Extreme uncertainty in wild animal welfare requires resilient model-building

This piece is part of a short se­ries on how the Wild An­i­mal Ini­ti­a­tive re­search team thinks about un­cer­tainty. Th­ese posts rep­re­sent the opinions of in­di­vi­d­ual staff mem­bers, and not nec­es­sar­ily the or­ga­ni­za­tional po­si­tion of Wild An­i­mal Ini­ti­a­tive on un­cer­tainty. This sum­mer, we are rais­ing $50,000 to fur­ther our re­search—you can learn more here and sup­port our work here. This es­say is cross­posted from the WAI blog.


A re­cent ar­ti­cle offered new es­ti­mates of wild an­i­mal num­bers that differed for some taxa by an or­der of mag­ni­tude from Brian To­masik’s of­ten-cited es­ti­mates, high­light­ing the un­cer­tainty we face about some­thing as ba­sic as how many wild an­i­mals there are. In in­tro­duc­ing the prob­lem of wild an­i­mal suffer­ing, the au­thor wrote, “We can all agree for ex­am­ple that 10 units of suffer­ing by one in­di­vi­d­ual is much worse than 10 in­di­vi­d­u­als suffer­ing one unit of pain.” I was very sur­prised to read this. While the idea is in­tu­itive to me, I don’t con­sider the in­tu­ition jus­tified, and would class it un­der ‘ex­treme moral un­cer­tainty’. Uncer­tainty like this about fun­da­men­tal ethics is, in my view, the great­est road­block to large-scale in­ter­ven­tion for wild an­i­mal welfare. How­ever, there is also plenty of un­cer­tainty on more em­piri­cal ques­tions. For­tu­nately, these ques­tions seem much more tractable, and an­swers to some may re­veal ac­tions we can take that are ro­bustly good un­der a range of ethics. Here I will sum­ma­rize the ma­jor points of moral and prac­ti­cal un­cer­tainty that im­pact my own re­search at Wild An­i­mal Ini­ti­a­tive and how I think un­cer­tainty should be ad­dressed in this kind of work go­ing for­ward.


Which an­i­mals are moral pa­tients?

If two an­i­mals are both ca­pa­ble of ex­pe­rienc­ing suffer­ing and plea­sure, their in­ter­ests mat­ter equally. How­ever, in­di­vi­d­u­als of differ­ent species may have dra­mat­i­cally differ­ent needs and in­ter­ests. For ex­am­ple, for an an­i­mal with a greater ca­pac­ity for learn­ing there may be a greater va­ri­ety of stim­uli that can trig­ger a feel­ing of fear, or their psy­cholog­i­cal re­sponse to a sim­ple stim­u­lus like food de­pri­va­tion may be more com­plex if they have a larger reper­toire of be­hav­iors to ex­press, in­stincts to satisfy, or ex­pe­riences to as­so­ci­ate with. It is not clear how to weigh up these differ­ences. If in­di­vi­d­u­als of one species have fewer/​sim­pler prefer­ences, do those prefer­ences count more on an in­di­vi­d­ual ba­sis be­cause they con­sume more of the an­i­mal’s emo­tional at­ten­tion at any given time? For ex­am­ple, an an­i­mal with a longer time hori­zon, who is able to form prefer­ences about to­mor­row, might feel less strongly about a bad time to­day than an an­i­mal who can­not con­ceive of any­thing be­yond what they are ex­pe­rienc­ing in the pre­sent. This is ul­ti­mately an em­piri­cal ques­tion, whether some­thing as pri­mal as the feel­ing of hunger gen­er­ates more psy­cholog­i­cal stress in cog­ni­tively sim­pler an­i­mals.

The ca­pac­ity to differ­en­ti­ate en­joy­able and un­en­joy­able ex­pe­riences has of­ten been used as a proxy for moral pa­tient­hood. Much great work has gone into the ques­tion of which species have this ca­pac­ity; this is es­pe­cially im­por­tant to ad­dress for the most nu­mer­ous an­i­mal groups, such as the arthro­pods and ne­ma­todes. How­ever, there is also a ques­tion as to which de­vel­op­men­tal stages of a given species are moral pa­tients and to what de­gree. In­sects vastly out­num­ber ver­te­brates, and in most species, eggs vastly out­num­ber lar­vae which in turn vastly out­num­ber ma­ture an­i­mals. For ex­am­ple, while we may be con­fi­dent that adult fish are sen­tient, many of them pro­duce num­bers of eggs/​lar­vae ri­val­ing in­sects, the over­whelming ma­jor­ity of whom die within a few days. The an­swers to these ques­tions will de­ter­mine which in­ter­ven­tions we should ul­ti­mately pri­ori­tize.


A lin­ear scale of suffer­ing and plea­sure?

Re­turn­ing to the claim that “10 units of suffer­ing by one in­di­vi­d­ual is much worse than 10 in­di­vi­d­u­als suffer­ing one unit of pain,” let’s con­sider the lay­ers of un­cer­tainty fac­ing this state­ment, and their im­pli­ca­tions for study­ing wild an­i­mal welfare. First, the real-world mean­ing of a unit of suffer­ing is am­bigu­ous. A unit of suffer­ing may de­pend on an in­di­vi­d­ual’s per­cep­tion, as with the self-as­sessed pain scale used in hos­pi­tals. Un­for­tu­nately, most wild an­i­mals can’t self-as­sess, so it may be more prac­ti­cal to mea­sure suffer­ing based on some­thing like the du­ra­tion of a pleas­ant or un­pleas­ant ex­pe­rience – a day of hunger, a minute of sex, or a sec­ond of be­ing eaten al­ive. How an an­i­mal’s psy­cholog­i­cal well-be­ing scales with these real-world mea­sures pre­sents an­other layer of em­piri­cal un­cer­tainty on top of the con­tro­ver­sial pop­u­la­tion ethics.

Uncer­tainty sur­round­ing the scal­ing of suffer­ing against time or other met­rics has ma­jor im­pli­ca­tions for which pa­ram­e­ters are im­por­tant in mod­els of wild an­i­mal welfare and what ecolog­i­cal data we need to col­lect. For ex­am­ple, if ex­tremes of suffer­ing/​plea­sure are of ex­po­nen­tially greater con­cern than mod­er­ate val­ues, then the var­i­ance in welfare across a pop­u­la­tion or even across an in­di­vi­d­ual’s life­time may be even more im­por­tant than its av­er­age. Similarly, ex­po­nen­tially-scal­ing suffer­ing might re­quire us to place great im­por­tance on cause of death in mod­els of wild an­i­mal welfare based on the sever­ity of the ex­pe­rience, even if it only con­sumes a small frac­tion of an in­di­vi­d­ual’s life­time. As a ‘unit’ of suffer­ing or plea­sure re­mains poorly defined, it is also not clear whether the place­ment of these ex­pe­riences at op­po­site ends of a sin­gle scale of ‘welfare’ is eth­i­cally jus­tified or just a con­ve­nient way of sum­ma­riz­ing a col­lec­tion of af­fec­tive states.


How do cause-spe­cific mor­tal­ity rates in­ter­act with one an­other?

As Abra­ham Rowe dis­cussed in his re­cent post, non-tar­get effects of in­ter­ven­tions are a ma­jor con­cern (though not unique to wild an­i­mal welfare). Jane Capozzelli also high­lighted nu­mer­ous in­ter­ven­tions from the ‘Con­ser­va­tion Ev­i­dence Database’ rele­vant to wild an­i­mal welfare, men­tion­ing each of their po­ten­tial non-tar­get effects on other species in the same ecosys­tem.

At the pop­u­la­tion level, there is a similar need for re­search into po­ten­tial non-tar­get effects of in­ter­ven­tions aimed at re­duc­ing mor­tal­ity due to spe­cific causes – such as erad­i­cat­ing a dis­ease or ex­clud­ing preda­tors – or among spe­cific age groups. For ex­am­ple, if many of the an­i­mals who are kil­led by preda­tors would have died shortly there­after from hunger, then it is im­per­a­tive to un­der­stand the rel­a­tive sever­ity of these types of deaths. On the other hand, suffer­ing from hunger might leave an­i­mals vuln­er­a­ble to dis­ease or pre­da­tion, so re­duc­ing hunger might have an out­size effect on over­all mor­tal­ity rate. Any re­duc­tion in mor­tal­ity will also in­crease the pop­u­la­tion size by some amount, which may be good or bad de­pend­ing on whether the lives of those an­i­mals are typ­i­cally worth liv­ing.


How does pop­u­la­tion den­sity af­fect sur­vival and welfare?

If a wild pop­u­la­tion grows with­out ac­com­pa­ny­ing growth in its habitat or re­source availa­bil­ity, ev­ery in­di­vi­d­ual’s life may be­come a bit more challeng­ing. Thus, in­creas­ing pop­u­la­tion den­sity typ­i­cally re­duces life ex­pec­tancy as well as, pre­sum­ably, day-to-day qual­ity of life. Un­der­stand­ing the mag­ni­tude of this effect is cru­cial for pre­dict­ing the im­pact of any in­ter­ven­tion that would dis­pro­por­tionately af­fect ei­ther pop­u­la­tion size or habitat. The im­pli­ca­tions of pop­u­la­tion den­sity also in­ter­act with our un­cer­tainty about fun­da­men­tal ethics: where to draw the line be­tween net-pos­i­tive and net-nega­tive welfare, whether to weigh suffer­ing and plea­sure equally, and whether to value a pop­u­la­tion based on the sum, av­er­age, or some other statis­tic of its con­stituents’ welfare. This is be­cause av­er­age per-in­di­vi­d­ual welfare – and there­fore the prob­a­bil­ity that their welfare is net-pos­i­tive – will be higher in a low-den­sity pop­u­la­tion, but the num­ber of in­di­vi­d­u­als ex­pe­rienc­ing that welfare will be lower.


How to deal with uncertainty

For re­searchers, ex­treme un­cer­tainty re­quires re­silient model-build­ing. Welfare mod­els should be as philo­soph­i­cally open-ended as pos­si­ble so that the math works, and their out­put is in­for­ma­tive, on a range of views. Of course, there are limits to this: non-util­i­tar­i­ans might ob­ject in prin­ci­ple to quan­ti­ta­tive mod­els of welfare, and we can’t re­al­is­ti­cally ac­count for ev­ery con­ceiv­able moral virtue. An al­ter­na­tive ap­proach could be to build a com­pre­hen­sive library of more philo­soph­i­cally de­tailed mod­els. But apart from the ad­di­tional work­load of that ap­proach, at this early stage of welfare biol­ogy it may be es­pe­cially im­por­tant to de­velop rel­a­tively sim­ple paradig­matic mod­els, more for their abil­ity to unify the field and in­spire new re­search av­enues than for their im­me­di­ate real-world ap­pli­ca­tion. As the field ma­tures, the prac­ti­cal­ity of ex­plor­ing more pre­cise mod­els should in­crease.

An ex­am­ple of re­silient model-build­ing un­der un­cer­tainty comes from the early his­tory of an­other field: the Drake Equa­tion in as­tro­biol­ogy. This model at­tempts to es­ti­mate the num­ber of in­tel­li­gent, com­mu­ni­cat­ing civ­i­liza­tions in the galaxy based on a mix of tra­di­tional phys­i­cal pa­ram­e­ters and mas­sively un­cer­tain prob­a­bil­ity terms. One term rep­re­sents the “frac­tion of stars with hab­it­able planets.” The model’s con­tinued rele­vance to its field owes to the fact that it doesn’t rely on any spe­cific defi­ni­tion of hab­it­a­bil­ity, for ex­am­ple, al­low­ing the model to ac­com­mo­date new dis­cov­er­ies and ways of think­ing about each pa­ram­e­ter.

Welfare-fo­cused mod­els of ecolog­i­cal pro­cesses should fol­low a differ­ent ap­proach, be­ing as ex­plicit about biolog­i­cal mechanisms and in­di­vi­d­ual out­comes as pos­si­ble. Given our profound un­cer­tainty about whether most an­i­mals’ lives are dom­i­nated by plea­sure or pain, pop­u­la­tion sizes by them­selves can’t tell us much about welfare. How­ever, mod­els that break pop­u­la­tion num­bers down ac­cord­ing to life his­tory fea­tures like age/​lifes­pan or cause-of-death provide in­for­ma­tion that might be use­ful for un­der­stand­ing and im­prov­ing wild an­i­mal welfare. Even if we are not con­fi­dent whether the an­i­mals in ques­tion have lives worth liv­ing, these mod­els can help iden­tify the likely best and worst fates ex­pe­rienced by an­i­mals of a given species.

Prac­ti­cal con­sid­er­a­tions may also over­ride our eth­i­cal un­cer­tainty in the near term. Even if we are clas­si­cal util­i­tar­i­ans, who make no dis­tinc­tion be­tween re­duc­ing suffer­ing and in­creas­ing hap­piness, it will be eas­ier as we be­gin to learn about wild an­i­mal welfare to iden­tify sources of pain and how to re­duce them than it will be to iden­tify sources of plea­sure and how to en­hance them. In the longer term, when we are con­fi­dent enough to in­ter­vene, sci­en­tific un­der­stand­ing of ecol­ogy and how welfare varies in a pop­u­la­tion can help de­sign courses of ac­tion that are es­pe­cially re­silient to philo­soph­i­cal un­cer­tainty.