, hives, breed­ing grounds, food caches, wa­ter­ing holes, for­ag­ing sites, or nest­ing beaches) mul­ti­ple times over the course of their lives. To do so re­quires nav­i­ga­tional skills.

Nav­i­ga­tional skills are rele­vant for the de­tec­tion of valenced ex­pe­rience in virtue of their pos­si­ble con­nec­tion to con­scious ex­pe­rience more gen­er­ally. Ac­cord­ing to some re­searchers, the evolu­tion­ary func­tion of con­scious­ness is to pro­duce an in­te­grated and ego­cen­tric spa­tial model to guide an an­i­mal as it nav­i­gates a com­plex en­vi­ron­ment. If con­scious­ness evolved speci­fi­cally to aid motile an­i­mals in com­plex en­vi­ron­ments, then var­i­ous nav­i­ga­tional feats could plau­si­bly be in­ter­preted as ev­i­dence of con­scious­ness in evolved crea­tures.

Peter Car­ruthers, a philoso­pher and cog­ni­tive sci­en­tist at the Univer­sity of Mary­land, writes, “at least some in­ver­te­brates (specifi­cally honey bees and jump­ing spi­ders) pos­sess a be­lief-de­sire-plan­ning cog­ni­tive ar­chi­tec­ture much like our own, as re­vealed by their so­phis­ti­cated nav­i­ga­tion abil­ities” (282). More­over, so­phis­ti­cated nav­i­ga­tional abil­ities, es­pe­cially those re­quiring de­tour be­hav­ior, are ev­i­dence of cog­ni­tive com­plex­ity, which, as was dis­cussed above, is of­ten taken to be rele­vant to con­scious­ness.[22] Spa­tial mem­ory is po­ten­tially rele­vant to con­scious­ness be­cause of its re­la­tion­ship to what Michael Trest­man, a philoso­pher of biol­ogy at In­di­ana Univer­sity, calls a “com­plex ac­tive body” (CAB). A CAB is ca­pa­ble of in­de­pen­dent, per­cep­tu­ally-guided, pow­ered-mo­tion (e.g., swim­ming, fly­ing, crawl­ing). Ac­cord­ing to Trest­man, the evolu­tion of CABs re­quires the ca­pac­ity for in­te­grated, em­bod­ied spa­tial cog­ni­tion, in­clud­ing spa­tial mem­ory. If one adopts a global workspace the­ory of con­scious­ness, a so­phis­ti­cated, in­te­grated form of cog­ni­tion (of which spa­tial mem­ory forms a part) of this sort is ev­i­dence for con­scious­ness.

Which Taxa?

To de­ter­mine which taxa to in­clude in our database, we faced two per­ti­nent ques­tions: (1) Which parts of the biolog­i­cal world would help us un­der­stand in­ver­te­brate sen­tience the best? and (2) At which tax­o­nomic rank ought we to in­ves­ti­gate those parts of the biolog­i­cal world? I’ll con­sider these ques­tions in turn.

Which Life­forms?

Ul­ti­mately, we want to know whether in­ver­te­brate welfare is a cause area worth pri­ori­tiz­ing within the an­i­mal welfare move­ment. In or­der to make that de­ter­mi­na­tion, we need to be able to ap­pro­pri­ately es­ti­mate the like­li­hood that in­ver­te­brates are ca­pa­ble of valenced ex­pe­rience. To get a han­dle on that like­li­hood, it’s not enough to merely in­ves­ti­gate in­ver­te­brates.

In­ver­te­brates rep­re­sent (at least as far as we are con­cerned) an edge case of con­scious­ness. There are plau­si­ble rea­sons to think that con­scious ex­pe­rience is con­fined to a rel­a­tively small cor­ner of evolu­tion­ary space (which ex­cludes in­ver­te­brates), but there are also plau­si­ble rea­sons to think that con­scious ex­pe­rience evolved very early and is con­served even among most in­ver­te­brate phyla.[23] When in­ves­ti­gat­ing an edge case, it’s a good idea to com­pare the edge case to clearer cases on ei­ther side. We wanted a database with the breadth of taxa re­quired to do that for con­scious­ness. To that end we in­cluded some fa­mil­iar ver­te­brate species: cows (Bos tau­rus), chick­ens (Gal­lus gal­lus do­mes­ti­cus), and of course hu­mans. Th­ese species give a sense for how com­mon the fea­tures we in­ves­ti­gated are in crea­tures widely con­sid­ered to be con­scious. We also in­cluded plants (king­dom Plan­tae), pro­tists, and prokary­otes to illus­trate how fre­quently these fea­tures are found in en­tities widely be­lieved to lack con­scious aware­ness.

In­clud­ing both pos­i­tive and nega­tive ex­am­ples will help iden­tify which fea­tures are rele­vant for de­tect­ing in­ver­te­brate con­scious­ness. For in­stance, there is some ev­i­dence that plants have the ca­pac­ity for as­so­ci­a­tive learn­ing. If that’s true, one might dis­count the rele­vance of as­so­ci­a­tive learn­ing in in­ver­te­brates, which is wide­spread.[24] On the other hand, even cows fail the clas­sic mir­ror test for self-recog­ni­tion. With that knowl­edge in mind, one might think the fact that no in­ver­te­brate has ever passed the test isn’t es­pe­cially im­por­tant for as­sess­ing whether in­ver­te­brates are con­scious.

When it came time to de­cide which in­ver­te­brates to in­ves­ti­gate, there were sev­eral fac­tors that went into our de­ci­sion: (a) how much sci­en­tific data is available for a given type of an­i­mal?, (b) how nu­mer­ous is a given type of an­i­mal?, (c) how likely is it that we could find ac­tion­able in­ter­ven­tions to pre­vent suffer­ing in a given type of an­i­mal?, (d) how prima fa­cie likely do we think it is that a given type of an­i­mal is con­scious?

We chose to in­ves­ti­gate sev­eral taxa that in­clude model or­ganisms from the in­ver­te­brate world, in­clud­ing the com­mon fruit fly, Drosophila melanogaster; the Cal­ifor­nia sea hare, Aplysia cal­ifor­nica; and the ne­ma­tode Caenorhab­di­tis el­e­gans. Th­ese species af­forded us a wealth of sci­en­tific data to an­a­lyze. We in­ves­ti­gated ants (fam­ily Formi­ci­dae) in part be­cause they are so nu­mer­ous, con­sti­tut­ing 15-20% of ter­res­trial an­i­mal bio­mass. We in­ves­ti­gated honey bees (genus Apis) in part be­cause they are ex­ten­sively ex­ploited by hu­mans.[25] We in­ves­ti­gated two de­ca­pod crus­taceans, crabs (in­fraorder Brachyura) and crayfish (fam­ily Cam­bari­dae), for similar rea­sons. We in­ves­ti­gated oc­to­puses (fam­ily Oc­topo­di­dae) be­cause oc­to­puses are widely con­sid­ered the most cog­ni­tively so­phis­ti­cated in­ver­te­brate. We wanted rep­re­sen­ta­tives from other com­mon in­ver­te­brate phyla, so we in­ves­ti­gated earth­worms (Lum­bri­cus ter­restris) from the phy­lum An­nelida and moon jel­lyfish (genus Aure­lia) from the phy­lum Cnidaria. To round out the in­ver­te­brate sec­tion of the database, we also in­ves­ti­gated cock­roaches (genus Peri­planeta) and spi­ders (or­der Araneae).

One fi­nal ‘taxon’ mer­its spe­cial dis­cus­sion. It is an em­piri­cal ques­tion which be­hav­iors are caused or ac­com­panied by con­scious ex­pe­rience and which are not. Even for sen­tient crea­tures such as hu­mans, a great many be­hav­iors, such as the patel­lar re­flex, will be me­di­ated solely by un­con­scious pro­cesses. If a be­hav­ior can be performed un­con­sciously by hu­mans, that’s some ev­i­dence that the same be­hav­ior does not op­er­ate con­sciously in non­hu­man an­i­mals. Dist­in­guish­ing which fea­tures can be performed un­con­sciously (and to what ex­tent) from those be­hav­iors which are always performed con­sciously is not easy. Thus, we de­cided to in­clude a column in our database ded­i­cated to an­swer­ing the ques­tion, for ev­ery fea­ture, ‘can this be­hav­ior be performed un­con­sciously in hu­mans?’ For each cell in this column we sum­ma­rize the best available sci­en­tific ev­i­dence for the fea­ture in ques­tion, of­ten with mixed re­sults. This column may make de­ter­min­ing the rel­a­tive im­por­tance of the fea­tures eas­ier.

What Tax­o­nomic Rank?

After de­cid­ing which types of an­i­mals we wanted to in­ves­ti­gate, we next needed to de­cide the tax­o­nomic level at which we would in­ves­ti­gate those an­i­mals. There are com­pet­ing con­sid­er­a­tions at play in this de­ci­sion. On the one hand, there is pres­sure to drill down to a fairly nar­row taxon (genus or species, say). 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.

On the other hand, there is also pres­sure to ex­am­ine each type of an­i­mal at the high­est tax­o­nomic rank that re­mains in­for­ma­tive. To a cer­tain ex­tent, we want our re­sults to gen­er­al­ize. There is lit­tle point in learn­ing that a par­tic­u­lar sub­species of jump­ing spi­der ex­pe­riences pain and plea­sure if we have no idea what this fact tells us about other arach­nids. Another, more prac­ti­cal, source of up­ward pres­sure was the amount of re­search which had been con­ducted at each tax­o­nomic rank. The higher up the tax­o­nomic hi­er­ar­chy one goes, the more re­search be­comes available for us to an­a­lyze. For all but the most com­monly stud­ied in­ver­te­brates (e.g., Drosophila melanogaster and Caenorhab­di­tis el­e­gans), it would have been difficult for us to fill in the database at the level of species.

Ul­ti­mately, our de­ci­sions were guided by a bal­ance of these con­sid­er­a­tions. With a few no­table ex­cep­tions, we opted for taxa that were rel­a­tively ho­mo­ge­neous, in hopes that in­for­ma­tion about in­di­vi­d­ual species within the taxon would gen­er­al­ize to other mem­bers of the taxon.[26] How­ever, where the quan­tity of available re­search was limited, we pushed up­ward the tax­o­nomic hi­er­ar­chy, some­times up to the level of or­der, for ex­am­ple with spi­ders (or­der Araneae). This is higher up the tax­o­nomic hi­er­ar­chy than we oth­er­wise would like to go, but given the rele­vant prac­ti­cal limits, it was the best we could do. As more re­search on in­ver­te­brate sen­tience be­comes available, we hope to be able to offer a more fine-grained anal­y­sis.

Pro­ject Limitations

In this sec­tion we de­scribe nine limi­ta­tions of our database.

Mor­gan’s Canon vs. the Pre­cau­tion­ary Principle

Our database alone can­not be used to draw any defini­tive con­clu­sions about the dis­tri­bu­tion of valenced ex­pe­rience among in­ver­te­brates (or any other part of the biolog­i­cal world). The in­for­ma­tion in the database must be suit­ably in­ter­preted ac­cord­ing to one’s fa­vored epistemic and prag­matic prin­ci­ples.

Ac­cord­ing to Mor­gan’s Canon,[27] an­i­mal be­hav­ior should not be ex­plained in terms of “higher” psy­cholog­i­cal pro­cesses (in our case, that means con­scious ex­pe­riences) if the be­hav­ior can be fairly ex­plained in terms of “lower” psy­cholog­i­cal pro­cesses (un­con­scious re­flexes and in­stincts). Mor­gan’s Canon does not pre­clude at­tribut­ing valenced ex­pe­rience to non­hu­man an­i­mals, so long as it can be shown that similar ex­pla­na­tions which do not in­voke valenced ex­pe­rience are in­ad­e­quate. Mor­gan’s Canon does, how­ever, place the bur­den of proof squarely on those who would at­tribute valenced ex­pe­rience to non­hu­man an­i­mals. Mor­gan’s Canon is an epistemic prin­ci­ple. It pro­vides a crite­rion for in­ter­pret­ing an­i­mal be­hav­ior. It aims to shape our be­liefs about an­i­mal con­scious­ness.

The pre­cau­tion­ary prin­ci­ple, on the other hand, is a a prag­matic prin­ci­ple and thus aims to guide our ac­tions. Ac­cord­ing to the pre­cau­tion­ary prin­ci­ple, when the ev­i­dence that an an­i­mal pos­sesses the ca­pac­ity for valenced ex­pe­rience is mixed, we should err on the side of cau­tion when in­ter­act­ing with the an­i­mal. When for­mu­lat­ing an­i­mal welfare strate­gies, the pre­cau­tion­ary prin­ci­ple tells us that, so long as there is some rea­son­able chance that an an­i­mal has the ca­pac­ity for valenced ex­pe­rience, we ought to af­ford the an­i­mal the pro­tec­tions and moral con­cerns that would be war­ranted if the an­i­mal did pos­sess the ca­pac­ity for valenced ex­pe­rience. Ac­cord­ing to the pre­cau­tion­ary prin­ci­ple, even if we be­lieve an an­i­mal lacks the ca­pac­ity for valenced ex­pe­rience, so long as there is a rea­son­able chance we are wrong, we ought to give the an­i­mal the benefit of the doubt by act­ing as if it did have the ca­pac­ity for valenced ex­pe­rience.

Mor­gan’s Canon and the pre­cau­tion­ary prin­ci­ple need not be in ten­sion. Mor­gan’s Canon tells us how we ought to set our cre­dences with re­spect to an­i­mal con­scious­ness. The pre­cau­tion­ary prin­ci­ple tells us where we ought to set the ev­i­den­tial bar for treat­ing an­i­mals as if they don’t pos­sess the ca­pac­ity for valenced ex­pe­rience. Taken to­gether, both Mor­gan’s Canon and the pre­cau­tion­ary prin­ci­ple can play a role in a well-in­formed cost-benefit anal­y­sis of the im­por­tance of in­ver­te­brate welfare. One might con­clude that the odds that in­ver­te­brates have the ca­pac­ity for valenced ex­pe­rience are low, but that the gains to be had from ad­dress­ing in­ver­te­brate welfare are po­ten­tially suffi­ciently high enough as to make it a cause area worth pri­ori­tiz­ing.

The Selec­tion of Fea­tures In­fluences the Results

The num­ber and com­po­si­tion of fea­tures se­lected for a pro­ject such as ours nec­es­sar­ily af­fect the broad con­tours of the in­ves­ti­ga­tion. This fact com­pli­cates in­ter­pre­ta­tion of our re­sults. It would be con­ve­nient if there were some func­tion that took us from the per­centage of fea­tures that a taxon satis­fies to a prob­a­bil­ity that an­i­mals in that taxon are con­scious. Un­for­tu­nately, with­out know­ing the nec­es­sary and suffi­cient con­di­tions on con­scious­ness (as­sum­ing there are such con­di­tions), there is no way to know which set of fea­tures would jus­tify this in­fer­ence over all po­ten­tially sen­tient or­ganisms. The type of ev­i­dence for sen­tience we do have available is not always ap­pli­ca­ble to all life­forms. For this pro­ject, we were ex­plic­itly aiming to eval­u­ate the ev­i­dence for sen­tience rele­vant to in­ver­te­brates. Hence, we did not want to con­sider fea­tures that would ob­vi­ously only be ap­pli­ca­ble to higher mam­mals.[28] So, for in­stance, we did not in­clude ‘abil­ity to pro­cess lan­guage’ or ‘abil­ity to re­port ex­pe­riences to sci­en­tists’ be­cause in­ver­te­brates are plainly not able to satisfy these fea­tures.[29] For this rea­son, we ad­vise against think­ing about our re­sults solely in terms of what per­centage of fea­tures a taxon satis­fies. For prac­ti­cal rea­sons, the se­lec­tion of fea­tures in a pro­ject such as ours will re­flect the in­ter­ests and em­phases of the re­searchers. Some­what ar­bi­trary de­ci­sions will then come to af­fect what per­centage of fea­tures a given taxon satis­fies.

Not All Fea­tures Are Rele­vant for All Species

There is prob­a­bly no set of fea­tures which is uni­ver­sally rele­vant for the de­tec­tion of valenced ex­pe­rience. Owing to anatom­i­cal, en­vi­ron­men­tal, and so­cial differ­ences among an­i­mals, valenced ex­pe­riences are apt to be ex­pressed in be­hav­iorally and neu­ro­biolog­i­cally dis­tinct ways. Differ­ent species of an­i­mal, af­ter all, are differ­ent. To take a triv­ial ex­am­ple: most of the time, when hu­mans are in pain, they gri­mace. But the hard ex­oskele­ton of an in­sect does not al­low for gri­mac­ing. Is the fact that in­sects don’t gri­mace (a small piece of) ev­i­dence that they don’t feel feel pain? Pre­sum­ably not. In­sects can’t gri­mace, so the fact that they don’t gri­mace is ir­rele­vant. More sub­tly, we might ex­pect an­i­mals in differ­ent en­vi­ron­ments to be sen­si­tive to differ­ent types of nox­ious stim­uli. Aquatic an­i­mals, for ex­am­ple, might not have no­ci­cep­tors that reg­ister ex­treme differ­ences in tem­per­a­ture be­cause they never have to deal with the threat of burn­ing.[30] But just be­cause aquatic an­i­mals can’t ex­pe­rience one type of pain doesn’t mean that they can’t ex­pe­rience pain at all. When con­struct­ing similar­ity ar­gu­ments ei­ther for or against a par­tic­u­lar species’ ca­pac­ity for valenced ex­pe­rience, one should keep in mind the en­vi­ron­ment in which the crea­ture evolved and how differ­ent evolu­tion­ary pres­sures might lead valenced ex­pe­riences to be ex­pressed differ­ently.[31]


Another worry is that the lifestyle and en­vi­ron­ment of phy­lo­ge­net­i­cally dis­tant an­i­mals may be so alien that their be­hav­ior can­not even be de­scribed with­out re­sort­ing to prob­le­matic an­thro­po­mor­phiz­ing. Honey bees, for in­stance, last shared a com­mon an­ces­tor with hu­mans about 797 mil­lion years ago. Honey bees have four dis­tinct life stages: egg, larva, pupa, and adult. The typ­i­cal worker bee does not usu­ally sur­vive longer than two or three months. Honey bees com­mu­ni­cate with each other via com­pli­cated dances; they use pheromones to sig­nal alarm, ori­en­ta­tion, and colony recog­ni­tion; they rely on ol­fac­tion to lo­cate novel food sources. Given these rad­i­cal differ­ences, is it pos­si­ble to de­tect honey bee emo­tions? Ac­cord­ing to some re­searchers, yes. Agi­tated honey bees allegedly ex­hibit pes­simistic cog­ni­tive bi­ases, a hal­l­mark of hu­man anx­iety. Hence, we have ev­i­dence for nega­tive emo­tional states in honey bees.

Such stud­ies should not be dis­missed out of hand. Honey bees differ dra­mat­i­cally from hu­mans, but we should not make the mis­take of sup­pos­ing that emo­tional similar­ity, at least to a cer­tain de­gree, is im­pos­si­ble. Still, it may be helpful to re­gard such stud­ies some­what skep­ti­cally. Terms like ‘anx­ious’ and ‘tense’ come loaded with an­thro­pocen­tric con­no­ta­tions. Ap­ply­ing these terms to in­ver­te­brates may in­evitably be mis­lead­ing.

The Liter­a­ture Might Be Bi­ased Toward Sur­pris­ing Results

The in­ves­ti­ga­tion of valenced ex­pe­rience in in­ver­te­brates is very much in its in­fancy. Be­cause there may be a gen­eral bias to­ward non-null ex­per­i­men­tal re­sults in the sci­ences, es­pe­cially for rel­a­tively im­ma­ture fields, we should be cau­tious about the con­clu­sions of any one study. There is rea­son to think that aca­demic jour­nals fa­vor pa­pers with sur­pris­ing re­sults over pa­pers which merely con­firm the ex­pected. In our case that might mean that stud­ies which pur­port to demon­strate novel be­hav­ior in in­ver­te­brates are over­rep­re­sented in the liter­a­ture. Repli­ca­tion stud­ies are, in gen­eral, un­der-re­warded in academia, so cor­rect­ing for this over­rep­re­sen­ta­tion may take years or even decades. Be­cause much of the liter­a­ture we can­vassed is so re­cent, some of the con­clu­sions we reached about in­ver­te­brate be­hav­ior will un­doubt­edly be proven mis­taken in the fu­ture.[32]

We Don’t Con­sider How Wide­spread a Fea­ture Is Within a Taxon

Ac­cord­ing to our method­ol­ogy, so long as a sin­gle species within a given taxon pos­sesses the fea­ture in ques­tion, the en­tire taxon counts as pos­sess­ing the fea­ture. We did not sys­tem­at­i­cally in­ves­ti­gate how wide­spread a fea­ture is within each taxon. Such an in­ves­ti­ga­tion would have been ei­ther ex­tremely time-in­ten­sive (if species-level data within a taxon are abun­dant) or, more likely, in­con­clu­sive (if species-level data within a taxon are sparse). For rel­a­tively small, rel­a­tively ho­mo­ge­neous taxa, this method­ol­ogy is un­prob­le­matic. How­ever, in some cases available re­search con­straints pushed us to con­sider taxa at a higher rank than we oth­er­wise would have liked. (For ex­am­ple, we in­ves­ti­gated spi­ders at the rank of or­der. Order Araneae con­tains ap­prox­i­mately 45,000 species.) In these cases, gen­er­al­iz­ing from a sin­gle species to the rest of the taxon is po­ten­tially prob­le­matic. At the ex­treme, a small num­ber of un­rep­re­sen­ta­tive species may bias the en­tire taxon. The case of plants (king­dom Plan­tae) is a good ex­am­ple. For in­stance, very few plants are ca­pa­ble of move­ment rapid enough to be de­tected with the naked eye, but these species are dis­pro­por­tionately rep­re­sented in our database.[33]

Most Fea­tures Come in Degrees

One of the most im­por­tant limi­ta­tions of our database is that many (per­haps most) of the fea­tures we in­ves­ti­gated come in de­grees. For ex­am­ple, be­cause cen­tral­iza­tion is not an all-or-noth­ing phe­nomenon, the ques­tion of whether some an­i­mal pos­sesses a cen­tral­ized ner­vous sys­tem is bet­ter an­swered with a scalar value than a “yes” or “no.” Un­for­tu­nately, our database is not equipped to provide such in­for­ma­tion. Our four cre­dence buck­ets (0-.25, .25-.50, .50-.75, .75-1.00) rep­re­sent our po­si­tion re­gard­ing whether an an­i­mal pos­sesses a fea­ture; they do not nec­es­sar­ily rep­re­sent the ex­tent to which the an­i­mal pos­sesses the fea­ture. For ex­am­ple, we might be­lieve with high cre­dence (.8) that a given an­i­mal pos­sesses a ner­vous sys­tem only 40% cen­tral­ized rel­a­tive to some rele­vant stan­dard. On the other hand, we might be­lieve with low cre­dence (.4) that a given an­i­mal pos­sesses a ner­vous sys­tem 80% cen­tral­ized rel­a­tive to some rele­vant stan­dard. Th­ese differ­ent epistemic po­si­tions might call for sub­tly differ­ent re­ac­tions. Un­for­tu­nately, to prop­erly ac­count for such differ­ences, we would need to in­clude a sec­ond axis for each fea­ture, an in­clu­sion which would have made the database sig­nifi­cantly more com­pli­cated. We opted for the sim­plic­ity of a sin­gle axis sys­tem over the added nu­ance of a two-axis sys­tem.[34]

It’s Some­times Difficult to Ascer­tain Whether a Crea­ture Possesses a Feature

Even when a fea­ture doesn’t come in de­grees, it is of­ten difficult to de­ter­mine with high con­fi­dence if an an­i­mal pos­sesses the fea­ture un­der in­ves­ti­ga­tion. Some­times this un­cer­tainty stems from a dearth of sci­en­tific data. (In these cases, the rele­vant cells have been marked ‘un­known.’) More of­ten, a fair amount of re­search is available, but the data are in­con­clu­sive for one or more rea­sons. Self-recog­ni­tion, for in­stance, is plau­si­bly an im­por­tant fea­ture to in­ves­ti­gate. (If self-recog­ni­tion im­plies self-aware­ness, then on some the­o­ries of mind, self-recog­ni­tion is pretty di­rect ev­i­dence of con­scious­ness.) Self-recog­ni­tion in non­hu­man an­i­mals is gen­er­ally mea­sured via the mir­ror test. The test as­sesses self-recog­ni­tion by de­ter­min­ing whether an an­i­mal can rec­og­nize its own re­flec­tion in a mir­ror. This is ac­com­plished by se­cretly mark­ing the an­i­mal with a small dot that is only visi­ble by look­ing in the mir­ror. If the an­i­mal sees the dot in the mir­ror then touches the dot on its body (in­di­cat­ing it un­der­stands the re­la­tion­ship be­tween it­self and the crea­ture in the mir­ror), it passes the test.

The mir­ror test has sev­eral well-known method­olog­i­cal short­com­ings. For ex­am­ple, be­cause the test uses vi­sual per­cep­tion, crea­tures which per­ceive the world pri­mar­ily through other sense modal­ities are at a dis­ad­van­tage. (Dogs, who have re­peat­edly failed the clas­sic mir­ror test, nav­i­gate the world mostly by a com­bi­na­tion of ol­fac­tion and au­di­tion.) Crea­tures for whom eye con­tact is a sign of ag­gres­sion also seem to be at a dis­ad­van­tage be­cause they will ei­ther re­fuse to di­rectly in­ves­ti­gate their re­flec­tion, or, if they do make pro­tracted eye con­tact, will move to coun­ter­act the per­ceived ag­gres­sion of the ‘for­eign’ crea­ture be­fore they have an op­por­tu­nity to no­tice the dot. (Go­rillas, who are mostly re­ported to fail the mir­ror test, fit this pro­file.)

Th­ese flaws lead to false nega­tives. But false pos­i­tives are also pos­si­ble. The mir­ror test is an im­perfect mea­sure of self-recog­ni­tion. But even if the mir­ror test were a perfect mea­sure, it would still be difficult in some cases to defini­tively say whether some an­i­mal passes the test. For in­stance, ac­cord­ing to a re­cent study, a species of bony fish (the bluestreak cleaner wrasse, Labroides dimi­di­a­tus) passes the mir­ror test, the first an­i­mal out­side birds and mam­mals to do so. From our non-spe­cial­ist per­spec­tive, it is un­clear how to eval­u­ate such a new, con­tro­ver­sial find­ing, es­pe­cially given the small sam­ple size of the study. Thus, for this fea­ture and many oth­ers, we have been forced to make some difficult judg­ment calls. Again, we have tried to in­di­cate as much in com­ments at­tached to the rele­vant cells.

It’s Un­clear How Much Ev­i­den­tial Weight to As­sign the Features

Not all fea­tures provide equal ev­i­dence for the ca­pac­ity for valenced ex­pe­rience, and the range in ev­i­den­tial power among the fea­tures is prob­a­bly fairly ex­treme. (In­deed, some fea­tures might not provide any ev­i­dence for the ca­pac­ity for valenced ex­pe­rience.) Know­ing the ev­i­den­tial power of the fea­tures would make the con­struc­tion of sound similar­ity ar­gu­ments much eas­ier. Nonethe­less, we have made no at­tempt to as­sign ev­i­den­tial weights to the fea­tures. The rea­son we have not done so is be­cause such a task is too am­bi­tious to com­plete within the time con­straints we im­posed on this pro­ject. As­sign­ing ev­i­den­tial weights is a difficult task, re­quiring more sci­en­tific in­ves­ti­ga­tion and philo­soph­i­cal re­flec­tion than we are cur­rently pre­pared to un­der­take. Even af­ter such an un­der­tak­ing, chances are that the weights we could con­fi­dently as­sign the fea­tures would come in wide ranges, un­der­min­ing their use­ful­ness. In this post we have briefly ex­plained why we in­ves­ti­gated the fea­tures that we did. Th­ese jus­tifi­ca­tions offer hints as to our ten­ta­tive views on the im­por­tance of var­i­ous fea­tures. Th­ese ten­ta­tive views, how­ever, are too hazy and in­com­plete to eas­ily sys­tem­atize in a way that would al­low for fruit­ful cross-com­par­i­son of fea­tures. As­sign­ing ev­i­den­tial weights to these fea­tures is an en­deavor we leave to other re­searchers (in­clud­ing, po­ten­tially, our fu­ture selves).

Con­clu­sion and Forth­com­ing Work

This post sum­ma­rizes Re­think Pri­ori­ties’ effort to bet­ter un­der­stand the is­sue of in­ver­te­brate sen­tience. We cre­ated a com­pre­hen­sive database of the ex­tant sci­en­tific liter­a­ture rele­vant to in­ver­te­brate sen­tience, cov­er­ing 53 fea­tures across 18 biolog­i­cal taxa. This database will help oth­ers gauge the dis­tri­bu­tion of sen­tience out­side the sub­phy­lum Ver­te­brata. In forth­com­ing work, we draw on this database (as well as a va­ri­ety of other in­puts) to an­a­lyze whether in­ver­te­brate welfare is a promis­ing cause area.


This es­say is a pro­ject of Re­think Pri­ori­ties. It was writ­ten by Ja­son Schukraft with con­tri­bu­tions from Daniela R. Wald­horn, Mar­cus A. Davis, Peter Hur­ford, and Max Carpen­dale. Kim Cud­ding­ton, Mar­cus A. Davis, Neil Dul­laghan, Peter Hur­ford, Te­gan McCaslin, David Moss, and Daniela R. Wald­horn pro­vided helpful com­ments on this es­say. If you like our work, please con­sider sub­scribing to our newslet­ter. You can see all our work to date here.


  1. Ver­te­brates con­sti­tute a sub­phy­lum in the phy­lum Chor­data. Cladis­ti­cally, it would be more pre­cise to speak of ‘chor­dates’ and ‘non-chor­dates.’ In us­ing the terms ‘ver­te­brates’ and ‘in­ver­te­brates’ we defer to com­mon us­age. How­ever, the num­ber of in­ver­te­brates in the phy­lum Chor­data is triv­ial com­pared to the num­ber of in­ver­te­brates out­side Chor­data, so com­mon us­age is not wholly in­ac­cu­rate. ↩︎

  2. Elas­mo­branch fish (i.e., car­tilag­i­nous fish, such as sharks) may be an ex­cep­tion be­cause they ap­pear to lack no­ci­cep­tors. See, in­ter alia, Ewan Smith and Gary Lewin. 2009. “No­ci­cep­tors: A Phy­lo­ge­netic View.” Jour­nal of Com­par­a­tive Phys­iol­ogy A Vol. 195, Is­sue 12: 1096. ↩︎

  3. See Table S1 in Bar-On, Y. M., Phillips, R., & Milo, R. (2018). The bio­mass dis­tri­bu­tion on Earth. Pro­ceed­ings of the Na­tional Academy of Sciences, 115(25), 6506-6511. ↩︎

  4. We use the terms ‘sen­tience,’ ‘phe­nom­e­nal con­scious­ness,’ and ‘sub­jec­tive ex­pe­rience’ in­ter­change­ably. An or­ganism is sen­tient just in case there is some­thing it is like to be that or­ganism. ‘Valenced ex­pe­rience’ de­notes a proper sub­set of con­scious ex­pe­rience in which ex­pe­riences take on a pos­i­tive or nega­tive af­fect. All crea­tures with the ca­pac­ity for valenced ex­pe­rience are nec­es­sar­ily sen­tient, but not all sen­tient crea­tures nec­es­sar­ily have the ca­pac­ity for valenced ex­pe­rience. Note: ‘sen­tience’ gets used in differ­ent ways by differ­ent philo­soph­i­cal com­mu­ni­ties. In philos­o­phy of mind, the term is nor­mally used in its broad sense, to mean ‘phe­nom­e­nal con­scious­ness.’ (See, in­ter alia, this SEP ar­ti­cle on an­i­mal con­scious­ness.) In moral philos­o­phy, the term is nor­mally used in its nar­row sense, to mean ‘valenced ex­pe­rience.’ (See, in­ter alia, this SEP ar­ti­cle on the grounds of moral sta­tus.) We have adopted the philos­o­phy of mind us­age. ↩︎

  5. The pro­ject could also be of use to philoso­phers and neu­ro­scien­tists study­ing con­scious­ness more gen­er­ally. ↩︎

  6. Analog­i­cal ar­gu­ments are in­duc­tive. In­fer­ence to the best ex­pla­na­tion, by con­trast, is ab­duc­tive. Ab­duc­tive ar­gu­ments are non-de­duc­tive like tra­di­tional in­duc­tive ar­gu­ments, but, un­like tra­di­tional in­duc­tive ar­gu­ments which are jus­tified em­piri­cally, ab­duc­tive ar­gu­ments are jus­tified a pri­ori. We are jus­tified in us­ing in­duc­tion be­cause, as a mat­ter of con­tin­gent fact, in­duc­tion has worked well in the past. In­stances of ab­duc­tive rea­son­ing, in con­trast, are gen­er­ally held to in­stan­ti­ate prin­ci­ples of ra­tio­nal­ity, which, if they are known at all, are known a pri­ori. ↩︎

  7. Other ex­pla­na­tions are more com­pli­cated be­cause they raise more ques­tions than they re­solve. Why, for in­stance, would some­one cre­ate so­phis­ti­cated robots pro­grammed to be­have as I do? ↩︎

  8. It’s im­por­tant to note that one can pre­fer an ex­pla­na­tion with­out fully be­liev­ing the ex­pla­na­tion. If there are nu­mer­ous plau­si­ble ex­pla­na­tions, the best ex­pla­na­tion might only war­rant a cre­dence of .2. For ex­am­ple, it’s con­sis­tent to have a fairly low cre­dence in the claim that in­ver­te­brates feel pain and yet think that that ex­pla­na­tion of their be­hav­ior is more likely than any other ex­pla­na­tion of their be­hav­ior. See Michael Tye. 2017. Tense Bees and Shell-Shocked Crabs. Oxford Univer­sity Press: 68. ↩︎

  9. See also Robert El­wood et al.’s 2009 “Pain and Stress in Crus­taceans?”, Vic­to­ria Braith­waite’s 2010 book Do Fish Feel Pain?, Gary Varner’s 2012 book Per­son­hood, Ethics, and An­i­mal Cog­ni­tion, Todd Fein­berg and Jon Mal­latt’s 2016 book The An­cient Ori­gins of Con­scious­ness, and Luke Muehlhauser’s Open Philan­thropy re­port cited ear­lier. ↩︎

  10. See Adam Shriver. 2006. “Mind­ing Mam­mals.” Philo­soph­i­cal Psy­chol­ogy Vol. 19: 433-442 (es­pe­cially §2) for an ac­cessible overview. ↩︎

  11. Here I’m un­der­stand­ing “pos­i­tive stim­u­lus” as an event that is, in some sense, ob­jec­tively good for the or­ganism, such as the con­sump­tion of a nu­tri­ent-dense food source, and “nega­tive stim­u­lus” as an event that is, in some sense, ob­jec­tively bad for the or­ganism, such as an at­tack from a preda­tor which causes bod­ily dam­age. ↩︎

  12. Paula Droege. 2017. “The Lives of Others: Pain in Non-Hu­man An­i­mals,” in Jen­nifer Corns (ed.) The Rout­ledge Hand­book of Philos­o­phy of Pain: 196. She adds, “dis­t­in­guish­ing flex­ible from fixed re­sponse is not a sim­ple mat­ter. The ba­sic idea be­hind flex­i­bil­ity is that an an­i­mal no longer sim­ply acts based on past as­so­ci­a­tions; it gen­er­al­izes on past learn­ing to an­ti­ci­pate which sort of ac­tion is best” (ibid.). ↩︎

  13. Con­scious­ness may also be a byproduct of a cer­tain de­gree of cog­ni­tive so­phis­ti­ca­tion. ↩︎

  14. Gen­uine de­cep­tion (also some­times called “tac­ti­cal” or “in­ten­tional” de­cep­tion) is dis­tinct from pas­sive adap­ta­tions, such as mimicry or cam­ou­flage. In prac­tice, dis­t­in­guish­ing gen­uine de­cep­tion from mere as­so­ci­a­tive con­di­tion­ing will be difficult (but not im­pos­si­ble). ↩︎

  15. A po­ten­tial ex­am­ple of this type of de­cep­tion is cut­tlefish mat­ing de­cep­tion. “Males de­ceive ri­val males by dis­play­ing male courtship pat­terns to re­cep­tive fe­males on one side of the body, and si­mul­ta­neously dis­play­ing fe­male pat­terns to a sin­gle ri­val male on the other, thus pre­vent­ing the ri­val from dis­rupt­ing courtship.” ↩︎

  16. As such, the makeshift shells some­times ac­quired by her­mit crabs do not qual­ify as tool use be­cause the shell is effec­tively in use the whole time the crab in­hab­its it. ↩︎

  17. See, for ex­am­ple, the case of soft-sed­i­ment dwelling oc­to­puses re­triev­ing co­conut shell halves dis­carded by the lo­cal hu­man pop­u­la­tion and later as­sem­bling the shell halves into pro­tec­tive shelters. The awk­ward man­ner in which the oc­to­puses must move while car­ry­ing these shells (the au­thors de­scribe it as “stilt-walk­ing”) rep­re­sents a cost in terms of en­ergy and in­creased pre­da­tion risk, which is only re­couped later when the shelves are suc­cess­fully as­sem­bled into a sur­face shelter or en­cap­su­lat­ing lair. Im­por­tantly, the only known source of these clean and lightweight shells is the coastal hu­man com­mu­ni­ties, and thus the oc­to­puses have not in­ter­acted with these items on an evolu­tion­ary timescale. ↩︎

  18. Of course, some trade­off be­hav­ior is sure to be purely re­flex­ive, in­stinc­tual, or oth­er­wise pre-pro­grammed. Trade­off be­hav­ior that oc­curs in novel situ­a­tions demon­strates the plas­tic­ity that one would ex­pect of a crea­ture with the ca­pac­ity for valenced ex­pe­rience. Trade­off be­hav­ior that oc­curs in re­sponse to situ­a­tions re­peat­edly en­coun­tered in the wild does not re­quire the same level of plas­tic­ity and hence may not re­quire the ca­pac­ity for valenced ex­pe­rience. ↩︎

  19. Mered­ith Root-Bern­stein. 2010. “Dis­place­ment Ac­tivi­ties dur­ing the Honey­bee Tran­si­tion from Wag­gle Dance to For­ag­ing.” An­i­mal Be­havi­our 79 (4) : 935–38. ↩︎

  20. Fos­sat, P., Bac­qué-Cazenave, J., De Deur­waerdère, P., Delbecque, J. P., & Cat­taert, D. (2014). Anx­iety-like be­hav­ior in crayfish is con­trol­led by sero­tonin. Science, 344(6189), 1293-1297. Note that the pop­u­lar in­tro­duc­tion to the ar­ti­cle mis­tak­enly states that the crayfish were afraid of the dark, not the light. A cor­rected ti­tle for the pop­u­lar in­tro­duc­tion is available here. ↩︎

  21. Separately, the in­jec­tion of sero­tonin in un­shocked crayfish in­duced light-aver­sion be­hav­ior that was also elimi­nated by chlor­diazepox­ide. In hu­mans, ele­vated lev­els of sero­tonin is as­so­ci­ated with anx­iety. ↩︎

  22. The strength of the ev­i­dence is un­der­cut some­what by stud­ies which pur­port to show that uni­cel­lu­lar slime molds can ‘solve’ rel­a­tively com­plex mazes. ↩︎

  23. There is an­other pos­si­bil­ity, that con­scious­ness evolved in­de­pen­dently among cer­tain ver­te­brate and in­ver­te­brate lineages. It might turn out, for ex­am­ple, that cephalo­pod con­scious­ness evolved in­de­pen­dently of mam­malian con­scious­ness. ↩︎

  24. Of course, one might just take this as (slight) ev­i­dence that plants are con­scious. ↩︎

  25. At any given time, there are hun­dreds of billions of do­mes­ti­cated honey bees across the globe. By one es­ti­mate there are more do­mes­ti­cated honey bees than all other ter­res­trial do­mes­ti­cated an­i­mals com­bined. ↩︎

  26. The three big ex­cep­tions are plants, pro­tists, and prokary­otes. ↩︎

  27. Mor­gan’s Canon is a spe­cial case of the more gen­eral prin­ci­ple of par­si­mony, as ap­plied to an­i­mal psy­chol­ogy. ↩︎

  28. To analo­gize: if you were in­ves­ti­gat­ing the ev­i­dence for sen­tience in hu­mans, you wouldn’t in­clude the fea­ture ‘can re­port ex­pe­riences in English’ be­cause that fea­ture is plainly only ap­pli­ca­ble to a sub­set of hu­mans. ↩︎

  29. Th­ese fea­tures are plau­si­bly im­por­tant, and the fact that in­ver­te­brates don’t satisfy them is per­haps a mark against them, but the in­clu­sion of these fea­tures would not have added any­thing of value to the pro­ject. See §3.2.2 of Luke Muehlhauser’s Open Phil re­port for more cog­ni­tive fea­tures we chose not to in­clude. ↩︎

  30. See Lynne Sned­don. 2015. “Pain in Aquatic An­i­mals.” Jour­nal of Ex­per­i­men­tal Biol­ogy 2015 218: 967-976; doi: 10.1242/​jeb.088823 http://​​jeb.biol­o­​​con­tent/​​218/​​7/​​967 ↩︎

  31. At the ex­treme, one might worry that very few of the fea­tures we in­ves­ti­gated are rele­vant for crea­tures that lack an evolu­tion­ary his­tory (i.e., var­i­ous forms of ar­tifi­cial in­tel­li­gence). ↩︎

  32. This con­cern is well-sum­ma­rized in Muehlhauser (2017), Ap­pendix Z8. ↩︎

  33. e.g., Mi­mosa pu­dica, Dion­aea mus­cipula, Morus alba, and Co­dar­i­o­ca­lyx mo­to­rius ↩︎

  34. How­ever, where im­por­tant qual­ifi­ca­tions were war­ranted, we added ap­pro­pri­ate com­ments (in the form of tooltips and ap­pen­dices) to the database. ↩︎