A review of what affective neuroscience knows about suffering & valence. (TLDR: Affective Neuroscience is very confused about what suffering is.)

Part 1

A sig­nifi­cant frac­tion of the EA move­ment is con­cerned with suffer­ing, and all else be­ing equal, think there should be less of it. I think this is an ex­traor­di­nar­ily no­ble goal.

But what *is* suffer­ing? There are roughly as many work­ing defi­ni­tions of suffer­ing in the EA move­ment as there are fac­tions in the EA move­ment. Wor­ry­ingly, these defi­ni­tions of­ten im­plic­itly or ex­plic­itly con­flict, and only the fact that the EA pie is grow­ing rel­a­tively rapidly pre­vents a de­scent into fac­tional war­fare over re­sources be­ing wasted on ‘in­cor­rect’ un­der­stand­ings of suffer­ing.

In­tu­itively, one would hope that grad­ual progress in af­fec­tive neu­ro­science will make this prob­lem less press­ing- that given enough time&effort&re­sources, differ­ent ap­proaches to defin­ing suffer­ing will co­here, and this prob­lem will fade away.

I am here to in­form you that this is not go­ing to hap­pen: this out­side view that “af­fec­tive neu­ro­science is slowly set­tling on a con­sen­sus view of suffer­ing” is not hap­pen­ing, and this hur­dle to co­or­di­na­tion will not re­solve it­self. In­stead, the more af­fec­tive neu­ro­science has learned about valence, the more con­fus­ing and di­ver­gent the pic­ture be­comes.

The fol­low­ing is an overview (adapted from my core re­search) of what af­fec­tive neu­ro­science knows about valence.

I’ll front-load some im­pli­ca­tions for dis­cus­sion:

  • There’s lots of philo­soph­i­cal con­fu­sion in valence/​suffer­ing re­search. In Kuh­nian terms, this would sug­gest that af­fec­tive neu­ro­science is ripe for a paradigm shift. Paradigm shifts of­ten come from out­side the field, and usu­ally have un­pre­dictable out­comes (it’s difficult to pre­dict how some fu­ture ver­sion of af­fec­tive neu­ro­science may define suffer­ing).

  • Or­ga­ni­za­tions who’ve been us­ing mod­els from af­fec­tive neu­ro­science, such as FRI, ACE, and OpenPhil, should be clearer about the caveats in­volved, and should con­sider hedg­ing their bets with some ‘ba­sic re­search’ plays.

  • The longer we don’t have a good model for what suffer­ing is, the worse off we’ll be with re­gard to move­ment co­or­di­na­tion.

-------------------------Be­gin re­view-------------------------

Why some things feel bet­ter than oth­ers: the view from neuroscience

Valence re­search tends to seg­re­gate into two buck­ets: func­tion and anatomy. The former at­tempts to provide a de­scrip­tion of how valence in­ter­acts with thought and be­hav­ior, whereas the lat­ter at­tempts to map valence states to the anatomy of the brain. The fol­low­ing are key high­lights from each ‘bucket’:

Valence as a func­tional com­po­nent of thought & be­hav­ior:

One of the most com­mon views of valence is that it’s the way the brain en­codes value:

Emo­tional feel­ings (af­fects) are in­trin­sic val­ues that in­form an­i­mals how they are far­ing in the quest to sur­vive. The var­i­ous pos­i­tive af­fects in­di­cate that an­i­mals are re­turn­ing to “com­fort zones” that sup­port sur­vival, and nega­tive af­fects re­flect “dis­com­fort zones” that in­di­cate that an­i­mals are in situ­a­tions that may im­pair sur­vival. They are an­ces­tral tools for liv­ing—evolu­tion­ary mem­o­ries of such im­por­tance that they were coded into the genome in rough form (as pri­mary brain pro­cesses), which are re­fined by ba­sic learn­ing mechanisms (sec­ondary pro­cesses) as well as by higher-or­der cog­ni­tions/​thoughts (ter­tiary pro­cesses). (Panksepp 2010a).

Similarly, valence seems to be a mechanism the brain uses to de­ter­mine or la­bel salience, or phe­nom­ena worth pay­ing at­ten­tion to (Cooper and Knut­son 2008), and to drive re­in­force­ment learn­ing (Bischoff-Grethe et al. 2009).

A com­mon thread in these the­o­ries is that valence is en­tan­gled with, and per­haps caused by, an ap­praisal of a situ­a­tion. Frijda de­scribes this idea as the law of situ­ated mean­ing: ‘‘In­put some event with its par­tic­u­lar mean­ing; out comes an emo­tion of a par­tic­u­lar kind’’ (Frijda 1988). Similarly, Clore et al. phrase this in terms of “The In­for­ma­tion Prin­ci­ple”, where “[e]mo­tional feel­ings provide con­scious in­for­ma­tion from un­con­scious ap­praisals of situ­a­tions.” (Clore, Gasper, and Garvin 2001) Within this frame­work, pos­i­tive valence is gen­er­ally mod­eled as the re­sult of an out­come be­ing bet­ter than ex­pected (Schultz 2015), or a sur­pris­ing de­crease in ‘re­ward pre­dic­tion er­rors’ (RPEs) (Joffily and Cori­celli 2013).

Com­pu­ta­tional af­fec­tive neu­ro­science is a rel­a­tively new sub­dis­ci­pline which at­tempts to for­mal­ize this ap­praisal frame­work into a unified model of cog­ni­tive-emo­tional-be­hav­ioral dy­nam­ics. A good ex­am­ple is “Mood as Rep­re­sen­ta­tion of Mo­men­tum” (El­dar et al. 2016), where moods (and valence states) are un­der­stood as pre-pack­aged be­hav­ioral and epistemic bi­ases which can be ap­plied to differ­ent strate­gies de­pend­ing on what kind of ‘re­ward pre­dic­tion er­rors’ are oc­cur­ring. E.g., if things are go­ing sur­pris­ingly well, the brain tries to take ad­van­tage of this mo­men­tum by shift­ing into a hap­pier state that is more suited to ex­plo­ra­tion & ex­ploita­tion. On the other hand, if things are go­ing sur­pris­ingly poorly, the brain shifts into a “hun­ker-down” mode which con­serves re­sources and op­tions.

How­ever- while these func­tional de­scrip­tions are in­tu­itive, el­e­gant, and ap­pear to ex­plain quite a lot about valence, frus­trat­ingly, they fall apart as meta­phys­i­cally-satis­fy­ing an­swers when we look closely at edge-cases and the anatomy of pain and plea­sure.

Valence as a product of neu­ro­chem­istry & neu­roanatomy:

The available neu­roanatom­i­cal ev­i­dence sug­gests that the above func­tional themes merely high­light cor­re­la­tions rather than meta­phys­i­cal truths, and for ev­ery func­tional story about the role of valence, there ex­ist counter-ex­am­ples. E.g.:

Valence is not the same as value or salience:

(Ber­ridge and Kringelbach 2013) find that “rep­re­sen­ta­tion [of value] and cau­sa­tion [of plea­sure] may ac­tu­ally re­flect some­what sep­a­rable neu­ropsy­cholog­i­cal func­tions”. Re­lat­edly, (Jensen et al. 2007) note that salience is also han­dled by differ­ent, non-perfectly-over­lap­ping sys­tems in the brain.

Valence should not be thought of in terms of prefer­ences, or re­in­force­ment learn­ing:

Even more in­ter­est­ingly, (Ber­ridge, Robin­son, and Al­dridge 2009) find that what we call ‘re­ward’ has three dis­tinct el­e­ments in the brain: ‘want­ing’, ‘lik­ing’, and ‘learn­ing’, and the neu­ral sys­tems sup­port­ing each are each rel­a­tively dis­tinct from each other. ‘Want­ing’, a.k.a. seek­ing, seems strongly (though not wholly) de­pen­dent upon the mesolim­bic dopamine sys­tem, whereas ‘lik­ing’, or the ac­tual sub­jec­tive ex­pe­rience of plea­sure, seems to de­pend upon the opi­oid, en­do­cannabinoid, and GABA-ben­zo­di­azepine neu­ro­trans­mit­ter sys­tems, but only within the con­text of a hand­ful of so-called “he­do­nic hotspots” (el­se­where, their pres­ence seems to only in­crease ‘want­ing’). With the right in­ter­ven­tions dis­abling each sys­tem, it looks like brains can ex­hibit any per­mu­ta­tion of these three: ‘want­ing and learn­ing with­out lik­ing’, ‘want­ing and lik­ing with­out learn­ing’, and so on. Like­wise with pain, we can roughly sep­a­rate the sen­sory/​dis­crim­i­na­tive com­po­nent from the af­fec­tive/​mo­ti­va­tional com­po­nent, each of which can be mod­u­lated in­de­pen­dently (Shriver 2016).

Th­ese dis­tinc­tions be­tween com­po­nents are em­piri­cally sig­nifi­cant but not nec­es­sar­ily the­o­ret­i­cally crisp: (Ber­ridge and Kringelbach 2013) sug­gest that the dopamine-me­di­ated, nov­elty-ac­ti­vated seek­ing state of mind in­volves at least some small amount of in­trin­sic plea­sure.

A strong theme in the af­fec­tive neu­ro­science liter­a­ture is that plea­sure seems highly linked to cer­tain spe­cial­ized brain re­gions /​ types of cir­cuits:

We note the re­ward­ing prop­er­ties for all plea­sures are likely to be gen­er­ated by he­do­nic brain cir­cuits that are dis­tinct from the me­di­a­tion of other fea­tures of the same events (e.g., sen­sory, cog­ni­tive). Thus plea­sure is never merely a sen­sa­tion or a thought, but is in­stead an ad­di­tional he­do­nic gloss gen­er­ated by the brain via ded­i­cated sys­tems. … Analo­gous to scat­tered is­lands that form a sin­gle archipelago, he­do­nic hotspots are anatom­i­cally dis­tributed but in­ter­act to form a func­tional in­te­grated cir­cuit. The cir­cuit obeys con­trol rules that are largely hi­er­ar­chi­cal and or­ga­nized into brain lev­els. Top lev­els func­tion to­gether as a co­op­er­a­tive het­er­ar­chy, so that, for ex­am­ple, mul­ti­ple unan­i­mous ‘votes’ in fa­vor from si­mul­ta­neously-par­ti­ci­pat­ing hotspots in the nu­cleus ac­cum­bens and ven­tral pal­li­dum are re­quired for opi­oid stim­u­la­tion in ei­ther fore­brain site to en­hance ‘lik­ing’ above nor­mal. (Kringelbach and Ber­ridge 2009a)

Some of these ‘he­do­nic hotspots’ are also im­pli­cated in pain, and ac­tivity in nor­mally-he­do­nic re­gions have been shown to pro­duce an aver­sive effect un­der cer­tain psy­cholog­i­cal con­di­tions, e.g., when threat­ened or sa­ti­ated (Ber­ridge and Kringelbach 2013). Fur­ther­more, dam­age to cer­tain re­gions of the brain (e.g., the ven­tral pal­li­dum) in rats changes their re­ac­tion to­ward nor­mally-plea­surable things to ac­tive ‘dis­lik­ing’ (Cromwell and Ber­ridge 1993; Smith et al. 2009). More­over, cer­tain painkil­lers such as ac­etaminophen blunt both pain and plea­sure (Durso, Lut­trell, and Way 2015). By im­pli­ca­tion, the cir­cuits or ac­tivity pat­terns that cause pain and plea­sure may have similar­i­ties not shared with ‘he­do­nically neu­tral’ cir­cuits. How­ever, pain does seem to be a slightly more ‘dis­tributed’ phe­nomenon than plea­sure, with fewer re­gions that con­sis­tently con­tribute.

Im­por­tantly, the key take­away from the neuro-anatom­i­cal re­search into valence is this: at this time we don’t have a clue as to what prop­er­ties are nec­es­sary or suffi­cient to make a given brain re­gion a so-called “plea­sure cen­ter” or “pain cen­ter”. In­stead, we just know that some re­gions of the brain ap­pear to con­tribute much more to valence than oth­ers.

Fi­nally, the core cir­cuitry im­pli­cated in emo­tions in gen­eral, and valence in par­tic­u­lar, is highly evolu­tion­ar­ily con­served, and all ex­ist­ing brains seem to gen­er­ate valence in similar ways: “Cross-species af­fec­tive neu­ro­science stud­ies con­firm that pri­mary-pro­cess emo­tional feel­ings are or­ga­nized within prim­i­tive sub­cor­ti­cal re­gions of the brain that are anatom­i­cally, neu­ro­chem­i­cally, and func­tion­ally ho­molo­gous in all mam­mals that have been stud­ied.” (Panksepp 2010b) Others have in­di­cated the opi­oid-me­di­ated ‘lik­ing’ re­ac­tion may be con­served across an in­cred­ibly broad range of brains, from the very com­plex (hu­mans & other mam­mals) to the very sim­ple (c. el­e­gans, with 302 neu­rons), and all known data points in be­tween- e.g., ver­te­brates, mol­luscs, crus­taceans, and in­sects (D’iakonova 2001). On the other hand, the role of dopamine may be sub­stan­tially differ­ent, and even be­hav­iorally in­verted (as­so­ci­ated with nega­tive valence and aver­sion) in cer­tain in­ver­te­brates like in­sects (Van Swinderen and An­dretic 2011) and oc­topi.

A tax­on­omy of valence?

How many types of pain and plea­sure are there? While neu­ro­science doesn’t offer a crisp tax­on­omy, there are some ap­par­ent dis­tinc­tions we can draw from phys­iolog­i­cal & phe­nomenolog­i­cal data:

  • There ap­pear to be at least three gen­eral types of phys­i­cal pain, each as­so­ci­ated with a cer­tain pro­file of ion chan­nel ac­ti­va­tion: ther­mal (heat, cold, cap­saicin), chem­i­cal (lac­tic acid buildup), and me­chan­i­cal (punc­tures, abra­sions, etc) (Os­teen et al. 2016).

  • More spec­u­la­tively, based on a di­men­sional anal­y­sis of psy­choac­tive sub­stances, there ap­pear to be at least three gen­eral types of plea­sure: ‘fast’ (co­caine, am­phetamines), ‘slow’ (mor­phine), and ‘spiritual’ (LSD, Mescal­ine, DMT) (Gomez Emils­son 2015).

  • Mu­ta­tions in the gene SCN9A can re­move the abil­ity to feel any pain me­di­ated by phys­i­cal no­ci­cep­tion (Marković, Janković, and Ve­selinović 2015; Drenth and Wax­man 2007)- how­ever, it ap­pears that this doesn’t im­pact the abil­ity to feel emo­tional pain (Heck­ert 2012).

How­ever, these dis­tinc­tions be­tween differ­ent types of pain & plea­sure ap­pear sub­stan­tially ar­tifi­cial:

  • He­donic plea­sure, so­cial plea­sure, eu­daimonic well-be­ing, etc all seem to be man­i­fes­ta­tions of the same un­der­ly­ing pro­cess. (Kringelbach and Ber­ridge 2009b) note: “The available ev­i­dence sug­gests that brain mechanisms in­volved in fun­da­men­tal plea­sures (food and sex­ual plea­sures) over­lap with those for higher-or­der plea­sures (for ex­am­ple, mon­e­tary, artis­tic, mu­si­cal, al­tru­is­tic, and tran­scen­dent plea­sures).” This seems to ex­press a rough neu­ro­scien­tific con­sen­sus (Kash­dan, Robert, and King 2008), albeit with some caveats.

  • Like­wise in sup­port of lump­ing emo­tional & phys­i­cal valence to­gether, com­mon painkil­lers such as ac­etaminophen help with both phys­i­cal and so­cial pain (De­wall et al. 2010).

A deeper ex­plo­ra­tion of the tax­on­omy of valence is hin­dered by the fact that the phys­iolo­gies of pain and plea­sure are frus­trat­ing in­verses of each other.

  • The core hur­dle to un­der­stand­ing plea­sure (in con­trast to pain) is that there’s no plea­sure-spe­cific cir­cuitry analo­gous to no­ci­cep­tors, sen­sors on the periph­ery of the ner­vous sys­tem which re­li­ably cause plea­sure, and whose phys­iol­ogy we can iso­late and re­verse-en­g­ineer.

  • The core hur­dle to un­der­stand­ing pain (in con­trast to plea­sure) is that there’s only weak and con­flict­ing ev­i­dence for pain-spe­cific cir­cuitry analo­gous to he­do­nic hotspots, re­gions deep in the in­te­rior of the ner­vous sys­tem which seem to cen­trally co­or­di­nate all pain, and whose phys­iolog­i­cal me­chan­ics & dy­nam­ics we can iso­late and re­verse-en­g­ineer.

I.e., pain is easy to cause, but hard to lo­cal­ize in the brain; plea­sure has a more definite foot­print in the brain, but is much harder to gen­er­ate on de­mand.

Philo­soph­i­cal con­fu­sion in valence re­search:

In spite of the progress af­fec­tive neu­ro­science con­tinues to make, our cur­rent un­der­stand­ing of valence and con­scious­ness is ex­tremely limited, and I offer that the core hur­dle for af­fec­tive neu­ro­science is philo­soph­i­cal con­fu­sion, not mere lack of data. I.e., per­haps our en­tire ap­proach de­serves to be ques­tioned. Sev­eral cri­tiques stand out:

Neu­roimag­ing is a poor tool for gath­er­ing data:

Much of what we know about valence in the brain has been in­formed by func­tional imag­ing tech­niques such as fMRI and PET. But neu­ro­scien­tist Martha Farah notes that these tech­niques de­pend upon a very large set of as­sump­tions, and that there’s a wide­spread worry in neu­ro­science “that [func­tional brain] images are more re­searcher in­ven­tions than re­searcher ob­ser­va­tions.” (Farah 2014) Farah notes the fol­low­ing flaws:

  • Neu­roimag­ing is built around in­di­rect and im­perfect prox­ies. Blood flow (which fMRI tracks) and metabolic rates (which PET tracks) are cor­re­lated with neu­ral ac­tivity, but ex­actly how and to what ex­tent it’s cor­re­lated is un­clear, and skep­tics abound. Psy­chol­o­gist William Ut­tal sug­gests that “fMRI is as dis­tant as the gal­vanic skin re­sponse or pulse rate from cog­ni­tive pro­cesses.” (Ut­tal 2011)

  • The el­e­gant-look­ing graph­ics neu­roimag­ing pro­duces are not di­rect pic­tures of any­thing: rather, they in­volve ex­ten­sive statis­ti­cal guess­work and ‘clean­ing ac­tions’ by many lay­ers of al­gorithms. This hid­den in­fer­en­tial dis­tance can lead to un­war­ranted con­fi­dence, es­pe­cially when most mod­els can’t con­trol for differ­ences in brain anatomy.

  • Neu­roimag­ing tools bias us to­ward the wrong sorts of ex­pla­na­tions. As Ut­tal puts it, neu­roimag­ing en­courages hy­pothe­ses “at the wrong (macro­scopic) level of anal­y­sis rather than the (cor­rect) micro­scopic level. … we are do­ing what we can do when we can­not do what we should do.” (Ut­tal 2011)

Neu­ro­science’s meth­ods for an­a­lyz­ing data aren’t as good as peo­ple think:

There’s a pop­u­lar be­lief that if only the above data-gath­er­ing prob­lems could be solved, neu­ro­science would be on firm foot­ing. (Jonas and Kord­ing 2016) at­tempted to test whether the field is merely data-limited (yet has good meth­ods) in a novel way: by tak­ing a micro­pro­ces­sor (where the ground truth is well-known, and un­limited amounts of ar­bi­trary data can be gath­ered) and at­tempt­ing to re­verse-en­g­ineer it via stan­dard neu­ro­scien­tific tech­niques such as le­sion stud­ies, whole-pro­ces­sor record­ings, pair­wise & granger causal­ity, and di­men­sion­al­ity re­duc­tion. This should be an eas­ier task than re­verse-en­g­ineer­ing brain func­tion, yet when they performed this anal­y­sis, they found that “the ap­proaches re­veal in­ter­est­ing struc­ture in the data but do not mean­ingfully de­scribe the hi­er­ar­chy of in­for­ma­tion pro­cess­ing in the pro­ces­sor. This sug­gests that cur­rent ap­proaches in neu­ro­science may fall short of pro­duc­ing mean­ingful mod­els of the brain.” The au­thors con­clude that we don’t un­der­stand the brain as well as we think we do, and we’ll need bet­ter the­o­ries and meth­ods to get there, not just more data.

Sub­jec­tive ex­pe­rience is hard to study ob­jec­tively:

Un­for­tu­nately, even if we im­proved our meth­ods for un­der­stand­ing the brain’s com­pu­ta­tional hi­er­ar­chy, it will be difficult to trans­late this into im­proved knowl­edge of sub­jec­tive men­tal states & prop­er­ties of ex­pe­rience (such as valence).

In study­ing con­scious­ness we’ve had to rely on ei­ther crude be­hav­ioral prox­ies, or sub­jec­tive re­ports of what we’re ex­pe­rienc­ing. Th­ese ‘sub­jec­tive re­ports of qualia’ are very low-band­width, are of un­known re­li­a­bil­ity and likely vary in com­plex, hid­den ways across sub­jects, and as (Tsuchiya et al. 2015) notes, the method­olog­i­cal challenge of gath­er­ing them “has bi­ased much of the neu­ral cor­re­lates of con­scious­ness (NCC) re­search away from con­scious­ness and to­wards neu­ral cor­re­lates of per­cep­tual re­ports”. I.e., if we ask some­one to press a but­ton when they have a cer­tain sen­sa­tion, then mea­sure their brain ac­tivity, we’ll of­ten mea­sure the brain ac­tivity as­so­ci­ated with press­ing but­tons, rather than the ac­tivity as­so­ci­ated with the sen­sa­tion we’re in­ter­ested in. We can and do at­tempt to con­trol for this with the ad­di­tion of ‘no-re­port’ paradigms, but largely they’re based on the sorts of neu­roimag­ing paradigms cri­tiqued above.

Affec­tive neu­ro­science has con­fused goals:

Lisa Bar­rett (Bar­rett 2006) goes fur­ther and sug­gests that study­ing emo­tions is a par­tic­u­larly hard task for neu­ro­science, since most emo­tions are not “nat­u­ral kinds” i.e.. things whose ob­jec­tive ex­is­tence makes it pos­si­ble to dis­cover durable facts about. In­stead, Bar­rett notes, “the nat­u­ral-kind view of emo­tion may be the re­sult of an er­ror of ar­bi­trary ag­gre­ga­tion. That is, our per­cep­tual pro­cesses lead us to ag­gre­gate emo­tional pro­cess­ing into cat­e­gories that do not nec­es­sar­ily re­veal the causal struc­ture of the emo­tional pro­cess­ing.” As such, many of the terms we use to speak about emo­tions have only an ad-hoc, fuzzy pseudo-ex­is­tence, and this sig­nifi­cantly un­der­mines the abil­ity of af­fec­tive neu­ro­science to stan­dard­ize on defi­ni­tions, meth­ods, and goals.


In sum­mary, af­fec­tive neu­ro­science suffers from (1) a lack of tools that gather un­bi­ased and func­tion­ally-rele­vant data about the brain, (2) a lack of for­mal meth­ods which can re­con­struct what the brain’s do­ing and how it’s do­ing it, (3) episte­molog­i­cal prob­lems in­ter­fac­ing with the sub­jec­tive na­ture of con­scious­ness, and (4) an ill-defined goal, as it’s un­clear just what it’s at­tempt­ing to re­verse-en­g­ineer in the first place.

Fig 1 sum­ma­rizes some core im­pli­ca­tions of cur­rent neu­ro­science and philo­soph­i­cal re­search. In short: valence in the hu­man brain is a com­plex phe­nomenon which defies sim­ple de­scrip­tion. Affec­tive neu­ro­science has been hugely use­ful at illu­mi­nat­ing the shape of this com­plex­ity, but is run­ning into hugely diminish­ing re­turns with its cur­rent paradigm, and offers mul­ti­ple con­flict­ing mod­els of what valence & suffer­ing could be.

Figure 1, core take­aways of af­fec­tive neu­ro­science on valence


Bar­rett, Lisa Feld­man. 2006. “Are Emo­tions Nat­u­ral Kinds?” Per­spec­tives on Psy­cholog­i­cal Science: A Jour­nal of the As­so­ci­a­tion for Psy­cholog­i­cal Science 1 (1): 28–58.

Ber­ridge, Kent C., and Morten L. Kringelbach. 2013. “Neu­ro­science of Affect: Brain Mechanisms of Plea­sure and Dis­plea­sure.” Cur­rent Opinion in Neu­ro­biol­ogy 23 (3): 294–303.

Ber­ridge, Kent C., Terry E. Robin­son, and J. Wayne Al­dridge. 2009. “Dis­sect­ing Com­po­nents of Re­ward: ‘lik­ing’, ‘want­ing’, and Learn­ing.” Cur­rent Opinion in Phar­ma­col­ogy 9 (1): 65–73.

Bischoff-Grethe, Amanda, Eliot Hazel­tine, Lind­sey Ber­gren, Richard B. Ivry, and Scott T. Graf­ton. 2009. “The In­fluence of Feed­back Valence in As­so­ci­a­tive Learn­ing.” Neu­roI­mage 44 (1): 243–51.

Clore, Ger­ald L., Karen Gasper, and Ericka Garvin. 2001. “Affect as In­for­ma­tion.” In Hand­book of Affect and So­cial Cog­ni­tion, ed­ited by J. P. For­gas, 121–44. Mah­wah, NJ: Lawrence Er­lbaum As­so­ci­ates.

Cooper, Jeffrey C., and Brian Knut­son. 2008. “Valence and Salience Con­tribute to Nu­cleus Ac­cum­bens Ac­ti­va­tion.” Neu­roI­mage 39 (1): 538–47.

Cromwell, Howard C., and Kent C. Ber­ridge. 1993. “Where Does Da­m­age Lead to En­hanced Food Aver­sion: The Ven­tral Pal­li­dum/​sub­stan­tia In­nom­i­nata or Lat­eral Hy­potha­la­mus?” Brain Re­search 624 (1-2): 1–10.

De­wall, C. Nathan, Ge­off Mac­don­ald, Gre­gory D. Web­ster, Car­rie L. Mas­ten, Roy F. Baumeister, Caitlin Pow­ell, David Combs, et al. 2010. “Ac­etaminophen Re­duces So­cial Pain: Be­hav­ioral and Neu­ral Ev­i­dence.” Psy­cholog­i­cal Science 21 (7): 931–37.

D’iakonova, V. E. 2001. “The role of opi­oid pep­tides in the in­ver­te­brate be­hav­ior.” Zhur­nal evoliut­sion­noi biokhimii i fiz­iologii 37 (4): 253–61.

Drenth, Joost P. H., and Stephen G. Wax­man. 2007. “Mu­ta­tions in Sodium-Chan­nel Gene SCN9A Cause a Spec­trum of Hu­man Ge­netic Pain Di­sor­ders.” The Jour­nal of Clini­cal In­ves­ti­ga­tion 117 (12): 3603–9.

Durso, Ge­offrey R. O., An­drew Lut­trell, and Bald­win M. Way. 2015. “Over-the-Counter Relief From Pains and Plea­sures Alike: Ac­etaminophen Blunts Eval­u­a­tion Sen­si­tivity to Both Nega­tive and Pos­i­tive Stim­uli.” Psy­cholog­i­cal Science 26 (6): 750–58.

El­dar, Eran, Robb B. Rut­ledge, Ray­mond J. Dolan, and Yael Niv. 2016. “Mood as Rep­re­sen­ta­tion of Mo­men­tum.” Trends in Cog­ni­tive Sciences 20 (1): 15–24.

Farah, Martha J. 2014. “Brain Images, Ba­bies, and Bath­wa­ter: Cri­tiquing Cri­tiques of Func­tional Neu­roimag­ing.” The Hast­ings Cen­ter Re­port Spec No (March): S19–30.

Frijda, Nico H. 1988. “The Laws of Emo­tion.” The Amer­i­can Psy­chol­o­gist 43 (5): 349–58.

Gomez Emils­son, An­dres. 2015. “State-Space of Drug Effects: Re­sults.” Quali­a­com­put­ing.com. June 9. https://​​quali­a­com­put­ing.com/​​2015/​​06/​​09/​​state-space-of-drug-effects-re­sults/​​.

Heck­ert, Justin. 2012. “The Hazards of Grow­ing Up Pain­lessly.” New York Times, Novem­ber 18.

Jensen, Jimmy, An­drew J. Smith, Matthäus Willeit, Adrian P. Crawley, David J. Mikulis, Irina Vitcu, and Shitij Ka­pur. 2007. “Separate Brain Re­gions Code for Salience vs. Valence dur­ing Re­ward Pre­dic­tion in Hu­mans.” Hu­man Brain Map­ping 28 (4): 294–302.

Joffily, Ma­teus, and Gior­gio Cori­celli. 2013. “Emo­tional Valence and the Free-En­ergy Prin­ci­ple.” PLoS Com­pu­ta­tional Biol­ogy 9 (6): e1003094.

Jonas, Eric, and Kon­rad Kord­ing. 2016. “Could a Neu­ro­scien­tist Un­der­stand a Micro­pro­ces­sor?” doi:10.1101/​055624.

Kash­dan, Todd B., Biswas-Diener Robert, and Laura A. King. 2008. “Re­con­sid­er­ing Hap­piness: The Costs of Dist­in­guish­ing be­tween He­donics and Eu­daimo­nia.” The Jour­nal of Pos­i­tive Psy­chol­ogy 3 (4): 219–33.

Kringelbach, Morten L., and Kent C. Ber­ridge. 2009a. “Towards a Func­tional Neu­roanatomy of Plea­sure and Hap­piness.” Trends in Cog­ni­tive Sciences 13 (11): 479–87.

Marković, Dan­ica, Rad­milo Janković, and Ines Ve­selinović. 2015. “Mu­ta­tions in Sodium Chan­nel Gene SCN9A and the Pain Per­cep­tion Di­sor­ders.” Ad­vances in Anes­the­siol­ogy 2015: 1–6.

Os­teen, Jeremiah D., Volker Herzig, John Gilchrist, Joshua J. Em­rick, Chuchu Zhang, Xi­dao Wang, Joel Cas­tro, et al. 2016. “Selec­tive Spi­der Tox­ins Re­veal a Role for the Nav1.1 Chan­nel in Me­chan­i­cal Pain.” Na­ture 534 (7608): 494–99.

Panksepp, Jaak. 2010a. “Affec­tive Neu­ro­science of the Emo­tional BrainMind: Evolu­tion­ary Per­spec­tives and Im­pli­ca­tions for Un­der­stand­ing De­pres­sion.” Dialogues in Clini­cal Neu­ro­science 12 (4): 533–45.

———. 2010b. “Affec­tive Neu­ro­science of the Emo­tional BrainMind: Evolu­tion­ary Per­spec­tives and Im­pli­ca­tions for Un­der­stand­ing De­pres­sion.” Dialogues in Clini­cal Neu­ro­science 12 (4): 533–45.

Schultz, Wolfram. 2015. “Neu­ronal Re­ward and De­ci­sion Sig­nals: From The­o­ries to Data.” Phys­iolog­i­cal Re­views 95 (3): 853–951.

Shriver, Adam. 2016. “The Un­pleas­ant­ness of Pain For Hu­mans and Other An­i­mals.” https://​​www.academia.edu/​​12621257/​​The_Un­pleas­ant­ness_of_Pain_For_Hu­mans_and_Other_An­i­mals.

Smith, Kyle S., Amy J. Tin­dell, J. Wayne Al­dridge, and Kent C. Ber­ridge. 2009. “Ven­tral Pal­li­dum Roles in Re­ward and Mo­ti­va­tion.” Be­havi­oural Brain Re­search 196 (2): 155–67.

Tsuchiya, Naot­sugu, Me­lanie Wilke, Ste­fan Frässle, and Vic­tor A. F. Lamme. 2015. “No-Re­port Paradigms: Ex­tract­ing the True Neu­ral Cor­re­lates of Con­scious­ness.” Trends in Cog­ni­tive Sciences 19 (12): 757–70.

Ut­tal, William R. 2011. Mind and Brain: A Crit­i­cal Ap­praisal of Cog­ni­tive Neu­ro­science. MIT Press.

Van Swinderen, B., and R. An­dretic. 2011. “Dopamine in Drosophila: Set­ting Arousal Thresh­olds in a Mi­ni­a­ture Brain.” Pro­ceed­ings of the Royal So­ciety B: Biolog­i­cal Sciences 278 (1707): 906–13.

-------------------------End re­view-------------------------

The above re­view ac­tu­ally un­der­states the challenge of get­ting a good model of suffer­ing, be­cause it mostly avoids prob­lems re­lat­ing to con­scious­ness (of which there are many). Still, my in­tent here isn’t to be dis­cour­ag­ing—or even to throw cold wa­ter on the idea that some­day, EA could have a good, in­te­gra­tive defi­ni­tion of suffer­ing we could con­fi­dently use for an­i­mal welfare, AI safety, and so­cial in­ter­ven­tions al­ike. It should be clear from my work that I do think that’s pos­si­ble.

Rather, my point is that EA should be re­al­is­tic about how bad the cur­rent state of knowl­edge about suffer­ing is, and that this prob­lem isn’t go­ing to solve it­self.

Mike John­son, Qualia Re­search Institute