, Unit

Keywords:Third-party condemnationEvolution of moralityMoral impartialityMoral judgmentMoralistic punishment

es of mrallye neg

actions among actors, victims, and third-parties to help illuminate condemnation. We

moratain a

that they serve some evolved function. However, the adap-tive functions of moral condemnation remain mysterious.

Modern biological accounts of morality have made sev-eral key revisions of Darwins theory. First, while someresearchers continue to appeal to group selection (e.g.,Haidt, 2007), other theorists have shifted focus to otherevolutionary pathways to altruism (e.g., de Waal, 1996;Ridley, 1996; Wright, 1994), especially kin selection (Ham-ilton, 1964) and reciprocal altruism (Trivers, 1971). In oneimportant variety of altruism-based theories, the evolution

0010-0277/$ - see front matter 2009 Elsevier B.V. All rights reserved.

* Corresponding author. Present address: Economic Science Institute,Chapman University, Orange, CA 92866, United States. Tel.: +1 215 9138569; fax: +1 215 898 7301.

E-mail addresses: descioli@psych.upenn.edu, pdescioli@yahoo.com (P.DeScioli).

Cognition 112 (2009) 281299

Contents lists available at ScienceDirect

Cognit

journal homepage: www.elsedoi:10.1016/j.cognition.2009.05.008helping others and to avoid certain actions such as in-cest (e.g., Haidt, 2007). These theories, by themselves, donot explain why people think that others should be pun-ished for moral violations. They do not explain third-partymoral judgment, moralistic punishment, and moral impar-tiality, as symbolized by the scale, sword, and blindfold ofLady Justice. These patterns of human cognition are spe-cies-typical and complex (see Sections 68), suggesting

non-human animals. He argued that sympathy evolvedinto morality by group selection groups with moralityout-competed groups without morality. Darwin recog-nized that an important problem for his theory was thewide variety in moral rules, especially useless and harmfulrules, such as Hindu food taboos. Darwin concluded thatmorality was designed to benet the group and that detri-mental rules are errors.1. Introduction

Current evolutionary theories ofpeople are motivated to perform cerargue that basic differences between the adaptive problems faced by actors and third-par-ties indicate that actor conscience and third-party condemnation are likely performed bydifferent cognitive mechanisms. Further, we argue that current theories of conscience donot easily explain its experimentally demonstrated insensitivity to consequences. How-ever, these results might be explicable if conscience functions, in part, as a defense systemfor avoiding third-party punishment. If conscience serves defensive functions, then itscomputational structure should be closely tailored to the details of condemnation mecha-nisms. This possibility underscores the need for a better understanding of condemnation,which is important not only in itself but also for explaining the nature of conscience. Weoutline three evolutionary mysteries of condemnation that require further attention:third-party judgment, moralistic punishment, and moral impartiality.

2009 Elsevier B.V. All rights reserved.

lity address whyctions such as

Darwin (1871) initiated the evolutionary biology ofmorality in The Descent of Man, devoting two chapters tothe subject. Darwin claimed that the foundation-stoneof morality is sympathy, a social instinct also found inAccepted 15 May 2009 result, few theoretical tools are available for understanding the rapidly accumulating datasurrounding third-party judgment and punishment. Here we consider the strategic inter-Mysteries of morality

Peter DeScioli *, Robert KurzbanDepartment of Psychology, University of Pennsylvania, Philadelphia, PA 19104

a r t i c l e i n f o

Article history:Received 3 October 2008Revised 5 May 2009

a b s t r a c t

Evolutionary theorialtruism. More genemechanisms) to thed States

orality, beginning with Darwin, have focused on explanations for, these accounts have concentrated on conscience (self-regulatorylect of condemnation (mechanisms for punishing others). As a

ion

vier .com/locate /COGNIT

282 P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299of cooperation hinges on punishment, occurring whengroups with norm-enforcing punishers out-compete othergroups (e.g., Boyd & Richerson, 1992, 2005; Fehr, Fischb-acher, & Gchter, 2002; Gintis, 2000). Second, severalresearchers have departed from Darwin, arguing that thediversity of moral rules points to multiple, independentlyevolved psychological systems underlying morality (Haidt& Joseph, 2004, 2008; Hauser, 2006; Krebs & Janicki,2004; Miller, 2007; Stich, 2006; Wilson, 1993), includingsystems associated with suffering, hierarchy, reciprocity,honesty, self-control, violence, fairness, and incestavoidance.

We contribute an adaptationist analysis of morality,focusing on moral condemnation. Adaptationism (Wil-liams, 1966) leverages the tight link between structureand function in biological systems. This link allows infer-ences about structure from function (e.g., predicting themicrostructure of water strider legs from their function,Gao & Jiang, 2004), and also allows inferences about func-tion from structure (e.g., material properties of the Achillestendon show design for running, not walking; Bramble &Lieberman, 2004). In Section 2, we consider whether moralcognition appears sufciently distinct and coherent that itmakes sense to investigate what functions the system per-forms. In Sections 3 and 4, we present two main adapta-tionist arguments about the functional organization ofmoral cognition. First, inferring structure from function,we draw on the strategic dynamics among perpetrator, vic-tim, and third-party condemner to argue for two distinctcomponents of moral cognition: One subsystem regulatesones own behavior (conscience) while another mechanismis specialized for judging others (condemnation). Second,inferring function from structure, we consider empiricalevidence showing that insensitivity to welfare conse-quences nonconsequentialism is pervasive in moraljudgment. Briey, we will argue that nonconsequentialismchallenges the prevailing view that moral conscience is de-signed to promote the welfare of family, friends, or groups.

In Section 5, motivated by the preceding arguments, weoutline a specic functional relationship between con-science and condemnation components of moral cognition.In particular, we suggest that condemnation mechanismscausally precede conscience, and that conscience func-tions, at least in part, as a defense system designed to avoidattacks from third-party condemners. This view impliesthat moral conscience cannot be fully explained indepen-dent of condemnation.

If so, then much about the evolution of morality re-mains mysterious. There is little work and still less agree-ment on the function of moral condemnation systems (butsee Boyd & Richerson, 1992, 2005). Sections 68 describethree problems of moral condemnation third-party judg-ment, punishment, and impartiality.

2. The moral dimension

Philosophers and laypeople alike debate whether par-ticular actions are right or wrong. These inquiries concernthe positions of actions along the moral dimension fromgood to evil. For instance, Plato (4th century BC/2004) con-demned suicide as wrong while Hume (1783) positionedsuicide on the right side of the moral spectrum. Thesemoral debates take for granted the moral dimension itself(Macnamara, 1991), i.e., the cognitive capacity to assignmoral values to actions.

This paper examines morality, meaning phenomenasurrounding the concepts right and wrong. Specically,we focus on the moral dimension rather than the positionsof actions along moral lines. The evolutionary explanationsfor why particular actions are assigned particular moralvalues have received considerable attention (e.g., Alexan-der, 1987; Darwin, 1871; de Waal, 1996; Haidt & Joseph,2004, 2008; Hauser, 2006; Krebs & Janicki, 2004; Lieber-man, Tooby, & Cosmides, 2003, 2007; Ridley, 1996; Wilson,1993; Wright, 1994). However, why incest is perceived aswrong and reciprocity right is a different question fromwhy human minds possess a right-wrong spectrum uponwhich events like incest and reciprocity might fall.

The moral dimension refers broadly to the capacity toexperience the world as morally textured and differenti-ated. What functions might be served by the cognitive sys-tems that generate moral distinctions? Before addressingthis question, this section considers whether the cognitivesystems underlying the moral dimension are sufcientlydistinct and coherent that they might perform someevolved function.

2.1. Irreducible Moore-ality

Is morality a distinct phenomenon? If right and wrongare synonymous with or derived from other concepts (ben-et/harm, obedience/disobedience, etc.) then the explana-tion of morality will necessarily be tied to explanations ofthese phenomena. Some, for example, regard morality asequivalent to concern for welfare. The moral philosopherGert (2005) wrote, It should be apparent that by an evil,I mean a harm. . . by a good, I mean a benet (p. 91). Instark contrast, Moore (1903) argued that right and wrongare among the innumerable objects of thought whichare themselves incapable of denition, because they arethe ultimate terms of reference to which whatever is capa-ble of denition must be dened (p. 10, emphasis origi-nal). Similarly, some cognitive psychologists regard moralrightness and wrongness as conceptual primitives (Macna-mara, 1991; Mikhail, 2007).

The idea that moral concepts are derived from welfareconcepts is inconsistent with important moral phenomena,including many sexual prohibitions (e.g., Stengers & VanNeck, 2001) and food taboos (Douglas, 1966; Fessler &Navarrete, 2003; Simoons, 1994) that are not explainedby welfare gains. The ethnographic record is replete withmoral rules that are understood not in terms of welfareor harm but in relation to purity, authority, divinity, loy-alty, etc. (Haidt, 2007; Haidt & Joseph, 2008; Shweder,Much, Mahapatra, & Park, 1997). Even when people ex-plain their moral judgments in terms of welfare, these jus-tications are not always accurate. Laboratory evidenceshows that, at least in some cases, stated welfare concernsdo not drive moral judgments but instead reect post hocrationalization (Haidt, 2001). In studies of harmless viola-tions (e.g., eating ones dog following a natural death),judgments of harmfulness did not predict moral judgments

(Haidt & Hersh, 2001; Haidt, Koller, & Dias, 1993). Morestarkly, people maintained that violations like incest werewrong even in situations when no harm could result,appearing morally dumbfounded, unable to explain theirown judgments (Haidt, 2001). Similarly, Tetlock (2000)presented subjects with policies such as markets for body

P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299 283organs; even after participants welfare objections weresatised with additional provisions, roughly half continuedto morally condemn the policies. These ndings cautionthat welfare justications cannot be simply taken at facevalue: People generatewelfare justications evenwhen theirwelfare judgments do not explain their moral judgments.

Further, actions that yield clear welfare gains aresometimes rejected for moral reasons. In the footbridgedilemma (see below), 90% of people thought it was imper-missible to kill one person to save ve others (Hauser,2006). Similarly, participants rejected welfare gainsproduced by trade-offs involving protected values, suchas making money by destroying natural forests (Baron &Spranca, 1997; Tetlock, 2003). These observed dissocia-tions among moral and welfare judgments show that theconcept morally wrong cannot be simply equivalent toharmful.1

Right and wrong also do not seem to be derived fromauthority concepts like permissible/forbidden, or lawful/unlawful (Macnamara, 1991). Key evidence comes fromdevelopmental psychology. Contradicting early claims byPiaget (1932) and Kohlberg (1981), children are skepticalof authority in the moral realm. Children as young as threeregard moral rules (but not social conventions) as indepen-dent of authority, custom, and explicit rules (Smetana &Braeges, 1990; Smetana, Schlagman, & Adams, 1993; Turi-el, 1998). Children often disobey authority when com-mands will result in harm (Laupa & Turiel, 1986). Theyalso reject the legitimacy of rules applied to personal is-sues such as hairstyle (Nucci, 1981). Further, children donot view violators of unjust laws as blameworthy (Helwig& Jasiobedzka, 2001). Even young children appear toappreciate the motto of the University of Pennsylvania,Leges sine moribus vanae: Laws without morals are useless.

In short, moral concepts are not reducible to welfareconcepts or authority concepts. People can conceive ofwelfare-increasing wrongs, welfare-decreasing rights,wrongful obedience, and rightful disobedience (unlikeunimaginable entities such as a four-sided triangle). Theconcepts of right and wrong, being irreducible, requireexplanation. Understanding these concepts, and the under-lying computational systems, is a central aim of moralpsychology.

2.2. Coherence across different types of violations

Cross-culturally, moral prohibitions concern diversephenomena, including violence, sex, food, communication,

1 The observed distinction between moral and welfare concepts does not,by itself, imply that moral cognition is insensitive to (conscious orunconscious) information about harm (e.g., the deleterious effects of inceston reproduction), but only that the concept of harm turns on differentinformation from moral concepts. Insensitivity to welfare in moral cogni-tion is addressed in Section 4.and witchcraft (e.g., Barton, 1919; Shweder, Mahapatra, &Miller, 1987; Shweder et al., 1997). In turn, different moralrules surround different biological problems (e.g., domi-nance hierarchies, mate guarding, foraging), which inmany species are each handled by specialized cognitivesystems (Krebs & Davies, 1993). This might suggest thatmorality is an umbrella term for a collection of differentpsychological systems (e.g., Haidt & Joseph, 2004, 2008;Hauser, 2006; Krebs & Janicki, 2004; Miller, 2007; Stich,2006; Wilson, 1993). Morality might be an articialgrouping rather than a natural kind, potentially explainingwhy moral concepts seem irreducible.

However, there is coherence inmoral cognition that cutsacross content domains. Moral condemnation, in particular,shows the same fundamental properties across diverse of-fenses viz., third-party judgment, moralistic punishment,and (the pretense of) moral impartiality (reviewed in Sec-tions 68). Also, the causal and intentional structure ofmoral actions seems to be evaluated similarly across of-fense types, perhaps indicating a common representationalformat (Mikhail, 2007). Further, moral cognition differsfrom other types of normative judgments (e.g., convention,authority, precaution) in patterns of reasoning (Fiddick,2004; Turiel, 1998) and, more visibly, in emotional andbehavioral consequences. Last, research on the emotionselicited bymoral violations (e.g., guilt, shame, righteous an-ger) has found that it is not the events, per se, that deter-mine which emotion is experienced but rather how eventsare appraised (i.e., appraisals about causality and actorsdispositions; Tracy & Robins, 2006, p. 1339; see also Tang-ney, Stuewig, & Mashek, 2007). The emotion of guilt, for in-stance, can be elicited by assault, theft, lying, indelity, etc.(Keltner & Buswell, 1996).

Coherence seems apparent when the moral value of anaction changes, and the many accoutrements of moraljudgment are activated or deactivated in a coordinatedfashion. Moralization, when objects or activities thatwere previously morally neutral acquire a moral compo-nent (Rozin, Markwith, & Stoess, 1997, p. 67), and the re-verse, amoralization, show that actions undergo discreteshifts in moral status. For instance, a precautionary rulemight shift to a moral rule, or vice versa (e.g., smoking, Ro-zin & Singh, 1999). Importantly, newly minted rules recruitthe full suite of moral machinery, including prohibition,outrage, censure, overjustication, internalization, and en-hanced parent-to-child transmission (Rozin, 1999). It is asthough moral cognition has an insert here parameter,processing diverse moral rules with the same computa-tional architecture (cf. Pinker, 1999).

Features of moral cognition that cohere across contentdomains suggest limitations to theories that tie theevolved function of morality to particular content domains.Previous work has focused on some particular subset ofmoral phenomena, dismissing moral variety as error (e.g.,Darwin, 1871, pp. 9599). Most often theorists concentrateon moral rules about altruism (Alexander, 1987; Darwin,1871; de Waal, 1996; Ridley, 1996; Wright, 1994), or a col-lection of domains, such as incest, trade, honesty, and adul-tery (e.g., Haidt & Joseph, 2004, 2008; Hauser, 2006; Krebs& Janicki, 2004; Lieberman et al., 2003, 2007; Miller, 2007;Wilson, 1993). If, however, there are mechanisms that

Although the scenario could be about human crime-ght-

In line with adaptationism (Williams, 1966), the reliablelink between strategic function and mechanism structureimplies that information about adaptive problems can beused to make inferences about mechanism structure. Illus-trating this reliable link, the three-player gameabove causedfunctionally equivalent devices to evolve (independently) intobacco, cotton, strawberry, and cucumber, as well as inseveral species of wasps and mites (Takabayashi & Dicke,1996). In this section, we draw on the strategic dynamicamong perpetrator, victim, and third-party condemner tomake inferences about the structure of moral cognition.

3.1. Morality as (at least) a three-player game

Moral interactions frequently involve three roles orplayers: We will refer to these individuals as actors, sec-ond parties, and third parties (see Fig. 1). In this frame-work, actors affect second parties, and third parties judgeactors and sometimes intervene. Third parties are recur-rent elements of moral interactions. For instance, US crimestatistics (19931999) show that in cases of assault acommon moral violation third parties were usually pres-ent (72% of cases) and roughly half became involved in the

284 P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299ing, it actually describes the strategic dynamic amongmaize, worm, and wasp. When attacked by moth larvae,maize emits a specialized chemical signal to alert parasiticwasps (Takabayashi & Dicke, 1996). Wasps attack theworms by injecting eggs that soon hatch into wasp larvae,which eat worms from the inside out.

This example illustrates that strategic games can shapeadaptations (Maynard Smith, 1982). Specically, the sce-nario involves a three-player game with asymmetric posi-tions, each requiring different adaptations (see also Sinervo& Lively, 1996). Different strategic problems lead us to ex-pect maize to have signaling mechanisms and wasps tohave signal reception systems, rather than vice versa. In-deed, this is correct. Maize even emits different chemicalcompositions for different worm species, each signal draw-ing different specialist wasps (De Moraes, Lewis, Pare, Al-born, & Tumlinson, 1998).operate across domains, then they might have evolvedfunctions that cut across domains. These overarching func-tions would be explicable not in terms of violation-specicadaptive problems (e.g., inbreeding avoidance), but interms of adaptive problems shared in common across thevariety of moral judgments.

In sum, the moral dimension is what remains constantacross moral judgments about altruism, trade, ghting,sex, food taboos, drug use, blackmagic, etc. Across domains,moral cognition features a rich psychology surroundingmoral judgment of oneself and others; condemnation, forinstance, includes monitoring, gossip, judgment, punish-ment, and impartiality (see Sections 68). Focusing on themoral dimension itself, rather than the objects on whichit operates, highlights evolutionary questions about the dis-tinct and coherent features of moral cognition.

3. The problem of morality

A better understanding of the adaptive problems sur-rounding moral interactions can help guide research to un-cover the cognitive mechanisms that evolved to handlemoral situations. Moral interactions involve multiple indi-viduals who each make decisions that affect their ownand others outcomes. This type of adaptive problem canbe described in terms of game theory, the study of games,or strategic environments in which individuals (players)can affect each others outcomes. At its core, game theoreticanalysis involves taking the perspective of each player in aninteraction and assessing how players can pursue theirgoals, given information about how the other players arelikely to behave. Game theory has proved extremely usefulfor understanding the evolution of mechanisms for strate-gic behavior in humans and non-human species (Krebs &Davies, 1993; Maynard Smith, 1982). What strategic prob-lems, or games, shaped the cognitive systems that hu-mans use to handle moral situations? Further, given thebroad outline of moral interactions, what can be inferredabout the structure and composition of moral cognition?

Consider this scenario: A tall beauty is attacked by a hun-gry horde of grubby onlookers, and her screams attract a sky-soaring savior who lays waste to the villainous worms.situation (Planty, 2002).The presence of third parties in moral interactions sets

up a strategic game among actors, second parties, and thirdparties. Each role can be regarded as a different position ina moral game. These positions are dened by their respec-tive problems, in the same way that tennis servers andreceivers confront different problems (Walker & Wooders,2001). Further, individuals might change positions (as intennis), taking, in turn, the role of actor, second party, orthird party.

We begin from the perspective of third parties, i.e.,bystanders who observe and judge an actor whose behav-ior affects a second party. As previously mentioned, thebroad adaptive functions served by third-party interven-

Fig. 1. Diagram of a strategic interaction among actor, second party, andthird party. One important case is among perpetrator, victim, andcondemner. Arrows depict the activity of one individual toward anotheras described in the parentheses. We focus on cognitive processes formanaging the actor position (conscience) as distinct frommechanisms formanaging the third party position (condemnation). For example, con-science might apply moral rules to questions such as Should I takemoney from a victim? or Should I lie to a victim? whereas condem-nation would apply moral rules to ask Should I condemn an actor whotakes money from a victim? or Should I condemn an actor who lies to avictim?.

tion are unclear. Here we describe proximate problemsthat third-parties confront and resolve, leaving aside forthe moment the ultimate goals that intervention supports.

Third parties condemn actors moral wrongs and theyalso praise actors moral virtues. Here we focus on third-party condemnation rather than praise.2 We propose thatone of the central moral games is the interaction amongthird-party condemner, perpetrator (actor), and victim (sec-ond party).

In this game, third-party condemners confront theproblem of detecting, judging, and punishing perpetratorsviolations, while recruiting others support and avoidingcounter-accusations. Condemners judgments are primar-ily retrospective, occurring after the fact, and they usuallyinvolve uncertainty about actors motives and behavior.Moreover, perpetrators typically respond to condemnationwith denial, counter-accusations, and retaliation. In decid-ing whether to condemn, third parties must consider the

P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299 285probability of guilt, whether an accusation will be believedby others, and whether their own innocence is demonstra-ble (to defend against counter-accusations). If third partieschoose to condemn a perpetrator, they must further decidethe severity of the offense, how much punishment is de-served, and whether or not to perform punishment them-selves. If condemners decide to punish, they face theproblem of minimizing the costs associated with perpetra-tor retaliation (see Section 7). Third parties might reducepunishment costs by recruiting others help or by deliver-ing sanctions opportunistically, e.g., when the violatorskin are absent (see Knauft, 1987).

Actors, or potential perpetrators, face different prob-lems. Actors must compute the costs and benets of theirpotential actions and choose the actions that best achievetheir goals. Some actions might involve moral violationsagainst a second party, which could draw revenge fromvictims or sanctions from third parties. Whereas third par-ties seek to detect violations, perpetrators confront theopposite problem: avoiding detection by concealing viola-tions. Whereas third parties make moral judgments retro-spectively, actors make moral judgments prospectively,assessing the wrongness of a behavior before selecting acourse of action.3 Whereas third parties have incomplete

2 Some broad stylized facts point to a special role for condemnation.Condemnation far exceeds praise in moral discourse among children (Ross& den Bak-Lammers, 1998) and adults (Wiessner, 2005). Mythic moralheroes, like superheroes, deliver punches to villains rather than rewards tothe virtuous (DeScioli & Kurzban, 2008). Human legal systems sentencecriminals to prison, rather than sentencing upstanding citizens to DisneyWorld vacations. Looking at moral concepts, one common usage of rightrefers to not condemnable rather than praiseworthy violent assaultcan be regarded as right if provoked, and abandoning a spouse can beviewed as right if precipitated by indelity. However, the reverse is nottrue for wrong which means condemnable and not merely notpraiseworthy. Thus, moral concepts might especially hinge on condem-nation. Finally, people praise altruism, but it is unclear how much praise isgiven to moral behavior, which is different from altruism (see Section 3).

3 Third parties also make prospective moral judgments, e.g., to threatenpotential violators or to negotiate candidate moral rules. However, thesejudgments are an elaboration on a basically retrospective design: prospec-tive condemnation is meaningful only in terms of potential retrospectivejudgment and punishment after the violation occurs. Analogously, pro-spective conscience mechanisms are elaborated by retrospective guiltwhich is meaningful in connection with subsequent prospective judgments.information about actors intentions, perpetrators are cer-tain of their own intentions (although they might try to hideintentions from condemners). Whereas third parties com-pute how much punishment is deserved, actors computethe punishments they expect to incur, and they attempt toavoid sanctions, sometimes by inhibiting wrongful behavior.

Finally, consider the perspective of the second party,i.e., a potential victim who is affected by the behavior ofa perpetrator. Victims must avoid costs imposed by perpe-trators, while recruiting aid from condemners. Victims cantry to avoid being wronged by evading perpetrators, bymounting a defensive stand, or by engaging in revengeafter the infraction to deter future violations. These tasksare not specic to moral violations (e.g., revenge can beused to deter behavior that is not morally wrong). How-ever, in order to recruit help from third parties, victimsare aided by moral judgments and accusations againsttheir foes. Note that this suggests that second party moraljudgment is derivative: Its force depends on the willing-ness of third parties to condemn moral violators.

The strategic dynamic among perpetrator, victim, andcondemner might have shaped human adaptations. Recur-rent adaptive problems can shape content-rich cognitivesystems over evolutionary time (Tooby & Cosmides,1992). The moral game might have favored cognitive sys-tems that include representations of each position (perpe-trator, victim, and condemner) as well as computationsperformed over these representations. Crucially, individu-als in each position confront different problems, implyingthat they will require different computations to solvethem.

Understanding moral cognition will be aided by a care-ful dissection of the relevant strategic problems. Broadly,perpetrators confront the problem of avoiding detectionand punishment by condemners, and revenge by victims.Victims must avoid costs imposed by perpetrators, whilerecruiting aid from condemners. Condemners confrontthe problem of detecting and punishing perpetrators vio-lations, while recruiting others support and avoidingcounter-accusations. Because these very different prob-lems require different computational procedures, they arelikely handled by different specialized cognitive mecha-nisms (Marr, 1982; Sherry & Schacter, 1987; Tooby & Cos-mides, 1992).

3.2. Conscience and condemnation as different components ofmoral cognition

Inferring structure from function, we suggest that thethree-player moral game points to two components ofmoral cognition. The rst is a system that uses moral con-cepts to guide ones actions; we will refer to this set ofmechanisms as conscience. The second is a system that usesmoral concepts to judge and punish a perpetrator; we willrefer to this system as condemnation. While conscienceoperates in actors, condemnation operates in third-partycondemners, as they evaluate actors.

Note that our usage of the terms conscience and con-demnation stems from the three-player game among per-petrator, victim, and condemner. We classify asconscience the mechanisms in actors that select among

potential actions based on moral values computed for eachaction. We classify as condemnation the mechanisms inthird parties that deploy accusations and punishmentbased on moral values computed for actors past actions.This distinction ts reasonably well with common usage

4

selection (Miller, 2007). In The Moral Animal, Wright

286 P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299of these terms.The inference that conscience and condemnation are

distinct is supported by basic differences in the adaptiveproblems surrounding them. Actor conscience is primarilyprospective, includes complete information about inten-tions, and competes with other motivations. Third-partycondemnation is retrospective, informed by only incom-plete information about intentions, and can be more de-tached from other motives (but see below aboutpartiality, Section 8). Different functional demands can beused to infer different underlying cognitive mechanisms(Marr, 1982; Sherry & Schacter, 1987; Tooby & Cosmides,1992). We can expect conscience and condemnation tobe underlain by different specialized computational pro-grams, just as we expect different mechanisms for wormsignaling and wasp receiving.

The conscience/condemnation distinction is supportedby work on moral emotions (Tangney, Stuewig, & Mashek,2007). This literature distinguishes the self-conscious emo-tions of shame, embarrassment, and guilt (Tangney et al.,2007), from the other-critical emotions of contempt, anger,and disgust (Rozin, Lowery, Imada, & Haidt, 1999). Thesesystems might differ in their targets (self vs. other) becausethey are designed for deployment in different strategicpositions (actor vs. third-party).

Understanding conscience and condemnation requiresa characterization of the functional relationship betweenthem. Consider an analogy with human language. Lan-guage production and comprehension are served by func-tionally distinct systems, both neuroanatomically and atthe level of gross morphology (ears as opposed to vocalcords and tongue, etc.). Further, language production andcomprehension have a specic functional relationship. Ingeneral, theories of the evolution of communication regardreception/comprehension systems as driving the evolutionof signaling/production systems (Dawkins & Krebs, 1978;Maynard Smith, & Harper, 2003). That is, the propertiesof signal reception systems cause the properties of signalproduction systems, rather than the opposite, because asignal cannot affect the receivers behavior unless the re-ceiver has a system capable of interpreting the signal.Analogously, condemnation mechanisms might havecaused the details of conscience, or vice versa.

The possibility that conscience and condemnation aredifferent component mechanisms draws attention to ques-tions about their functional relationship. The issue is as ba-sic as whether the nose is designed to hold spectacles, orspectacles are designed to t noses (see Voltaires Candide,1759). To use an evolutionary example, warblers and cuck-oos have matching colored eggs, leading to questions about

4 In one example from ction, Jiminy Cricket (in Disneys Pinocchio) wasassigned to be Pinocchios conscience, Lord High Keeper of the knowledgeof right and wrong. . .and guide along the straight and narrow path. Actingas conscience, Jiminy guided Pinocchios own actions (versus urging him tocondemn or accuse others for wrongdoing).(1994) says of morality, Darwins sometimes diffuse spec-ulations about the social instincts have given way to the-ories rmly grounded in logic and fact, the theories ofreciprocal altruism and kin selection (p. 328). In The Ori-gins of Virtue, Ridley (1996) focuses on reciprocity, discuss-ing the Prisoners Dilemma, public goods, and gains intrade, closing with the line: We must encourage socialand material exchange between equals for that is the rawmaterial of trust, and trust is the foundation of virtue (p.265). In Good Natured: The Origins of Right and Wrong in Hu-mans and Other Animals, de Waal (1996) reviews primatewhether the color of cuckoo eggs is shaped by the color ofwarbler eggs or vice versa. Research favors a warbler-cen-tered theory holding that warbler eggs are mimicked bycuckoo eggs, so that warblers will be tricked into caringfor cuckoo offspring (Brooke & Davies, 1988). Below, weexamine the functional relationship between actor con-science and third-party condemnation. Is conscience thecore of morality, with condemnation emerging as abyproduct? Or, does condemnation form the moral core,with conscience evolving as a downstream consequence?

4. Is morality conscience-centered?

Historically, evolutionary theorists have proposed con-science-centered explanations for moral cognition. Thatis, theorists have taken the problem of morality to beexplaining the cognitive mechanisms that operate in actorsto motivate their own behavior, answering questions suchas, why do people avoid incest? Much less attention hasbeen given to the mechanisms that cause third-party con-demnation, answering questions such as, why do peoplecondemn others for incest? Often, this focus is implicit:Theorists do not use the word conscience, but simply re-fer to mechanisms for self-regulation as morality, leavingcondemnation out of view. There are several importantexceptions that distinguish self-regulation from third-party condemnation. These accounts generally regard con-demnation as a byproduct or spillover from consciencemechanisms or other psychological mechanisms, such asdisgust (e.g., Fessler & Navarrete, 2004; Greene, 2008;Haidt & Joseph, 2004; Lieberman et al., 2003). In sum, the-ories of morality have generally been conscience-cen-tered in that they either focus exclusively on systems formoral self-regulation, assuming that conscience can be ex-plained independent of condemnation, or they explicitlyargue that third-party condemnation is a downstream con-sequence of conscience mechanisms.

The historical focus on conscience mechanisms explainswhy theories of morality have tended to invoke altruismmodels. Altruismmechanisms motivate an actors behaviorto benet others. Darwin (1871) invoked a group selectionmodel to explain morality, arguing that conscience is de-signed to benet the group. Modern theories continue toregard morality as an altruism device, although the evolu-tionary processes thought to be responsible have expandedbeyond group selection, and now include kin selection(e.g., Wright, 1994), reciprocity (e.g., Ridley, 1996), culturalgroup selection (e.g., Boyd & Richerson, 2005), and sexual

The issue is more general than altruism. For example,while Freud (1918) attributed incest avoidance to fear ofmoralistic punishment, recent evidence indicates thatchildhood co-residence causes sexual aversion (Fessler &Navarrete, 2004; Lieberman et al., 2003, 2007). Compliancewith incest prohibitions might be primarily caused by spe-cic incest aversion mechanisms rather than conscience.Similarly, Kant argued that suicide was morally wrongbut most people preserve their lives, to be sure, in accor-dance with duty, but not from duty (Kant, 1785/1993, p.10).

Isolating the specic effects of conscience is important

P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299 287altruism, including chapters entitled Sympathy, Quidpro Quo, and Getting Along. De Waal opens the book:In addition to being human, we pride ourselves on beinghumane. What a brilliant way of establishing morality asthe hallmark of human nature by adopting our speciesname for charitable tendencies! (p. 1). Following Darwin,these accounts use theories of altruism to explain morality.More generally, they take conscience (mechanisms operat-ing in actors) to be the key explanandum of morality.

This section presents an alternative view. We look atthe structure of moral conscience to make inferences aboutits function. A central feature of moral cognition perva-sive nonconsequentialism poses difculties for altruismtheories of conscience, and more generally, challengesthe assumption that conscience can be explained indepen-dent of condemnation.

4.1. Compliance with moral rules can be caused by non-moral systems

Conscience, on the present view, is a cognitive systemthat uses the moral concepts right and wrong to guideactor behavior. The effects of conscience are difcult to iso-late because non-moral cognitive systems often causemorally compliant behavior (Kant, 1785/1993). For exam-ple, many birds are faithful mating partners but this behav-ior is (presumably) not caused by cognitive systems thatuse the moral concepts right and wrong. Mate delityin humans might also be caused, in part, by non-moral sys-tems, in addition to moral motivations. The challenge is topick out conscience against a background of morally com-pliant behavior caused by other systems.

Altruism is an important example of how morally com-pliant behavior is often caused by non-moral systems.Altruism can evolve by multiple pathways, including kinselection, reciprocity, and mutualism. Helping is wide-spread in organisms from bacteria (Grifn, West, & Buck-ling, 2004) to insects (Holldobler & Wilson, 1990) to non-human primates (de Waal, 1996; Schino, 2007). Likewise,many human altruism devices, such as parental care mech-anisms, evolved prior to moral cognition and continue tooperate independent of moral motives (e.g., umbilicalcords, mammary glands). The development of altruism isrevealing. By age one, children comfort distressed individ-uals with hugs and kisses, and by age two, many prosocialbehaviors emerge (Eisenberg & Fabes, 1998; Zahn-Waxler,Radke-Yarrow, & Wagner, 1992). For instance, Warnekenand Tomasello (2006) found that 18-month-olds helpedexperimenters reach an object dropped accidentally (butnot intentionally). However, moral cognition seems to be-gin to develop around age 3 and continues to changethroughout childhood (Darley & Shultz, 1990). During thistime, children increasingly show patterns typical of adultmoral judgment, such as distinguishing intentional andaccidental violations or foreseeable and unforeseeable vio-lations (Darley & Shultz, 1990). Importantly, these changesoccur after a large repertoire of prosocial behavior is al-ready in place. Taken together, phylogenetic and ontoge-netic evidence indicates that altruistic behavior can occurindependent of moral conscience (see also Batson, Klein,Highberger, & Shaw, 1995; Kant, 1785/1993).for understanding its function. Evolutionary theories ofmoral conscience generally take for granted that compli-ance with moral rules can be attributed to the operationof conscience. However, it is unclear to what extent con-science causes moral compliance over and above compli-ance produced by non-moral systems designed foraltruism, delity, honesty, disease-avoidance, etc. Indeed,experiments that control for alternative motives (e.g., rep-utation) show surprisingly little evidence that moral judg-ment motivates moral behavior. Instead, moral judgmentoften motivates efforts to appear moral (Batson, 2008).

4.2. Nonconsequentialism in moral cognition

Moral consequentialism is the philosophical view thatwhether an act is morally right depends only on conse-quences (Sinnott-Armstrong, 2006). Usually, consequen-tialist theories focus on how actions directly affectwelfare outcomes (e.g., utilitarianism), and we will use thissense of the term. That is, consequentialists judge an actionsolely based on the welfare consequences that the action isexpected to produce.5 In contrast, moral nonconsequential-ism is the view that moral rightness does not depend onlyon consequences. Commonly, theories in this genre focuson absolute rules of behavior, such as Kants categoricalimperatives. Consequentialism regards actions as com-pletely instrumental, to be judged solely according to theirexpected effects. In contrast, nonconsequentialism allowsjudgment on the basis of the properties of an action (e.g.,whether it involves lying) rather than only on the basis ofits effects.

The debate between these views extends beyond phi-losophy. Legal professionals debate whether the rule ofreason (consequentialism) or per se rules (nonconse-quentialism) should be used to decide cases (Arthur,2000). In the policy arena, these approaches often compete

5 Philosophers have proposed many forms of consequentialism (Sinnott-Armstrong, 2006). To avoid any potential confusion, we will specify furtherthe type of consequentialism we have in mind: The relevant consequencesare welfare consequences that are expected (versus actual) and direct(versus indirect). By welfare, we remain general by including all possibleweightings among self and others welfare as potential grounds forconsequentialist judgment. The focus on expectation further implies thatgreater weight is placed on immediate proximate effects, because moredistant effects should be discounted in proportion to uncertainty. The focuson direct effects means that the relevant consequences are the expectedwelfare consequences attached to the action under evaluation; the relevanteffects do not include consequences resulting from the moral judgmentitself, such as reputation effects for the judger.

288 P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299to shape decision-making (Baron, 1998), as signaled by thetitle of an important book on the subject, Fairness VersusWelfare (Kaplow & Shavell, 2002). For example, a debateabout HIV policy concerns whether programs should focuson behavior per se (abstinence) by denying services to ac-tive sex workers and drug users, or should serve these pop-ulations to reduce HIV transmission and the resulting harm(Marlatt, 1996; Rekart, 2005).

Experiments have found that peoples moral judgmentsare sometimes consequentialist and sometimes nonconse-quentialist (reviewed in Hauser, 2006). For example, in oneversion of the trolley problem (the switch dilemma), peo-ple generally made consequentialist decisions: most par-ticipants (90%) thought it was permissible to ip a switchto redirect a trolley, thereby killing one person to save ve.However, in the footbridge dilemma, people generallymade nonconsequentialist decisions: most people (90%)said it was impermissible to save ve people by pushingone person off of a footbridge in front of the trolley. Theseresults show that most people are nonconsequentialists(but not strict deontologists).

Nonconsequentialism is widespread and complex. Oneline of research found that people maintained that certainbehaviors were wrong and should be punished even whenno harmful consequences would result (e.g., non-reproduc-tive consensual incest, Haidt, 2001; Haidt & Hersh, 2001;Haidt et al., 1993; Tetlock, 2000). Others have shown thatthe intentional and causal structure of behavior inuencesnonconsequentialist decisions in complex ways suf-ciently complex that some researchers regard moral judg-ments as comparable in intricacy to the universal grammarof natural language (for ideas about a universal moralgrammar, see Hauser, 2006; Mikhail, 2007).

A narrow focus on behavior per se, rather than conse-quences, is a signature feature of moral cognition. Reason-ing surrounding the concepts right and wrong ofteninvolves judgments based on behavior per se rather thanthe instrumental effects of behavior. No other decision-making domain seems to share this peculiarity. For exam-ple, people often use rules in precautionary reasoning, suchas Do not drive through stop signs, but they hardly main-tain such rules even when this will lead to much greaterharm (e.g., a truck skidding out of control from behind).Outside of morality, a narrow concentration on behaviorper se, rather than consequences, is regarded as patholog-ical (obsessive-compulsive disorder, see Boyer & Lienard,2006).

4.3. Nonconsequentialism undermines consequentialistexplanations

Evolved mechanisms are organized to bring about par-ticular consequences, thereby allowing organisms to pro-tect, repair, and replicate themselves. Behavioraladaptations work by evaluating the environment and rec-onguring the organism in such a way that certain conse-quences (e.g., predatory attack) are more or less likely.These mechanisms tailor behavior to t the changingworld. Thermoregulation devices, for instance, track tem-perature changes and activate behaviors (e.g., shivering,sweating) that bring about specic consequences (targetbody temperature). Mate choice mechanisms track infor-mation about others sex/age/health and activate behaviors(e.g., courtship, copulation) that bring about specic conse-quences (fertilization). In short, evolution generally pro-duces consequentialist mechanisms. This does not, ofcourse, imply conscious calculation of consequences (e.g.,sweating), but only designs that are well organized tobring about adaptive outcomes (Dawkins, 1976).

Moral nonconsequentialism in humans is puzzling be-cause the underlying mechanisms focus on behavior perse rather than on consequences. This decision procedureseems to undermine the very raison dtre of behavior: todynamically respond to different problems. Rules of behav-ior that are insensitive to welfare consequences such asnever kill or never exchange sex for money (compareto never shiver) are, by denition, insensitive to context,which can lead to damaging sometimes disastrous con-sequences. What explains why moral judgment rejects cer-tain behaviors, even when the costs are severe?

Nonconsequentialism raises doubts about altruism the-ories of moral conscience. The a priori prediction of altru-ism theories seems clear: Moral conscience should beconsequentialist. In non-human animals, altruism (e.g.,parental care) is regulated by outcome-driven processesin which the performance target is increasing others wel-fare (e.g., offspring; Krebs & Davies, 1993). Kin selection fa-vors mechanisms that increase inclusive tness by helpingrelatives. When dilemmas arise, such as food shortages,many animals show their consequentialism by neglectingor even killing their offspring or siblings in order to allo-cate resources more efciently (Mock, 2004). Similarly,other pathways to altruism such as reciprocity or groupselection should yield consequentialist mechanisms, notinexible rules. Group selection, for instance, should favora design that sacrices one group member to save ve.

More broadly, the candidate adaptive problems thatsurround theorizing in the morality literature altruism,disease avoidance, indelity, etc. ought, barring excep-tional circumstances, to lead to adaptations designed tobring about a solution to the adaptive problem (deliveringbenets, avoiding pathogens, etc.). The striking and persis-tent deviations from consequentialism in moral conscienceundermine theories that posit direct consequentialistfunctions.

4.4. Nonconsequentialism: error or evidence?

A riposte to the above argument is that what appears tobe nonconsequentialism is actually error, arising from asystem implementing rule consequentialism. That is, simpleheuristics such as never kill increase welfare on balance,even though they produce some errors. These heuristicrules might be implemented by conscious reasoning (Bar-on, 1994; Sunstein, 2005) or by unconscious emotions(Gigerenzer, 2007; Greene, 2008, p. 60). Further, the arti-cial dilemmas used in experiments might exaggerate er-rors, giving a misleading impression of their magnitude.Natural selection can favor fast and frugal heuristics whentime, knowledge, or computational abilities are limited(Gigerenzer & Goldstein, 1996). Do moral rules functionas heuristics for welfare?

(Kant, 1785/1993). In fact, research has turned up surpris-ingly little evidence that moral judgment motivatesmorally compliant behavior; instead, it often motivates ef-forts to appear moral (Batson, 2008). Second, conscience is

P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299 289We think this error explanation is unlikely. What spe-cial limitations on time, knowledge, or computation ex-plain why moral problems require extremely simple rules(more than other decision domains)? All real world prob-lems are complex, but computational systems usuallymanage complexity by evaluating a few key dimensions(rather than none). A persuasive argument that moral rulesare heuristics will have to specify the properties of moralproblems that explain why extremely simple rules are re-quired to solve them. The current arguments are cast sobroadly that if they were correct, then all cognitive func-tions would be performed with commandment-like rules.Given that this is not the case, these arguments need tobe reconsidered.

Consider, for example, a design that focuses on severalcomputations. Conscience mechanisms could compute fornearby individuals the physical harm likely to occur inthe immediate future this system would favor killingone person to save ve. These computation requirementsdo not seem exceptionally burdensome, involving calcula-tions that people appear to already perform (in precaution-ary reasoning) and that seem readily available inindividuals welfare judgments. Indeed, in most of the rel-evant experiments, welfare costs were high and easy tocompute. Often, participants were asked to assess bothwelfare and moral wrongness (Haidt, 2001; Haidt & Hersh,2001; Haidt et al., 1993; Tetlock, 2000), and, unlike moraljudgments, welfare judgments accurately tracked welfareoutcomes. Given that people are able to compute welfarein these situations, simple moral rules offer no computa-tional advantage.

Further, moral conscience is not, in fact, cognitivelysimple. On the contrary, experimental ndings show thatpeople are sensitive to many dimensions of actor behav-ior, even when consequences are ignored (Hauser, 2006;Robinson & Darley, 1995; also see Section 6). As men-tioned above, the complexity of moral judgment has ledto theories surrounding a universal moral grammar(Hauser, 2006; Mikhail, 2007). Conscience is no cognitivemiser, so why are so little computational resources de-voted to evaluating welfare outcomes when making mor-al judgments?

Evaluated for ight, penguins wings appear awed andsuboptimal. At some point, it makes sense to considerwing features not as defects but rather as evidence of analternative function, like swimming. If moral nonconse-quentialism points to an alternative function for con-science, then attempts to explain conscience independentof condemnation are like trying to explain a penguin wingout of water. Because the evolutionary function of moralityis unknown, patterns in moral cognition cannot be safelydismissed as error. Instead, consistent features are bestviewed as potential clues to the adaptive problems thatshaped the system.

4.5. Summary

Current theories of moral conscience are problematic.First, non-moral systems can cause moral compliance,making it difcult to know the extent to which conscienceis responsible for altruism, delity, incest avoidance, etc.often nonconsequentialist, focusing narrowly on behaviorper se rather than consequences. Because this is peculiarto conscience, it seems safe to associate nonconsequential-ism with moral concepts, whereas consequentialist judg-ments are more difcult to trace to moral as opposed tonon-moral mechanisms. Last, a decision procedure thatleads one to absolutely avoid certain behaviors does notappear well designed to accomplish any direct actor goal,undermining theories that posit direct consequentialistfunctions.

We think that the key to unlocking the problem of non-consequentialism is the strategic dynamic among perpe-trator, victim, and condemner. Decision procedures thatare never advisable in individual decision-making aresometimes advantageous in multi-player strategic interac-tions because they inuence others behavior (Schelling,1960). For example, commitment mechanisms (e.g., envi-ronmentalists chaining themselves to threatened trees) re-duce agents own freedom of action, but this can inuenceother agents behaviors in desirable ways (e.g., Frank,1988; Hirshleifer, 1987). Perhaps the structure of moralconscience can be illuminated by considering how actorsinteract with third-party condemners.

5. Condemnation-centered morality

We now turn to an exploration of the potential value ofexamining moral systems starting with condemnationrather than conscience.

5.1. Conscience does not explain condemnation butcondemnation explains conscience

Theories proposed to explain actor conscience do not,without additional assumptions, concurrently explainthird-party condemnation. For example, it has been pro-posed that incest aversion functions to guide the organismto avoid inbreeding costs (e.g., Lieberman et al., 2003). Thisdoes not, by itself, explain why humans want others pun-ished for incest.6 In fact, individuals with a disposition topraise others tness-reducing moral violations, such as in-cest or suicide, could in theory gain an advantage by manip-ulating competitors to engage in self-destructive behavior.

There is at present no well-developed model that pro-vides a direct pathway from conscience to condemnation.That is, there is no reason to expect that moral systems de-signed to avoid certain acts (conscience) will cause repre-sentations that others should avoid those acts and bepunished if they do them (condemnation). One possibleconscience-to-condemnation relationship is that once con-science mechanisms evolved, this created selection favor-ing individuals with condemnation systems. However, it

6 For instance, in the United States legal system, all fty states and theDistrict of Columbia have some variation of a prohibition of incest on thebooks (Inbred Obscurity, 2006, p. 2469). In Massachusetts, consensualincest is punishable by up to 20 years in prison.

290 P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299is difcult to see how the presence of conscience mecha-nisms might favor mutations associated with condemna-tion mechanisms. Another possible approach would be toassume that conscience mechanisms are designed to regu-late ones own behavior but they sometimes overgeneral-ize to regulate others behavior. However, such extremeerrors failure to distinguish ones own behavior from oth-ers behavior imply an exceptionally poor design (Cos-mides & Tooby, 2000) and have been observed only incases of severe pathology (e.g., self-monitoring decits inschizophrenia; Frith, 1996). Moreover, whereas overgener-alization could cause individuals to experience vicariousaversion to others violations, it seems insufcient to ex-plain why people want violators to be punished. Moralisticpunishment is a new feature not present in conscience.

Whereas conscience does not explain third-party con-demnation, given third-party condemnation, an explanationfor conscience straightforwardly follows. In a population ofpeople who condemn others for certain behaviors, selec-tion will favor defense systems that guide individuals toavoid those behaviors (unless the benets outweigh thecosts of being punished). Indeed, models show that, givencertain assumptions, punishment can favor the evolutionof any behavioral system (Boyd & Richerson, 1992, 2005).Actors can estimate condemnation costs by representingthe computations of third-party condemners, whateverthose computations might be. In this scenario, conscienceis designed to apply the concepts of right and wrong to reg-ulate ones own behavior to defend against third-partycondemners.

Mencken (1949) wrote, Conscience is the inner voicethat warns us somebody may be looking (p. 617). If con-science functions, in part, as a defense system, then thismight help explain nonconsequentialist decisions. Con-science might focus on behavior per se rather than conse-quences because third-party condemners focus onbehavior. Of course, this does not solve the problem ofnonconsequentialism, but it potentially consolidates twopuzzles nonconsequentialism in both actors and con-demners into one puzzle nonconsequentialism incondemners.

5.2. Conscience as a defense system: evidence from moralhypocrisy

The idea that conscience functions as a defense systemmight explain a ubiquitous feature of human life: moralhypocrisy, the gap between individuals moral standardsand their behavior (Batson, Kobrynowicz, Dinnerstein,Kampf, & Wilson, 1997). People often engage in behaviorthat they believe is morally wrong. What explains this?One possibility is that immoral behavior results from a fail-ure of self-regulation mechanisms, perhaps due to compet-ing motives. This account posits errors in systemintegration between conscience and other cognitivemechanisms.

Another possibility is that conscience is a defense sys-tem that allows immoral behavior when condemnation isunlikely. For example, when persuasive justications havebeen identied, conscience might facilitate immoralbehavior; this could explain why justications can causemoral disengagement (Bandura, Barbaranelli, Caprara, &Pastorelli, 1996). After behaving immorally, maintaininga representation of the behavior as morally wrong canfacilitate concealment or, in the event of detection, deploy-ment of the necessary justications or reparations. Moralhypocrisy, therefore, might simply reect the functioningof conscience as a system for defense rather than directbehavior regulation.

Moral hypocrisy has been demonstrated in the labora-tory (Batson & Thompson, 2001; Batson, Thompson, Seu-ferling, Whitney, & Strongman, 1999; Batson et al., 1997).In a typical experiment, participants assigned themselvesand another individual to different tasks, one more desir-able than the other. Participants could ip a coin to chooserandomly or they could simply choose. Retrospectively,nearly all participants judged ipping the coin as morallyright. Roughly half of participants ipped the coin, butwhether they used the coin or not, 90% assigned them-selves to the desirable task, showing that participants typ-ically ignored the coin ip. Batson and Thompson (2001)concluded that at least some individuals want to appearmoral while, if possible, avoiding the cost of actually beingmoral (p. 54).

Similarly, studies of childrens tattling show they ea-gerly report others wrongs but ignore their own transgres-sions (den Bak & Ross, 1996; Dunn & Munn, 1985). Talwar,Lee, Bala, and Lindsay (2002) found no relationship be-tween childrens understanding of lying and truth-tellingand their own lying behavior. Most children correctly dis-tinguished lies from truth, judged a ctional liar as wrong,recommended that other children should tell the truth, andclaimed that they would tell the truth. However, most ofthese same children lied to conceal their own cheating.In a similar study (Talwar, Lee, Bala, & Lindsay, 2004), chil-dren lied to hide their parents transgression unless theythemselves might be implicated in the infraction.

5.3. Lady Justice

In contrast to the historical focus on actor conscience intheories of morality, we propose that third-party condem-nation is a key perhaps the key explanandum of moral-ity. If conscience functions, in part, as a defense system,then its structure cannot be understood independent ofthe structure of condemnation, which conscience shouldclosely parallel.

The potential centrality of condemnation has importantimplications for empirical research. For example, considerthe action-omission effect, which leads people to view kill-ing someone as more wrong than letting someone die. Thisphenomenon can be approached from two perspectives:(1) How does the action-omission effect benet actors?or, (2) How does the action-omission effect benet third-party condemners? The condemnation-centered view fore-grounds the latter question.

This returns us to the threefold mystery of moralitysymbolized by Lady Justices scale, sword, and blindfold(see Fig. 2). First, why do people care, at all, about moralviolations occurring among unrelated others? Second,why do people punish moral violators, even when punish-ing is costly? Third, why do people try to appear impartial,

gressors punished, whether in fact or ction.

evaluate others altruism/selshness. Broadly, this predicts

IES evalordin

ts onht dr

luables imp

ysteri

P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299 291Research on childrens tattling shows that moral inter-est begins early in development (den Bak & Ross, 1996;Dunn & Munn, 1985; Ross & Den Bak-Lammers, 1998).Even before age two, children call attention to others vio-lations (while ignoring their own). Roughly half of thetime, the tattler is unaffected by the violation. Further, tat-tling persists even when parents react negatively to snitch-claiming to neglect relationships with kin, friends, and al-lies in their moral decisions and actions? Just as sexualreproduction initially appeared biologically bizarre withseemingly insuperable costs (Hamilton, 2001; Williams,1975), third-party judgment, punishment, and impartialityrepresent biological mysteries of the rst order.

6. Moral judgment

In the biological world, organisms monitor othersinteractions only when the benets exceed the costs.Third-party monitoring can sometimes help organismsnd food (e.g., wasps, see above), quality mates (matecopying, White & Galef, 1999), or avoid costly ghts (usingtransitive inference, Doutrelant, McGregor, & Oliveira,2001). A condemnation-centered view of morality high-lights the question: What benets explain why peoplemonitor others moral behavior, even when they them-selves are unaffected?

Interest in moral infractions is enormous. The pressdraws audiences with reports of celebrity debauchery,political corruption, and corporate fraud. Medieval publicexecutions and modern superhero movies (DeScioli &Kurzban, 2008) attest to widespread desire to see trans-

THREE MYSTER

Moral Judgment Humans monitor andunrelated others acc

Moral Punishment Humans impose cosrules, even if this mig

Moral Impartiality Humans damage vato enforce moral rule

Fig. 2. Three ming. Finally, tattling precedes, and is uncorrelated with,positive talk, leading Ross and den Bak-Lammers (1998)to conclude that tattling does not emerge from the moregeneral experience of sharing information about the sib-ling with the parent, but is an earlier, largely unrelatedform of parent-child discourse (p. 294).

Continuing into adulthood, moral judgment is a stapleof everyday conversation. Gossip, often about wrongdoing,is cross-culturally ubiquitous (Barkow, 1992; De Backer,2005; Dunbar, 1996; Gluckman, 1963; Haviland, 1977).Roughly 65% of conversation is about social topics (Dunbar,2004). As in childrens tattling, moral discourse is dispro-portionately negative. This pattern has been observedcross-culturally. For instance, among the Ju/hoansi Bush-that moral judgment should be a function of intended ben-et or harm. But this seems inconsistent with observedinsensitivities to welfare outcomes (e.g., the footbridge di-lemma, Section 4). If moral judgment measures altruism,then why dont welfare outcomes dominate judgment?Similarly confusing are ndings showing that the severityof unintended consequences can inuence moral judgment(Rucker, Polifroni, Tetlock, & Scott, 2004). Participantsjudged a hypothetical carjacker who harmlessly ejected apassenger as less wrong than a carjacker who inadver-tently caused a heart attack. As Rucker et al. noted, Surely,we learn nothing new about the moral character of the car-jacker in the high-severity versus the low-severity condi-tion (p. 673). These observations suggest that moralmen, 56% of conversation contained moralistic criticism,whereas 7% contained praise (Wiessner, 2005).

Moral surveillance is an active process. People seek evi-dence, interrogate claimants, and interview eyewitnesses,whose testimony is given more credence than secondarysources (Wilson,Wilczynski,Wells, &Weiser, 2000). Peopleare sophisticated judges, comparing facts from multiplesources, attributing greater credibility to facts repeated bymultiple independent sources, and discounting informationfrom sources with vested interests (Hess & Hagen, 2002).

Humans seek, gather, evaluate, and communicate infor-mation about wrongdoing among others. What benetsoffset the costs of these time-consuming behaviors?

6.1. What is moral judgment good for?

One possibility is that moral judgment functions to

OF MORALITY uate interactions among g to moral rules of behavior.

individuals who violate moral aw retaliation.

social relationships in order artially.

es of morality.judgment performs some alternative or additional func-tions other than assessing others altruism.

Rather than consequences, moral judgment seems to beparticularly attuned to properties of perpetrator behavior,including whether actors: (1) act or fail to act, (2) act inten-tionally or accidentally, (3) act with contact or at a dis-tance, and (4) violate directly or as a byproduct (seeHauser, 2006). First, acts are judged more harshly thaninaction, even when the latter causes more harm (Baron& Ritov, 1994, 2004; Cushman, Young, & Hauser, 2006;Haidt & Baron, 1996; Ritov & Baron, 1990, 1992, 1995,1999; Spranca, Minsk, & Baron, 1991). For instance, recom-mending a salad dressing with cayenne pepper to an aller-gic tennis competitor was seen as more wrong than

distinction between moralistic punishment and second-

292 P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299remaining silent when the competitor chose himself (Spr-anca et al., 1991). The effect has also been found in children(8 years: Baron, Granato, & Spranca, 1993; 1117 years:Keltikangas-Jrvinen & Lindeman, 1997).

Second, intended violations are perceived as morewrong than accidental ones (Robinson & Darley, 1995),holding goals constant. Pizarro, Uhlmann, and Bloom(2003) looked at causal deviance, in which an agent in-tends to cause an outcome but causes the outcome in amanner other than the intended one. Participants evalu-ated a protagonist who planned to stab his enemy andeither did so, or, was bumped and stabbed him acciden-tally. The accidental killing was viewed as less blamewor-thy. The intention/accident distinction has been found asearly as age three (Ferguson & Gail, 1988; Grueneich,1982; Leon, 1984; Nuez & Harris, 1998; Sedlak, 1979).

Third, transgressions involving physical contact areseen as more immoral. Recall that when an individual fac-ing the Trolley Problem can ip a switch to divert the trol-ley (rather than push the man), killing one to save ve,judgments change dramatically: Only 10% viewed this aswrong (vs. 90% for pushing the man; Hauser, 2006).

Fourth, violations occurring as a foreseen byproduct ofanother act are seen as less wrong than violations usedas a means to an end. In another variation of the TrolleyProblem, there is a sidetrack that loops back to the originaltrack. Diverting the trolley to the sidetrack can save vepeople, but only if there is a heavy object on the sidetrack.When the object was a heavy man, 50% of participantsjudged diverting the trolley as impermissible. When theobject was a weight but a man standing in front of theweight will be killed only 25% judged diverting the trol-ley as impermissible (Hauser, 2006; see also Royzman &Baron, 2002; Waldmann & Dieterich, 2007).

In sum, moral judgment is especially attuned to proper-ties of actor behavior. Moreover, people are often unawarethat these factors play a role in their decisions (Cushman,Young, & Hauser, 2006). What function might explainthese cognitive processes? It might help to think aboutthe problems confronted by third-party condemners. Oneproblem, for instance, is recruiting others support, whichoften requires substantiating evidence. Perhaps, for exam-ple, the action-omission distinction tracks the likelihood ofa persuasive moral accusation. Importantly, unpersuasiveaccusations can draw dangerous retaliation from the ac-cused and their allies. If providing evidence for a wrongfulact is easier than demonstrating a violating omission, thenthird-party condemners might benet by feeling less moraloutrage for omissions relative to actions (DeScioli, 2008).This is one way that the action-omission distinction mightcontribute to a condemnation function.

6.2. Scales of justice

Other features of moral judgment require explanation.First, the ontology of legitimate perpetrators or victims(animals, inanimate objects, etc.) shows striking variation(Kadri, 2005; Singer, 1981). Next, people are motivated toremain consistent in their moral judgments, leading tostrong order effects in experiments (Hauser, 2006). Also,people dislike moral diversity, preferring group membersparty revenge (see also Nozick, 1981). Morality and ven-geance intersect in moral rules that place limits on retalia-tion (e.g., lex talionis, e.g., eye for an eye). However, moralrules about revenge do not imply an equivalence betweenthese forms of punishment. An evolutionary perspectiveunderscores important differences between revenge andmoralistic punishment.

Second-party punishment is widespread in nature(Clutton-Brock & Parker, 1995). Design for damaging oth-ers, even when costly, can yield an evolutionary advantageby altering others behavior. Punishment occurs in a vari-ety of contexts, including dominance relationships, compe-tition for resources, mating interactions, parent-offspringconict, and cooperative breeding (Clutton-Brock & Parker,1995). Egret chicks punish their smaller siblings for com-peting to get food brought by the parent (Mock, 2004).Mother elephant bite unrelated pups who try to drink theirmilk (Reiter, Stinson, & Le Boeuf, 1978). In several primatespecies, males punish females who refuse their attempts tomate (Smuts & Smuts, 1993). These examples illustratehow animals use punishment in a variety of contexts to de-ter or coerce others.

In contrast, third-party punishment is rare among non-humans, with several possible exceptions. Social insects killworkers for laying eggs (Foster & Ratnieks, 2001; Gobin, Bil-len, & Peeters, 1999), hyenas intervene in others ghts(Engh, Siebert, Greenberg, & Holekamp, 2005), and chim-panzees intervene on behalf of unrelated allies (de Waal,to agree with them (Haidt, Rosenberg, & Hom, 2003).Further, individuals readily compare violations acrossdomains, and severity rankings sometimes show remark-able consistency (Robinson & Kurzban, 2007).

Moral judgment is complex. While moral judgments arenot uniquely determined by welfare, authority, custom,precedent, etc., each of these factors can inuence judg-ment. Emotions such as disgust and anger can exert stronginuences (Haidt & Hersh, 2001; Haidt et al., 1993; Rozinet al., 1999). Also intriguing but infrequently discussed isthe historically widespread practice of trial by ordeal orcombat (Kadri, 2005), in which surviving re, boiling oil,or dueling determines moral rightness. Why can moraljudgment be inuenced by arbitrary contests?

These complexities suggest that the psychologicalscales of moral judgment are tipped by many factors. Mor-al dilemmas might occur when welfare is pitted againstauthority, custom contradicts emotion, and so on. Culturalvariation is expected because it is unlikely that preciselythe same balance of forces will be struck for all moral is-sues across different societies (Shweder et al., 1997). Vari-ation should taper when factors reach extremes orconverge (welfare and disgust both oppose boiling babiesfor fun). The reason that moral judgment possesses thesesensitivities should be claried by asking the question:What adaptive functions are served by monitoring othersmoral violations?

7. Moralistic punishment

Focusing on third-party condemnation highlights the

1996). Among pigtail macaques, Flack, de Waal, and Kraka-uer (2005) found that third parties intervened in 72% ofconicts; 13% of interventions drew subsequent retaliation.

Explanations for second-party punishment are rela-tively straightforward (e.g., relative advantage, deterrence,

While the previous studies examined the prevalence of

P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299 293extortion), but third-party punishment is generally moredifcult to explain (kin selection for social insects; perhapsalliance-building for somemammalian cases). The problemis especially difcult because third-party condemners suf-fer costly retaliation from perpetrators. What benets off-set the costs of third-party punishment?

7.1. Field and laboratory observations

Moralistic punishment varies in intensity and costs.7

Dissemination of information about others infractions dam-ages violators reputations (Hess & Hagen, 2002). Even thisform of punishment can be costly. Among the Ashanti ofGhana, tale-bearing concerning the private affairs of a Chiefwas often punished by cutting off the lips (Rattray, 1929, p.327). Publicly ridiculing, shaming, or complaining abouttransgressors risks undermining valuable relationships(Kaplan & Hill, 1985). Miller (2003) investigated third-partysanctions of people who illegally parked in handicappedspaces. Punishment ranged from cold stares to tire slashingand was sometimes met with retaliation; one condemnerwas physically assaulted by three men. Among the Gebusiof New Guinea, Knauft (1987) documented 82 executionsof alleged disease-causing sorcerers; 5% of the executionerswere themselves killed in retaliation by the sorcerers kin.Among the Ju/hoansi Bushmen of Botswana, Wiessner(2005) found that punishment increased in intensity as vio-lations were repeated, beginning with mockery and insults,followed by harsh criticism and nally violence. Most pun-ishments were administered by coalitions of three or moreindividuals, and 8% of these group punishments yieldedtransparent costs to punishers, usually by interrupting reci-procal sharing; 3% ended in vicious ghts.

Research in social psychology on bystander interven-tion (Latan & Nida, 1981) shows that people are oftenwilling to incur costs to stop others violations. Studieshave found high rates of third-party intervention forwrongdoing, including assault (65%, Shotland & Straw,1976; 44%, Fischer, Greitemeyer, Pollozek, & Frey, 2006),rape (65%, Harari, Harari, & White, 1985), theft (57%, How-ard & Crano, 1974; 28%, Gelfand, Hartmann, Walder, &Page, 1973), and grafti and littering (49% and 63%, respec-tively, Chekroun & Brauer, 2002).

More recently, researchers have used economic gamesto examine third-party punishment. Turillo, Folger, Lavelle,Umphress, and Gee (2002) found that third parties paid topunish, 15% of the time, individuals who unfairly dividedmoney with another individual. Fehr and Fischbacher(2004) found that 60% of third parties were willing to payat least a small amount to punish unfair individuals. Also,

7 Elsewhere we have investigated whether third-party punishment isdriven by reputation benets, arguing that third-party punishment is rarein anonymous laboratory environments. Here we focus, more generally, onthe theoretical challenges posed by third-party punishment, independentof whether it occurs in public or private.third-party punishment, economic games can also be usedto look at the inuences of key variables. Kurzban, DeScioli,and OBrien (2007) examined the inuence of anonymity.Third-party participants could pay to punish individualswho violated reciprocity in a sequential prisoners dilem-ma. When participants knew they would announce theirdecisions to other third parties, they spent roughly threetimes more on punishment (average $3.17) than (different)participants in an anonymous condition (average $1.06).Interestingly, in post-experiment surveys, not a single sub-ject in the public condition mentioned the audience as afactor in their decision. Audience effects might be specicto third-party punishment, as second-party revenge hasbeen observed to be insensitive to anonymity (Bolton &Zwick, 1995).

7.2. Accounting for cost

The costs of moralistic punishment imply an importantdistinction between: (1) wanting wrongdoers to be pun-ished, and (2) wanting to punish wrongdoers. This distinc-tion is often neglected (e.g., Singer et al., 2006). Because theunderlying systems should be sensitive to cost, the feelingthat a wrongdoer deserves punishment is distinct frommotivation to perform punishment oneself (Robinson,Kurzban, & Jones, 2007). Historically, public executionshave drawn large audiences, reecting a strong desire tosee wrongdoers punished; this does not show observerswillingness to perform costly punishment themselves.

In conclusion, while second-party punishment is bio-logically common, third-party punishment is rare in non-humans and presents special theoretical challenges. Pre-cisely what benets explain why people have cognitivesystems that perform costly third-party punishment?

8. Moral impartiality

Lady Justices blindfold symbolizes that condemnationshould not depend on who is helped or harmed. Moralimpartiality requires condemners to ignore kinship, friend-ship, and group loyalty. Sometimes people show moralimpartiality, like when David Kaczynski turned in hisbrother Theodore, the Unabomber, for killing three peo-ple. Sometimes people are partial, like when politicianWil-liam Bulger refused to help authorities nd his brother,wanted for 19 murders. Here we consider the theoreticalchallenges posed by moral impartiality. We do not, ofthird-party parties spent 8% of their endowment to punishpeople who defected when their partner cooperated in aPrisoners Dilemma game. Carpenter and Matthews(2005) examined punishment in a Public Goods Game (Led-yard, 1995). They found that 10% of participants showedthird-party punishment of non-contributors, spending anaverage of only $0.10 total on third-party sanctions. Hen-rich et al. (2006) looked at third-party punishment for un-fair divisions in 15 cultures around the world. Overall,66% of third parties were willing to pay to punish the mostunfair division (100% allocated to oneself), with this gureranging from 28% to over 90% across cultures.

294 P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299course, assume that people are always impartial, or eventhat they are usually impartial. Instead, we ask why peopleare ever impartial to any degree at all, why they often claimto be impartial, why they praise impartiality, and why hu-mans possess the concept of moral impartiality. The mys-tery of moral impartiality surrounds its inevitableconicts with loyalty.

Focusing on third-party condemnation shifts howimpartiality itself is construed. From the perspective ofconscience, impartiality is the opposite of selshness, i.e.,actors own preferences should not inuence moral judg-ments. The view from third-party condemnation is mark-edly different. For third parties, with little or no directinterest in the interaction, moral impartialitys oppositeis favoritism. From the third-party perspective, partialityis not selshness, but loyalty to kin, friends, and groups.

Third-party impartiality is a frictionless machine ofbiology seemingly impossible. Theories of altruism re-quire partiality (toward kin, reciprocating partners, orgroups), which is made possible by specialized discrimina-tion mechanisms (Axelrod & Hamilton, 1981; Cosmides &Tooby, 1992; Trivers, 1971). That is, partiality is a specicstrategy underpinned by specialized cognitive machinery.Impartiality undermines this machinery, threatening valu-able relationships. Even the pretense of impartiality couldweaken trust and damage relationships.

Third-party punishment among non-human animalstends to be partial. Among hyenas, when third partiesjoined ghts among unrelated others, they sided with thehigher status individual 94% of the time (Engh et al.,2005). Chimpanzees possess a system of revenge inwhich individuals side against those who side againstthem (de Waal & Luttrell, 1988). There are, however, someexceptions: Some high-ranking individuals in primategroups take a control role in managing conicts (see deWaal, 1996). Among pigtail macaques, 29% of third-partyinterventions were impartial: the third party targeted bothcombatants or stood neutrally between them (Flack et al.,2005).

Impartiality is rare in nature and poses evolutionarychallenges. Importantly, moral impartiality is not ex-plained by theories of altruism. In fact, altruism poses theproblem:Why do people ever put moral impartiality abovefamily, friends, and coalitions?

8.1. The concept of moral impartiality

The concept of impartiality is distinct from being unbi-ased or consistent (Gert, 2005). A baseball umpire might bebiased toward a narrow strike zone, but can still be impar-tial with respect to the two teams. An inconsistent umpire,randomly switching strike zones, can also be impartial. In-stead, the concept of impartiality requires the specicationof the relevant group and the respect in which the actor isimpartial (Gert, 2005). Gert (2005) proposed the denition:A is impartial in respect R with regard to group G if andonly if As actions in respect R are not inuenced by whichmember(s) of G benet or are harmed by these actions (p.132). Moral impartiality is the specic case in which moralrules are applied impartially to the group of moral agents(Gert, 2005).8.2. Field and laboratory observations

When US soldiers were asked whether they would re-port a unit member for intentional and unnecessary vio-lence against innocent foreign civilians, 45% said theywould not (Morgan, 2007). On the view presented here,what requires explanation is the 55% who claimed theywould report a unit member. Why would someone con-demn fellow group members for violations againststrangers?

Moral impartiality is reected in various cultural forms.In philosophy, it is seen in Kants (1785/1993) universaliz-ability criterion, Smiths (1759) impartial spectator, andRawls (1971) veil of ignorance. Impartiality is reectedin age-old principles of justice such as audi alteram partem(hear both sides). The Bible states, Wrongdoers will payfor the wrong they do; there will be no favoritism (Colos-sians 3:25). Modern superheroes are portrayed as impar-tial, even saving evildoers to turn them over to theauthorities (despite the fact that leaving villains aliveendangers humanity).

The ideal of moral impartiality is present cross-cultur-ally. Judges among the Lozi of Zambia announced to liti-gants that they decide, not by sobozi (partiality orprejudice), but by bupaki (evidence), lisupo (indications,probabilities, presumptions), and libaka (reasons, reason-ing) (Gluckman, 1967, p. 105). In Gebusi sorcery inquests,sickness-causing sorcerers were identied by spiritualmediums who, to prevent partiality, would often be drawnfrom an outside community (Knauft, 1987).

The costs of impartiality are visible in cultural means ofmitigating them. Among the Ashanti of Ghana, people be-lieve that the Chiefs decisions channel the ancestral spir-its, rather than reecting favoritism. Rattray (1929)reports, His judgments were regarded, not as emanatingfrom his own mouth, but as being the decisions of his deadancestors. He was thus placed largely beyond the risk ofopprobrium, which a just decision, but one displeasing toa certain and possibly powerful faction, might haveoccasioned (p. 289). Among the Azande of Central Africa,chiefs avoid close blood-brotherhood relationships to helpmaintain impartiality: Chiefs have to settle cases and dis-pense justice and direct administration. An alliance ofblood would militate against the fairness of their judg-ments and paralyze their execution (Evans-Pritchard,1933, p. 374). Finally, impartiality is sometimes used foronly the most severe offenses. Among the Ifugao of thePhilippines, people are obliged to support kin in all dis-putes, but the only exception to this rule is sorcery, an of-fense so severe that people turn against family members(Barton, 1919, p. 70).

Turning to laboratory research, Lieberman and Linke(2007) examined the inuence of kinship and group mem-bership on third-party judgments (wrongness and punish-ment). Wrongness judgments were not inuenced bysocial category. However, perpetrators of theft werejudged as deserving less punishment (but not less wrong)when they were kin (versus nonkin) and when they wereingroup (versus outgroup) members. This might indicatethat wrongness judgments tend to be more impartial thanpunishment decisions. However, wrongness ratings were

pants showed less retributive emotion for ingroup than

nonconsequentialist, casting doubt on theories that posit

P. DeScioli, R. Kurzban / Cognition 112 (2009) 281299 295outgroup violators (ingroup bias), but the reverse occurred(outgroup bias) when guilt was certain.

Bernhard, Fischbacher, and Fehr (2006) looked at in-group partiality among two indigenous groups in PapuaNew Guinea. In the rst stage of an experimental game,Player A divided money with Player B (Dictator Game).In the second stage, a third party could punish player Afor making an unfair division; for each monetary unitspent on punishment, Player As payoff was reduced bythree units. There were four conditions manipulating in-group membership: The punisher was either in the samegroup as the (potential) perpetrator and victim, the per-petrator only, the victim only, or neither. Interestingly,perpetrators group memberships did not affect punish-ment decisions. People did punish more, however, whenthe victim was an ingroup member (versus outgroupmember).

8.3. Loyalty vs. moral impartiality

Accounts that link morality with loyalty and personalties (e.g., Fletcher, 1993; Gilligan, 1977) are underminedby the inevitable conict between altruism and moralimpartiality (see also Batson et al., 1995). Loyalty and per-sonal caring are alternatives to moral impartiality. This ten-sion is illustrated by the Brazilian maxim, For friends,everything; for enemies, the law, and the colloquialism,My country right or wrong. One cannot be both a goodaltruist and a good (impartial) moralist. For this reason, itwould seem that people should prefer partial rather thanimpartial partners. The possibility of false accusation high-lights the crucial difference. With ambiguous evidence,loyal altruists, not impartial condemners, will grant thebenet of the doubt. In fact, the potentially lethal threatof false accusation an inevitable consequence of moraljudgment should favor the evolution of intensied loyal-ties to family, friends, and groups to counteract the threatof impartial condemnation.

Despite this, however, humans across cultures advocateneglecting loyalties when making moral judgments. Whena moralist condemns a perpetrator, the moralist is judgedby the yard