PSY22-6a

Neuron, Vol. 44, 389–400, October 14, 2004, Copyright 2004 by Cell Press

The Neural Bases of Cognitive Conflict and Control in Moral Judgment

or emotional processes. Here we use quotation marks to indicate the latter usage.)

Personal moral dilemmas and judgments concern the

Joshua D. Greene,1,2,* Leigh E. Nystrom,1,2

Andrew D. Engell,1,2 John M. Darley,1

and Jonathan D. Cohen1,2

appropriateness of personal moral violations, and we1Department of Psychology consider a moral violation to be personal if it meetsPrinceton University three criteria: First, the violation must be likely to cause2 Center for the Study of Brain, serious bodily harm. Second, this harm must befall aMind, and Behavior particular person or set of persons. Third, the harm mustPrinceton University not result from the deflection of an existing threat ontoPrinceton, New Jersey 08544 a different party. One can think of these three criteria in terms of “ME HURT YOU.” The “HURT” criterion picks out the most primitive kinds of harmful violations (e.g.,

Summary assault rather than insider trading) while the “YOU” crite- rion ensures that the victim be vividly represented as

Traditional theories of moral psychology emphasize an individual. Finally, the “ME” condition captures a no-

reasoning and “higher cognition,” while more recent tion of “agency,” requiring that the action spring in a

work emphasizes the role of emotion. The present direct way from the agent’s will, that it be “authored” fMRI data support a theory of moral judgment ac- rather than merely “edited” by the agent. Dilemmas that cording to which both “cognitive” and emotional pro- fail to meet these three criteria are classified as “imper- cesses play crucial and sometimes mutually competi- sonal.” As noted previously (Greene et al., 2001), these tive roles. The present results indicate that brain three criteria reflect a provisional attempt to capture regions associated with abstract reasoning and cogni- what we suppose is a natural distinction in moral psy- tive control (including dorsolateral prefrontal cortex chology and will likely be revised in light of future re- and anterior cingulate cortex) are recruited to resolve search. difficult personal moral dilemmas in which utilitarian An example of an impersonal moral dilemma is the values require “personal” moral violations, violations trolley dilemma (Thomson, 1986): A runaway trolley is that have previously been associated with increased headed for five people who will be killed if it proceeds activity in emotion-related brain regions. Several re- on its present course. The only way to save them is to gions of frontal and parietal cortex predict intertrial hit a switch that will turn the trolley onto an alternate differences in moral judgment behavior, exhibiting set of tracks where it will kill one person instead of five. greater activity for utilitarian judgments. We speculate Should you turn the trolley in order to save five people that the controversy surrounding utilitarian moral phi- at the expense of one? Most people say yes (Greene et losophy reflects an underlying tension between com- al., 2001). An example of a personal moral dilemma is peting subsystems in the brain. the footbridge dilemma (Thomson, 1986): As before, a

trolley threatens to kill five people. You are standing next to a large stranger on a footbridge spanning theIntroduction tracks, in-between the oncoming trolley and the hapless five. This time, the only way to save them is to push thisFor decades, moral psychology was dominated by de- stranger off the bridge and onto the tracks below. Hevelopmental theories that emphasized the role of rea- will die if you do this, but his body will stop the trolleysoning and “higher cognition” in the moral judgment of from reaching the others. Should you save the five othersmature adults (Kohlberg, 1969). A more recent trend by pushing this stranger to his death? Most people sayemphasizes the role of intuitive and emotional pro- no (Greene et al., 2001). The trolley dilemma, unlike thecesses in human decision making (Damasio, 1994) and footbridge dilemma, is impersonal because it involvessociality (Bargh and Chartrand, 1999; Devine, 1989), a the deflection of an existing threat (i.e., no agency—it

shift in perspective that has profoundly influenced re- is “editing” rather than “authoring”).

cent work in moral psychology (Haidt, 2001; Rozin et The rationale for distinguishing between personal and

al., 1999). Our previous work suggests a synthesis of impersonal moral violations/judgments is in part evolu-

these two perspectives (Greene and Haidt, 2002; Greene tionary. Evidence from observations of great apes sug-

et al., 2001). We have argued that some moral judg- gests that our common ancestors lived intensely social

ments, which we call “personal,” are driven largely by lives guided by emotions such as empathy, anger, grati-

social-emotional responses while other moral judg- tude, jealousy, joy, love, and a sense of fairness (de

ments, which we call “impersonal,” are driven less by Waal, 1996), and all of this in the apparent absence of social-emotional responses and more by “cognitive” moral reasoning. (By “reasoning” we refer to relatively processes. (As discussed below, the term “cognitive” slow and deliberative processes involving abstraction has two distinct uses, referring in some cases to infor- and at least some introspectively accessible compo- mation processing in general while at other times refer- nents [Haidt, 2001].). Thus, from an evolutionary point ring to a class of processes that contrast with affective of view, it would be strange if human behavior were

not driven in part by domain-specific social-emotional dispositions. At the same time, however, humans appear*Correspondence: [email protected]

Neuron 390

to possess a domain-general capacity for sophisticated time to respond (high-RT trials), as compared to trials in which the participant responds quickly (low-RT), re-abstract reasoning, and it would be surprising as well

if this capacity played no role in human moral judgment. flecting presumed conflict in processing. Likewise, we predicted that regions in the dorsolateral prefrontal cor-Thus, we sought evidence in support of the hypothesis

that moral judgment in response to violations familiar tex (DLPFC) would also exhibit increased activity for high-RT trials (as compared to low-RT trials), reflectingto our primate ancestors (personal violations) are driven

by social-emotional responses while moral judgment the engagement of abstract reasoning processes and cognitive control (Miller and Cohen, 2001).in response to distinctively human (impersonal) moral

violation is (or can be) more “cognitive.” Second, we tested the hypothesis that, in the dilem- mas under consideration, these control processes workOur previous results supported this hypothesis in two

ways (Greene and Haidt, 2002; Greene et al., 2001). First, against the social-emotional responses described above and in favor of utilitarian judgments, i.e., judgments thatwe found that brain areas associated with emotion and

social cognition (medial prefrontal cortex, posterior cin- maximize aggregate welfare (e.g., by sacrificing one life in order to save five others). In keeping with this hypothe-gulate/precuneus, and superior temporal sulcus/tem-

peroparietal junction) exhibited increased activity while sis, we predicted increased DLPFC activity for trials in which participants judged personal moral violations toparticipants considered personal moral dilemmas, while

“cognitive” brain areas associated with abstract reason- be appropriate, as compared to trials in which partici- pants judged personal moral violations to be inappropri-ing and problem solving exhibited increased activity

while participants considered impersonal moral di- ate. In other words, this hypothesis predicted that the level of activity in regions of DLPFC would correlatelemmas.

Second, we found that reaction times (RTs) were, on positively with utilitarian moral judgment. We emphasize that this prediction goes beyond those explored in ouraverage, considerably longer for trials in which partici-

pants judged personal moral violations to be appro- previous work. Previously, we found that different classes of moral dilemma (personal versus impersonal)priate, as compared to trials in which participants judged

personal moral violations to be inappropriate. No com- produce different patterns of neural activity in the brains of moral decision makers. Here we test the hypothesisparable effect was observed for impersonal moral judg-

ment. We compare this effect on RT to the Stroop effect that different patterns of neural activity in response to the same class of moral dilemma are correlated with(MacLeod, 1991; Stroop, 1935), in which people are slow

to name the color of the ink in which an incongruent differences in moral decision-making behavior. To test the predictions of this theory, we focused onword appears (e.g., “red” written in green ink). According

to our theory, personal moral violations elicit prepotent, a class of dilemmas that bring “cognitive” and emotional factors into more balanced tension than those featurednegative social-emotional responses that drive people

to deem such actions inappropriate. Therefore, in order in our previous work. For example, consider the follow- ing moral dilemma (the crying baby dilemma).to judge a personal moral violation to be appropriate

one must overcome a prepotent response, just as one Enemy soldiers have taken over your village. They have orders to kill all remaining civilians. You and somefaced with the color-naming Stroop task must overcome

the temptation to read the word “red” when it is written of your townspeople have sought refuge in the cellar of a large house. Outside, you hear the voices of soldiersin green ink. The sort of mental discipline required by

the Stroop task is known as “cognitive control,” the who have come to search the house for valuables. Your baby begins to cry loudly. You cover his mouthability to guide attention, thought, and action in accor-

dance with goals or intentions, particularly in the face to block the sound. If you remove your hand from his mouth, his crying will summon the attention of the sol-of competing behavioral pressures (Cohen et al., 1990;

Posner and Snyder, 1975; Shiffrin and Schneider, 1977). diers who will kill you, your child, and the others hiding out in the cellar. To save yourself and the others, youWe interpreted the behavioral results of our previous

study as evidence that when participants responded in must smother your child to death. Is it appropriate for you to smother your child in ordera utilitarian manner (judging personal moral violations

to be acceptable when they serve a greater good) such to save yourself and the other townspeople? This is a difficult personal moral dilemma. In responseresponses not only reflected the involvement of abstract

reasoning but also the engagement of cognitive control to this dilemma, participants tend to answer slowly, and they exhibit no consensus in their judgments. This di-in order to overcome prepotent social-emotional re-

sponses elicited by these dilemmas. lemma, like the other consistently difficult dilemmas used here, has a specific structure: in order to maximizeOur present aim was to further test our theory of moral

judgment by directly testing two specific hypotheses aggregate welfare (in this case, save the most lives), one must commit a personal moral violation (in this case,derived from the arguments above. First, we tested the

hypothesis that increased RT in response to personal smother the baby). According to our theory, this di- lemma is difficult because the negative social-emotionalmoral dilemmas results from the conflict associated with

competition between a strong prepotent response and response associated with the thought of killing one’s own child competes with a more abstract, “cognitive”a response supported by abstract reasoning and the

application of cognitive control. In keeping with this hy- understanding that, in terms of lives saved/lost, one has nothing to lose (relative to the alternative) and much topothesis, we predicted that the anterior cingulate cortex

(ACC), a brain region associated with cognitive conflict gain by carrying out this horrific act. We believe that the ACC responds to this conflict and that control-relatedin the Stroop and other tasks (Botvinick et al., 2001),

would exhibit increased activity during personal moral processes in the DLPFC tend to favor the aforemen- tioned “cognitive” response. We hypothesize that thesejudgment for trials in which the participant takes a long

Cognitive Conflict and Control in Moral Judgment 391

Figure 1. Relationships among Three Analyses

The present results are from three increas- ingly focused analyses of a single data set drawn from 41 participants who responded to moral dilemmas while having their brains scanned using fMRI.

control processes, insofar as they are effective, drive processes favoring a utilitarian response. Analysis 2 was performed only on high-RT trials because of the relativethe individual to the utilitarian conclusion that it is appro-

priate to smother the baby in order to save more lives. paucity of low-RT-utilitarian judgments and the need to control for RT.This case contrasts with “easy” personal moral dilem-

mas, ones that receive relatively rapid and uniform judg- ments (at least from the subjects within our sample). Results One such case is the infanticide dilemma in which a teenage mother must decide whether or not to kill her Replication of Previous Results

Previously, we distinguished between personal and im-unwanted newborn infant. According to our theory, this dilemma is relatively easy because the negative social- personal moral judgments and found that brain areas

associated with emotion and social cognition (medialemotional response associated with the thought of someone killing her own child dominates the weak or prefrontal cortex, BA 9/10; posterior cingulate/precu-

neus, BA 31/7; and bilateral superior temporal sulcusnonexistent “cognitive” case in favor of this action. Here there is no significant cognitive conflict and no need for (STS)/inferior parietal lobe, BA 39) exhibited relatively

greater activity for personal moral judgment, while brainextended reasoning or cognitive control. Thus, com- pared to the high-RT trials typically generated by cases areas associated with working memory and other char-

acteristically “cognitive” processes (right DLPFC, BA 46;like crying baby, the low-RT trials typically generated by cases such as infanticide should exhibit lower levels bilateral inferior parietal lobe, BA 40) exhibited relatively

greater activity for impersonal moral judgment (Greeneof activity in the ACC and DLPFC. The analyses required to test these assertions make et al., 2001). The present data, drawn from 41 partici-

pants, replicated each of these results (Table 1), as wellup a nested structure (Figure 1). Previously, we com- pared the neural activity associated with “personal” and as our previously reported behavioral results. The pres-

ent data set includes data from nine participants that“impersonal” moral judgments (Greene et al., 2001). In analysis 1, we tested our hypotheses concerning conflict were analyzed previously (Greene et al., 2001). A sepa-

rate analysis excluding data from these nine participantsmonitoring in the ACC and abstract reasoning and cog- nitive control in the DLPFC by comparing high-RT to yielded results consistent with those reported for the

full data set. This larger data set also revealed previouslylow-RT personal moral judgments. In analysis 2, we tested our hypothesis concerning the involvement of unobserved differences in neural activity between per-

sonal and impersonal moral judgment (Table 1), includ-DLPFC in “cognitive” processes underlying utilitarian judgments by subdividing the high-RT personal moral ing a bilateral increase in amygdala activity for personal,

as compared to impersonal, moral judgment.judgments according to the participant’s behavior, i.e., by comparing “utilitarian” judgments (“appropriate”) to nonutilitarian judgments (“inappropriate”). In each of Analysis 1: Difficult versus Easy Personal

Moral Judgmentthese difficult dilemmas, an action that normally would be judged immoral (e.g., smothering a baby) is favored Preliminary to analysis 1, we compared the neural activ-

ity associated with difficult personal moral judgmentsby strong utilitarian considerations (e.g., saving many lives). The participants, in each instance, must decide if to that of a fixation baseline in ROIs generated by our

previous experiment comparing personal and imper-the utilitarian action is “appropriate” or “inappropriate.” Our hypothesis is that judgments of “appropriate” will sonal moral judgment (Greene et al., 2001). This was

done to ensure that the difficult personal moral dilem-be associated with greater DLPFC activity than those of “inappropriate,” reflecting the influence of “cognitive” mas focused on here (and not just the easy personal

Neuron 392

Table 1. Brain Regions Exhibiting Differential Activity for Personal versus Impersonal Moral Judgment

Max t Score Cluster Size Talairach Coordinates Regions Right/Left Brodmann’s Area (df � 40) (Voxels) (x, y, z)

Personal � Impersonal Medial prefrontal cortex R/L 9/10 10.3 513 0, 54, 19 Posterior cingulate/precuneus R/L 31 10.01 275 �4, �50, 36 Putamena, caudate nucleus a, L N/A, 21 7.59 320 �24, 14, �9 Middle temporal gyrusa, �2, �7, 8 Amygdalaa �1, �18, 8

�53, �7, �13 �26, �8, �15

Mid cingulate R/L 24 6.17 17 0, �20, 36 Middle temporal gyrus R 21 5.96 109 49, �4, �13 Amygdala R N/A 5.82 44 25, �4, �13 Anterior cingulate R/L 24 5.47 27 �2, 22, 16 Superior temporal sulcus R 39 5.11 233 46, �49, 19

L 39 5.09 211 �45, �58, 17 Lingual gyrus R 19 4.88 39 21, �65, �5

Impersonal � Personal Inferior parietal lobe R 40 11.44 505 25, �64, 35

L 40 10.02 386 �31, �61, 36 Inferior frontal gyrus R 44 6.52 81 43, 7, 25

L 44 6.48 77 �48, 6, 26 Posterior cingulate R/L 23/31 6.09 20 �4, �31, 29 Middle frontal gyrus R 46 5.97 101 39, 28, 25 Middle temporal gyrus R 21 5.32 23 43, �49, �5 Inferior temporal gyrus L 37 4.42 15 �40, �57, �5

Voxelwise significance threshold p � 0.0005; minimum cluster size 8 voxels. a Denotes distinct focus of activation within larger ROI.

moral dilemmas, which previously were not distin- (see Table 2 and Figure 2) and inferior parietal lobes (BA 40/39). Also as predicted, we found that difficult, asguished from difficult personal moral dilemmas) engage

the previously identified brain regions associated with compared to easy, personal moral dilemmas were asso- ciated with increased ACC activity (see Table 2 andemotion and social cognition. As predicted, we found

that each of the three brain regions previously exhibiting Figure 2). Finally, this contrast also revealed activity in the posterior cingulate cortex (BA 23/31).greater activity for personal, as compared to impersonal,

moral judgment (medial prefrontal cortex, BA 9/10; pos- terior cingulate/precuneus, BA 31/7; and bilateral supe- Analysis 2: Utilitarian Personal Moral Judgment

To test the hypothesis that utilitarian moral judgmentsrior temporal sulcus (STS)/inferior parietal lobe, BA 39) also exhibited above-baseline activity for difficult per- engage brain areas associated with “cognitive” pro-

cesses, we compared the neural activity associated withsonal moral judgments in the current study (p � 0.05, cluster size � 8 voxels). We note that this baseline com- utilitarian judgments (accepting a personal moral viola-

tion in favor of a greater good) to nonutilitarian judg-parison is a particularly strong test of the relevant re- gions’ engagement in our task because these regions ments (prohibiting a personal moral violation despite its

utilitarian value). We conducted a planned contrast us-are most often found to exhibit decreased neural activity relative to fixation baseline in other studies (Gusnard ing the ROIs generated by analysis 1 (high- versus low-

RT). Here we found increased activity for utilitarian, asand Raichle, 2001). To test the hypothesis that difficult, as compared to compared to nonutilitarian, moral judgment bilaterally

in the anterior DLPFC (BA 10) and in the right inferioreasy, personal moral dilemmas also engage brain areas associated with abstract reasoning, cognitive conflict, parietal lobe (BA 40) (see Table 3 and Figure 3). In addi-

tion, we found increased activity for utilitarian moraland cognitive control, we directly compared the neural activity associated with difficult and easy personal moral judgments in the more anterior region of the posterior

cingulate (BA 23/31) mentioned above. Other brain re-dilemmas. More specifically, we divided personal moral judgment trials into three categories based on individu- gions exhibit this effect as well (see Table 3). We note

that the utilitarian and nonutilitarian trials compared hereally normalized reaction time (see Experimental Proce- dures) and compared the neural activity associated with were matched for average RT (see Experimental Proce-

dures).the most difficult trials (upper third/high-RT) to the easi- est trials (lower third/low-RT). Mean RT for high- and As a supplemental exploration, we conducted a

whole-brain analysis making the same comparison. Aslow-RT trials were 8.38 and 2.83 s, respectively. Note that the same number of time points was compared for before, we observed increased activity for utilitarian

judgments in the right anterior DLPFC (BA 10), inferioreach condition (see Experimental Procedures). As predicted, we found that difficult, as compared parietal lobe (BA 40; this time on the left side), and

posterior cingulate (BA 23/31) (see Table 4 and Figureto easy, personal moral dilemmas involved increased activity bilaterally in both the anterior DLPFC (BA 10/46) 3). This DLPFC region is anterior to and contiguous with

Cognitive Conflict and Control in Moral Judgment 393

Figure 2. Difficult versus Easy Personal Moral Judgment

Selected brain regions (see Table 2) exhibiting significantly increased activity for difficult (high-RT), as compared to easy (low-RT), personal moral judgment: anterior cingulate cortex (BA 32), posterior cingulate cortex (BA 23/31), precuneus (BA 7), right and left middle frontal gyrus (BA 10/46). Statistical maps of voxelwise t scores were thresholded for significance (p � 0.0005) and cluster size (�8 voxels). (A) Sagittal slice plane is x � 0; (B) axial slice plane is z � �9 (Talairach and Tournoux, 1988). Image is reversed right to left according to radiologic convention.

the DLPFC region identified in the spatially restricted ing “cognitive” processes that drive people to approve of such violations in the relevant contexts. This hypothe-analysis described above. The same effect (utilitarian �

nonutilitarian) was observed in three locations within the sis stands in contrast to the hypothesis that difficult personal moral dilemmas elicit increased reaction timestemporal lobes: right superior temporal gyrus (BA 22/42),

right middle temporal gyrus (BA 21), and left inferior simply because they involve extended computation and not because of cognitive competition between incom-temporal gyrus (BA 19). patible behavioral responses. Evidence suggests that the ACC is responsive to processing conflict (BotvinickDiscussion et al., 2001; Carter et al., 1998), as in the Stroop task (Kerns et al., 2004; MacDonald et al., 2000). We thereforeAnalysis 1

In analysis 1, we tested two predictions concerning the predicted that difficult, as compared to easy, personal moral dilemmas would exhibit increased ACC activity,psychological processes engaged in moral judgment

and their neural implementation. First, according to our a prediction that was confirmed. Second, we hypothesized that the processes thattheory, difficult (high-RT) personal moral dilemmas such

as the crying baby dilemma involve a conflict between compete with social-emotional responses to difficult personal moral dilemmas are ones that rely on abstract(1) social-emotional responses that drive people to dis-

approve of personal moral violations and (2) countervail- reasoning and cognitive control. The anterior DLPFC is

Table 2. Brain Regions Exhibiting Differential Activity for Difficult versus Easy Personal Moral Judgment

Max t Score Cluster Size Talairach Coordinates Regions Right/Left Brodmann’s Area (df � 39) (Voxels) (x, y, z)

Difficult � Easy Anterior cingulate a R/L 32 8.82 443 0, 33, 25 Middle frontal gyrus a L 10/46 �28, 49, 7 Middle frontal gyrus a R 10 7.22 213 32, 47, 11 Anterior insula a/inferior frontal gyrus R N/A, 47 32, 17, �2 Posterior cingulate a R/L 23/31 7.15 380 �1, �27, 27 Precuneus a R/L 7/31 0, �69, 37 Inferior parietal lobe R 40/39 6.72 156 46, �54, 35 Inferior parietal lobe L 40/39 5.67 40 �42, �59, 35 Anterior insula L N/A 4.4 9 �37, 14, 0

Easy � Difficult Cuneus R/L 17/18/19 8.9 1314 1, �75, 8 Middle temporal gyrus R 21 6.99 87 50, �6, �7 Middle temporal gyrus a L 21 6.31 299 �55, �6, �7 Superior temporal sulcus a L 21/22 �58, �47, 8 Superior temporal gyrus R 22 6.22 67 46, �36, 8 Precentral gyrus L 4 5.13 18 �52, �9, 30 Red nucleus L N/A 4.92 49 �4, �23, �1 Caudate nucleus L N/A 4.55 14 �3, 4, 8 Inferior frontal gyrus L 44 4.15 8 �49, 5, 19

Voxelwise significance threshold p � .0005; min cluster size 8 voxels. a Denotes distinct focus of activation within larger ROI.

Neuron 394

Figure 3. Utilitarian versus Nonutilitarian Dif- ficult Personal Moral Judgment

Selected brain regions (see Tables 3–4) ex- hibiting significantly increased activity for utilitarian, as compared to nonutilitarian, diffi- cult personal moral judgment. Images are re- versed right to left according to radiologic convention. (A) A spatially restricted analysis (p � 0.05, cluster size � 8) of activity in the anterior dorsolateral prefrontal cortex (BA 10/ 46) revealed bilateral clusters of voxels exhib- iting increased activity during trials in which participants made utilitarian judgments. Axial slice plane is z � �8 (Talairach and Tournoux, 1988). (B) A whole-brain analysis (p � 0.005, cluster size � 8) revealed a contiguous and slightly anterior region on the right side exhib- iting the same effect (z � �13) . (C) Time course of activity in this region by participant response: utilitarian/“appropriate” (green) ver- sus nonutilitarian/“inappropriate” (red). Data are not adjusted for hemodynamic lag.

known for its role in these processes (Koechlin et al., tent with our hypothesis concerning “cognitive” pro- cesses. Activity in these regions has regularly been ob-2003; Miller and Cohen, 2001; Ramnani and Owen,

2004), and we therefore predicted that this region as served together with that of the DLPFC in tasks that engage working memory and other characteristicallywell would exhibit increased activity for difficult, as com-

pared to easy, personal moral dilemmas. This prediction “cognitive” processes (Wager and Smith, 2003). Second, we observed the same effect in an anterior region of thewas also confirmed. This result is particularly striking in

light of our previous finding that personal moral judg- posterior cingulate (BA 23/31). In the replication of our previous results (Table 1), this area exhibited relativelyment involved decreased activity in the DLPFC, as com-

pared to impersonal moral judgment (Greene et al., 2001). greater activation for impersonal, as compared to per- sonal, moral judgment. Thus, the activity in this regionThis analysis yielded several other results that deserve

attention. First, our finding that difficult, as compared appears once again to follow the “cognitive” pattern observed in the DLPFC and inferior parietal lobes, de-to easy, personal moral judgments involved increased

activity bilaterally in the inferior parietal lobes is consis- spite the fact that this region lies in the posterior cingu-

Table 3. Brain Regions Exhibiting Differential Activity for Utilitarian versus Nonutilitarian Personal Moral Judgment within Regions Exhibiting Differential Activity for Difficult versus Easy Personal Moral Judgment

Max t Score Cluster Size A Priori ROI Talairach Coordinates Regions Right/Left Brodmann’s Area (df � 38) (Voxels) Size (Voxels) (x, y, z)

Utilitarian � Nonutilitarian Posterior cingulate R/L 23/31 3.34 110 380a 0, �31, 32 Superior/middle frontal gyrus L 10 3.32 57 443a �22, 48, 8

R 10 2.76 62 213a 28, 49, 6 Precuneus L 7 3.19 11 380a �14, �67, 33

R 7 2.11 12 380a 7, �68, 42 Inferior parietal lobe R 40 3.01 40 156 36, �44, 36 Lingual gyrus R/L 18 2.57 15 1314 1, �67, 6

Voxelwise significance threshold p � 0.05, within a priori ROI, min cluster size 8 voxels. a Includes multiple foci of activation.

Cognitive Conflict and Control in Moral Judgment 395

Table 4. Brain Regions Exhibiting Differential Activity for Utilitarian versus Nonutilitarian Personal Moral Judgment

Max t Score Cluster Size Talairach Coordinates Regions Right/Left Brodmann’s Area (df � 38) (Voxels) (x, y, z)

Utilitarian � Nonutilitarian Inferior temporal gyrus L 19 3.93 68 �51, �69, �1 Middle temporal gyrus R 21 3.65 14 54, �57, 8 Superior frontal gyrus R 10 3.40 8 16, 56, 12 Posterior cingulate L/R 23/31 3.4 40 �5, �30, 32 Inferior parietal lobe L 40 3.4 21 �38, �45, 25 Superior temporal gyrus R 22/42 3.34 9 64, �31, 8

Voxelwise significance threshold p � .005; min cluster size 8 voxels.

late, a region previously associated with emotion (Mad- 2001), rather than attributing the observed increase in ACC activity to nonspecific processes related to timedock, 1999). (See also the discussion of analysis 2

below.) The remaining region exhibiting this effect, the on task or general difficulty. The conflict monitoring hy- pothesis, however, is controversial, and thus our inter-anterior insula, is also associated with emotion and,

more specifically, with disgust (Calder et al., 2001). Its pretation of these data may be questioned from the perspective of alternative theories of ACC function. Ac-activity has also been associated with risky decision

making in a gambling task (Paulus et al., 2003) and with cording to one theory, the ACC functions as an error detector (Coles et al., 1995). This theory, however, af-a tendency to reject unfair offers in the ultimatum game

(Sanfey et al., 2003). Thus, it seems that the insula sub- fords no clear interpretation of the present results be- cause, in the context of difficult moral judgment, it isserves negative affective states that bear on decision

making. In light of this, it is plausible that increased not clear what counts as an “error.” According to a different set of theories, the relevant function of the ACCinsula activity would be associated with difficult per-

sonal moral judgment specifically, but not with personal is regulative, subserving attention to (or selection for) action (e.g., Posner et al., 1988). The present data domoral judgment in general. In response to difficult per-

sonal moral dilemmas, people find themselves enter- not distinguish between such regulative theories and the conflict monitoring theory, as we have argued thattaining, and in some cases endorsing, actions that would

otherwise be considered morally repugnant. For exam- difficult personal moral dilemmas involve both increased conflict (ACC) and increased regulative control (DLPFC).ple, in the crying baby case, one is tempted to say that

it is acceptable to smother the baby in order to save Elsewhere we have attempted to dissociate these pro- cesses, localizing the detection of conflict to the ACCthe lives of the other people who are hiding. The consid-

eration of such repugnant acts in the context of difficult (Botvinick et al., 1999, 2001; MacDonald et al., 2000) and control to DLPFC (MacDonald et al., 2000; Kernsmoral dilemmas may elicit greater insula activity than

the consideration of similar acts in the context of easy et al., 2004). Finally, some have argued that increased ACC activity reflects the generation of autonomic statespersonal moral dilemmas, such as the infanticide case,

in which judgment is unanimous and swift and in which of cardiovascular arousal (Critchley et al., 2003), perhaps related to negative affective states. We are not ablethere appears to be little temptation to make a contro-

versial judgment. to assess this hypothesis directly because we did not acquire the necessary physiological data (heart rate vari-We note that dilemmas were classified as difficult

(high-RT) or easy (low-RT) on a subject-by-subject, per- ability, etc.). It is worth noting, however, that the conflict monitoring hypothesis may be consistent with the auto-trial basis. Thus, a dilemma that was easy for one person

could be difficult for another. Nevertheless, the RT re- nomic regulation hypothesis if, for example, conflict de- tection by the ACC is associated with negative affectsults for each dilemma were fairly consistent across

individuals, allowing us to refer to certain dilemmas as and signals the need for both cognitive and auto- nomic control.“difficult” or “easy.” The footbridge and infanticide di-

lemmas tended to be easy, while the crying baby di- lemma tended to be difficult. The trolley dilemma is Analysis 2

In analysis 2, we tested a second hypothesis concerningimpersonal and is therefore not involved in the present analysis (see Figure 1). the role of “cognitive processes” in moral judgment.

In addition to proposing that difficult personal moralSeveral alternative explanations concerning the pres- ent analysis also deserve attention. First, the compari- dilemmas involve increased reasoning and cognitive

control, we hypothesized that these “cognitive” pro-son between difficult and easy personal moral judg- ments is complicated by a potential confound of time on cesses have a preferred behavioral outcome, namely

that of favoring utilitarian moral judgments, at least intask: more difficult trials are defined as those associated with longer RT. However, longer RT could also reflect the the context of the difficult personal moral dilemmas em-

ployed in this study.prolonged engagement of other, nonspecific processes, such as visual processing and/or motor responding. We These dilemmas share a common structure: a per-

sonal moral violation is required to achieve a greateraddress this concern in our discussion of analysis 2 below. good, as in the crying baby case. These difficult cases

contrast with easy personal moral dilemmas such asSimilarly, our interpretation draws on the conflict mon- itoring hypothesis of ACC function (Botvinick et al., the infanticide case in which personal moral violations

Neuron 396

are proposed but in which the benefits sought are rela- suggests that the effects in these three regions were tively small compared to those available in the difficult related to processes occurring after the point of deci- cases. Thus, it is natural to suppose that a utilitarian, sion. These may have been related to participants’ reac- cost-benefit analysis is most often the basis for judging tions to their decisions, which may be more salient when personal moral violations to be appropriate in the diffi- the participant has recently approved of a personal cult cases. moral violation (e.g., smothering a baby). In support of

Reaching an overt judgment on utilitarian grounds this suggestion, we note that overlapping regions in the has two processing requirements. First, the abstract temporal lobes have been associated with the percep- reasoning that constitutes a utilitarian analysis must be tion of socially significant actions (Allison et al., 2000). conducted. Second, cognitive control must be engaged The significance of the effects observed in these regions to support successful competition of the behavior fa- is a matter for further research. vored by the outcome of that analysis against any in- The results of analysis 2 help to resolve a potential compatible behavioral pressures (e.g., an emotional re- concern about the results of analysis 1. As noted above, sponse favoring the opposite behavior). Thus, we might the comparison made in analysis 1 between difficult and expect to see neural activity associated with both of easy personal moral dilemmas, based on differences in these demands in the results of analysis 1. That is, diffi- RT, is subject to a confound of time on task. Analysis cult personal moral dilemmas, as compared to easy 2 used the brain areas identified in analysis 1 as a priori ones, will involve both utilitarian reasoning and (in many regions of interest in a comparison in which RT was cases) the application of cognitive control in favoring controlled. Therefore, the results of analysis 2 suggest the utilitarian response over its competitors. However, that the effects related to cognitive control that were the results of analysis 2 are expected to be more restric- predicted and observed in analysis 1 were not merely tive since they examined only difficult dilemmas, com- due to increased time on task and are consistent with our paring activity associated with utilitarian versus nonutili- theory concerning the deployment of cognitive control in tarian responses. Since difficult dilemmas were likely to responding to difficult moral dilemmas. have engaged abstract reasoning, irrespective of behav- ioral response (as evidenced by similar RTs for both Conflict and Cognitive Control types of trial), areas of activity associated with abstract We hypothesized and found evidence for the engage- reasoning are likely to have been “subtracted out” by ment of brain areas associated with the detection of analysis 2. Thus, voxels showing significantly greater conflict (ACC) and the deployment of cognitive control activity associated with a utilitarian response reflect pri- (DLPFC) in responding to difficult moral dilemmas. In marily the successful engagement of cognitive control previous work, we have proposed that a cardinal func- in support of that response. In summary, analysis 1 iden- tion of conflict detection by ACC is the recruitment of tified areas within DLPFC associated with the demands control mechanisms in PFC needed to resolve the con- for abstract reasoning as well as cognitive control, while flict (Botvinick et al., 2001). Support for this proposal analysis 2 identified a subset of these regions that we has come from studies in which behavioral evidence of presume was associated more specifically with the exe- conflict and corresponding ACC activity on one trial is cution of control. The latter finding is consistent with followed in the subsequent trial by improved perfor- other findings suggesting that the DLPFC (bilateral BA

mance (Botvinick et al., 1999) and a corresponding in- 46) plays an important role in the regulation of potentially

crease in PFC activity (Kerns et al., 2004). In the present counterproductive emotions in the context of social de-

study, however, difficult decisions were associated with cision making (Sanfey et al., 2003), in the context of the

increased ACC and DLPFC activity in the same trial. placebo effect (Wager et al., 2004), and in the evaluation

One interpretation of this finding is that the conflict asso- of tradeoffs between future and immediate rewards (Mc-

ciated with a difficult moral decision was detected by Clure et al., 2004). Our finding that different areas of

the ACC, which then recruited control mechanisms in PFC seem to have been differentially sensitive to the

DLPFC to help resolve conflict within the same trial. Thisengagement of reasoning and control is intriguing, sug- is plausible given the latency of behavioral responsesgesting that the neural mechanisms subserving these in this task (seconds) relative to simpler tasks involvingaspects of cognitive processing may be at least par- speeded responses (under 1 s) in which adjustments oftially dissociable. control are typically observed across trials. However,The same effect was observed bilaterally in the inferior another possibility is that it was the engagement of con-parietal lobes (BA 40), consistent with the common find- trol in the support of utilitarian responses that produceding of activity in these areas in tasks engaging cognitive the conflict associated with difficult decisions. That is,control (Wager and Smith, 2003). This effect was also the recruitment of control reflected in DLPFC activityobserved in the posterior cingulate region (BA 23/31) allowed the utilitarian “cognitive” response to competethat, as described above, exhibited increased activity for more effectively with the otherwise prepotent emotionaldifficult personal moral dilemmas as well as increased response, generating the conflict reflected in ACC activ-activity for impersonal, as compared to personal, moral ity. Adjudicating between these alternatives will requirejudgment. The activity in this region, which is more often greater temporal resolution than our methods provided.associated with emotion (Maddock, 1999), mirrors that

of the characteristically “cognitive” brain regions in the “Cognition” and Emotion in Moral PsychologyDLPFC and the inferior parietal lobes. Finally, in the whole- For decades, moral psychology was dominated by ratio-brain version of this analysis, this effect (utilitarian � non- nalist models according to which moral developmentutilitarian) was observed in three regions within the tempo-

ral lobes. An examination of the time courses of activity consisted of the use of increasingly sophisticated

Cognitive Conflict and Control in Moral Judgment 397

modes of abstract moral reasoning (Kohlberg, 1969). roles played by both emotion and “cognition” (Greene More recently, affective processes have taken center and Haidt, 2002). Our previous neuroimaging and behav- stage. According to Haidt’s social-intuitionist model ioral results suggest that emotional responses drive indi- (Haidt, 2001), moral judgment is driven primarily by viduals to disapprove of personal moral violations. Our rapid, affectively based, intuitive responses, with delib- present finding that increased “cognitive” activity in the erate moral reasoning engaged after the fact to provide DLPFC predicts utilitarian moral judgment behavior sug- rational justifications in response to social demands. gests that cognitive control processes can override The present data support a synthetic theory of moral these emotional responses, favoring personal moral vio- judgment according to which both of these viewpoints lations when the benefits sufficiently outweigh the costs. reflect important aspects of the truth (Greene and Haidt, Thus, both emotional and “cognitive” processes appear 2002). Our earlier work provided initial support for this to be crucial in producing the patterns of neural activity theory through our finding that personal moral judgment and behavior observed in these experiments. This con- involves relatively greater activity in brain areas associ- clusion is consistent with a growing body of literature ated with social-emotional processing, while impersonal concerning the respective roles of intuition and delibera- moral judgment involves relatively greater activity in tion in judgment and decision making (Kahneman, 2003). brain areas associated with characteristically “cogni- tive” processes such as working memory, abstract rea- The Relationship between “Cognition” soning, and problem solving. These neuroscientific find- and Emotion ings were complemented by RT data suggesting that The account we’ve offered is complicated by the fact some moral dilemmas elicit response conflict between that brain regions other than the DLPFC and inferior negative emotional responses and countervailing pro- parietal lobes predict utilitarian moral judgment. One of cesses, which we hypothesized to be “cognitive” in na- these regions is in the posterior cingulate (BA 23/31), ture. The present data provide further support for this which has been associated with emotion (Maddock, theory in two ways. First, they corroborate and extend 1999). This finding does not necessarily undermine our our earlier findings in a much larger sample size, provid- suggestion that “cognitive” processing tends to favor ing further support for our claims concerning the roles utilitarian judgment in response to the dilemmas em- of emotion and “cognition” in moral judgment. Second, ployed here. It does, however, challenge the overly sim- and more importantly, these data reveal that neural ac- ple view that utilitarian judgments are wholly allied with tivity in classically “cognitive” brain regions predicts a “cognition” while nonutilitarian judgments are wholly al- particular type of moral judgment behavior, thus provid- lied with “emotion.” Like David Hume (Hume, 1978), we ing strong support for the view that both “cognitive”

suspect that all action, whether driven by “cognitive” and emotional processes play crucial and sometimes

judgment or not, must have some affective basis. Even mutually competitive roles.

a cold, calculating utilitarian must be independently mo- In analysis 1, we tested and confirmed the prediction

tivated, first, to engage in the reasoning that utilitarian that brain regions involved in mediating response con-

judgment requires and, second, to respond in accor- flict (ACC) and the implementation of cognitive control

dance with such judgment. The ACC, a limbic region (DLPFC) exhibit increased activity during difficult, as

believed to recruit cognitive control (Botvinick et al., compared to easy, personal moral judgment. These find-

2001), is well suited to play the first of these motivational ings support the Kohlbergian claim that high-level cogni-

roles. We tentatively suggest that the region identified tive processes are marshaled in the resolution of difficult

in BA 23/31 of the posterior cingulate may play the sec- moral dilemmas and stand in tension with the social

ond of these roles. This area was engaged under condi-intuitionist claim that in nearly all cases moral judgments tions and in a manner that closely parallels other areasare more akin to perception than episodes of reasoning (in the DLPFC and parietal cortex) that have been consis-or reflection (Haidt, 2001). Likewise, the RT data raise tently associated with nonemotional processing. Thus,doubts about moral judgment as unreflective, as our it is possible that this brain area is involved in mediatingparticipants routinely exhibited RTs over 10 s, and in the interaction between purely “cognitive” processessome cases over 20 s, despite the fact that they were and the affective/motivational processes necessary fornot required to justify their answers at any point. The producing behavior. This interpretation draws conver-engagement of brain areas commonly associated with gent evidence from a recent study of spatial attentiondeliberative thought processes strengthens this view. (Small et al., 2003).In analysis 2, we tested and confirmed the prediction

Throughout this article, we have relied on a familiarthat utilitarian judgment, as compared to nonutilitarian distinction between “emotion” or “affect” on the onejudgment, involves increased activity in brain regions hand and “cognition” on the other. This distinction hasassociated with cognitive control, particularly in the proven useful, and yet it may be somewhat artificial. TheDLPFC. This finding challenges both rationalist (Kohl- term “cognition” is often defined in terms of “informationberg, 1969) and emotivist (Haidt, 2001) theories of moral processing,” but all of the processes considered here,psychology because both types of theory regard internal including those that we have labeled “emotional,” in-moral conflicts, insofar as they are thought to exist, volve information processing, thus calling into questionas conflicts between processes of the same general the usefulness of this definition of “cognition.” Alterna-type—rational processes in the case of rationalist mod- tively, one might render the emotion/cognition distinc-els and emotional processes in the case of emotivist tion in terms of a contrast between, on the one hand,models. In contrast, the results presented here and pre- representations that have direct motivational force and,viously (Greene et al., 2001) together suggest a synthetic

view of moral judgment that acknowledges the crucial on the other hand, representations that have no direct

Neuron 398

consent was obtained for each participant. In addition to the datamotivational force of their own, but that can be contin- drawn from these 41 participants, data from three participants weregently connected to affective/emotional states that do discarded for technical reasons related to the scanner, and data

have such force, thus producing behavior that is both from one participant were discarded due to highly abnormal behav-

flexible and goal directed. According to this view, the ioral responses. One participant’s data were discarded from analy- emotion/cognition distinction is real, but it is a matter ses 1 and 2 and another participant’s data were discarded from

analysis 2 due to unbalanced factors (which depend on the partici-of degree and, at the present time, not very well under- pants RTs and decision outcomes).stood. It is within a framework of this sort that we retain

and utilize the emotion/cognition distinction while rec- Stimuliognizing that this distinction is far from clear cut. We employed a battery of 60 practical dilemmas, available online at http://www.neuron.org/cgi/content/full/44/2/389/DC1/. (See

Broader Implications materials for “Experiment 2”.) These dilemmas were divided into “moral” and “nonmoral” categories based on the responses of pilotFor two centuries, Western moral philosophy has been participants. Data concerning nonmoral dilemmas were not ana-defined largely by a tension between two opposing view- lyzed for present purposes. Two independent coders evaluated eachpoints. Utilitarians (or, more broadly, “consequentialists”) moral dilemma using three criteria designed to capture the differ-

such as John Stuart Mill (Mill, 1998) argue that morality ence between the intuitively “up close and personal” (and putatively

is, or ought to be, a matter of promoting the “greater more emotional) sort of violation exhibited by the footbridge di- good,” while “deontologists” such as Immanuel Kant lemma and the more intuitively impersonal (and putatively less emo-

tional) violation exhibited by the trolley dilemma. First, coders indi-(Kant, 1959) argue that certain moral lines ought not be cated for each dilemma whether or not the action in question couldcrossed, that certain rights or duties must be respected, “reasonably be expected to lead to serious bodily harm.” Second,regardless of the greater good that might otherwise be they were asked to indicate whether or not this harm would be “the

achieved. Moral dilemmas of the sort employed here result of deflecting an existing threat onto a different party.” Our

boil this philosophical tension down to its essentials and use of this criterion, which parallels a distinction made by Thomson may help us understand its persistence. We propose (1986), is an attempt to operationalize an intuitive notion of “agency.”

Intuitively, when a harm is produced by means of deflecting anthat the tension between the utilitarian and deontologi- existing threat, the agent has merely “edited” and not “authored”cal perspectives in moral philosophy reflects a more the resulting harm. Finally, coders were asked to indicate whetherfundamental tension arising from the structure of the or not the resulting harm would “befall a particular person or a

human brain. The social-emotional responses that we’ve member or members of a particular group of people.” Here the

inherited from our primate ancestors (due, presumably, question, in intuitive terms, is whether the victim(s) is/are “on stage” to some adaptive advantage they conferred), shaped in the dilemma. The moral dilemmas of which the coders said that

the action in question (1) could reasonably be expected to lead toand refined by culture bound experience, undergird the serious bodily harm (2) to a particular person or a member or mem-absolute prohibitions that are central to deontology. In bers of a particular group of people (3) where this harm is not thecontrast, the “moral calculus” that defines utilitarianism result of deflecting an existing threat onto a different party were

is made possible by more recently evolved structures assigned to the personal moral judgment condition, the others to

in the frontal lobes that support abstract thinking and the impersonal condition. high-level cognitive control. We have adduced some For analysis 1, RTs were normalized by reexpressing each RT as

a percentage of the participant’s average RT for personal trials.support for this hypothesis, albeit using a limited set Personal trials were ordered by normalized RT and split evenly intoof testing materials. First, we have seen evidence of high-, medium-, and low-RT categories. Medium-RT dilemmas wereincreased social-emotional processing in cases in which discarded from analysis. For analysis 2, a two-tailed Student’s t test

deontological intuitions are prominent. Second, we have was performed to ensure that the utilitarian and nonutilitarian trial

seen greater activity in brain regions associated with groups were matched for average RT (p � 0.6). cognitive control when utilitarian judgments prevail. These brain regions house some of our species’ most Experimental Design recently evolved neural features (Allman et al., 2002) and Dilemmas were presented on a visual display projected into the

scanner in a series of 12 blocks of five trials each. Each dilemmaassociated cognitive abilities. was presented as text through a series of three screens, the firstWe emphasize that this cognitive account of the Kant two describing a scenario and the last posing a question about theversus Mill problem in ethics is speculative. Should this appropriateness of an action one might perform in that scenario

account prove correct, however, it will have the ironic (e.g., turning the trolley). Subjects read at their own pace, pressing implication that the Kantian, “rationalist” approach to a button to advance from the first to the second screen and from the moral philosophy is, psychologically speaking, grounded second to the third screen. After reading the third screen, subjects

responded by pressing one of two buttons (“appropriate” or “inap-not in principles of pure practical reason, but in a set of propriate”). Subjects were given a maximum of 46 s to read all threeemotional responses that are subsequently rationalized screens and respond. The intertrial interval (ITI) lasted for a minimum(Haidt, 2001). Whether this psychological thesis has any of 14 s (seven images) in each trial, allowing the hemodynamic

normative implications is a complicated matter that we response to return to baseline after each trial. During the ITI, parti- leave for treatment elsewhere (Greene, 2003; Greene cipants viewed a fixation cross. Stimuli were presented and behav- and Cohen, 2004). ioral responses were collected using PsyScope stimulus presenta-

tion software (Cohen et al., 1993), the PsyScope button box (http:// psyscope.psy.cmu.edu/bbox/), and a custom response handsetExperimental Procedures from Psychology Software Tools (http://www.pstnet.com/).

Participants Our participants were 41 healthy adult undergraduates (24 males, fMRI Acquisition

Images were acquired using a 3.0 T Siemens Allegra head-dedicated17 females). All participants spoke native English, were right handed, and were screened for a history of psychiatric and neurological scanner. A high-resolution (1 mm voxel MPRAGE) whole-brain struc-

tural scan was acquired prior to the functional imaging. Functionalproblems. All experimental procedures complied with the guidelines of Princeton University’s Internal Review Panel. Written informed images were acquired in 22 axial slices parallel to the AC-PC line

Cognitive Conflict and Control in Moral Judgment 399

using an EPI pulse sequence, with a TR of 2000 ms, a TE of 25 ms, of automatic processes: A parallel distributed processing account of the stroop effect. Psychol. Rev. 97, 332–361.a flip angle of 90, a FOV of 192 mm, 3.0 mm isotropic voxels, and

1 mm interslice spacing. Cohen, J.D., MacWhinney, B., Flatt, M., and Provost, J. (1993). Psy- scope: A new graphic interactive environment for designing psychol-

fMRI Preprocessing and Analysis ogy experiments. Behavioral Research Methods, Instruments, and Prior to statistical analysis, images for all participants were motion Computers 25, 257–271. corrected and coregistered using AIR (Woods et al., 1992). Statistical

Coles, M.G., Scheffers, M.K., and Fournier, L. (1995). Where did you analyses and related preprocessing were performed using the NIS

go wrong? Errors, partial errors, and the nature of human information software package (http://kraepelin.wpic.pitt.edu/nis/). Images were

processing. Acta Psychol. (Amst.) 90, 129–144. spatially smoothed with an 8 mm FWHM 3D Gaussian filter. Linear

Critchley, H.D., Mathias, C.J., Josephs, O., O’Doherty, J., Zanini, S.,trends were removed from each voxel’s time series for each run, Dewar, B.K., Cipolotti, L., Shallice, T., and Dolan, R.J. (2003). Humanand outliers beyond three standard deviations of the mean were cingulate cortex and autonomic control: Converging neuroimagingcorrected. Task-related activity was measured using a floating win- and clinical evidence. Brain 126, 2139–2152.dow of eight images surrounding (four prior to, one during, and three

following) the time of response. Thus, the entire window was 16 s Damasio, A.R. (1994). Descartes’ Error: Emotion, Reason, and the long. This window included three postresponse images in order to Human Brain (New York: G.P. Putnam). allow for the lag in the hemodynamic response (typically peaking

de Waal, F. (1996). Good Natured: The Origins of Right and Wrong 4–5 s following an eliciting neural response). For each participant,

in Humans and Other Animals (Cambridge, MA: Harvard Univer- the mean BOLD signal in each condition was averaged across the

sity Press). eight images in the response window; these means were then con-

Devine, P.G. (1989). Stereotypes and prejudice: Their automatic andtrasted across subjects with a pairwise t test. controlled components. J. Pers. Soc. Psychol. 56, 5–18.Whole-brain exploratory analyses were performed with a voxel-

wise significance threshold of p � 0.0005, while analyses focusing Forman, S.D., Cohen, J.D., Fitzgerald, M., Eddy, W.F., Mintun, M.A., on a restricted number of a priori ROI voxels were performed with and Noll, D.C. (1995). Improved assessment of significant activation a reduced voxelwise significance threshold of p � 0.05. All analyses in functional magnetic resonance imaging (fmri): Use of a cluster- used a cluster size threshold (8 voxels) to reduce type 1 error from size threshold. Magn. Reson. Med. 33, 636–647. multiple comparisons (Forman et al., 1995). The analysis preliminary Greene, J. (2003). From neural ‘is’ to moral ‘ought’: What are the to analysis 1 was restricted to voxels contained within a priori ROIs moral implications of neuroscientific moral psychology? Nat. Rev. generated by experiment 1 from Greene et al. (2001). For the spatially Neurosci. 4, 846–849. restricted version of analysis 2, an ROI mask generated by analysis

Greene, J., and Cohen, J. (2004). For the law, neuroscience changes 1 (high-RT versus low-RT) was used to delimit a priori ROIs. The

nothing and everything. Philos. Trans. R. Soc. Lond. B Biol. Sci., subsequent exploratory whole-brain analysis was performed at a

in press. lower statistical threshold (p � 0.005) after an initial analysis at our

Greene, J., and Haidt, J. (2002). How (and where) does moral judg-standard threshold (p � 0.0005) failed to identify any significant ment work? Trends Cogn. Sci. 6, 517–523.voxels. For generation of the event-related average graph (Figure

3C), baseline activity was defined as the mean signal across the Greene, J.D., Sommerville, R.B., Nystrom, L.E., Darley, J.M., and first two images of the trial, reflecting activity during the prior ITI. Cohen, J.D. (2001). An fmri investigation of emotional engagement

in moral judgment. Science 293, 2105–2108. Acknowledgments

Gusnard, D.A., and Raichle, M.E. (2001). Searching for a baseline: Functional imaging and the resting human brain. Nat. Rev. Neurosci.

This research was supported by a fellowship from the National Insti- 2, 685–694.

tutes of Health awarded to J.D.G. (MH067410) and by grants from Haidt, J. (2001). The emotional dog and its rational tail: A socialthe National Science Foundation awarded to J.D.C, J.M.D., and intuitionist approach to moral judgment. Psychol. Rev. 108, 814–834.J.D.G. (BCS-0351996) and to Paul Kantor and Stephen J. Hanson

(CCF-0205178). Many thanks to Jason Chein and Sam McClure for Hume, D. (1978). A Treatise of Human Nature, L.A. Selby-Bigge and helpful suggestions. P.H. Nidditch, eds. (Oxford: Oxford University Press).

Kahneman, D. (2003). A perspective on judgment and choice: Map- Received: June 3, 2004 ping bounded rationality. Am. Psychol. 58, 697–720. Revised: September 13, 2004

Kant, I. (1959). Foundation of the Metaphysics of Morals (Indianapo- Accepted: September 27, 2004

lis, IN: Bobbs-Merrill). Published: October 13, 2004

Kerns, J.G., Cohen, J.D., MacDonald, A.W., 3rd, Cho, R.Y., Stenger, V.A., and Carter, C.S. (2004). Anterior cingulate conflict monitoringReferences and adjustments in control. Science 303, 1023–1026.

Allison, T., Puce, A., and McCarthy, G. (2000). Social perception from Koechlin, E., Ody, C., and Kouneiher, F. (2003). The architecture of visual cues: Role of the sts region. Trends Cogn. Sci. 4, 267–278. cognitive control in the human prefrontal cortex. Science 302, 1181–

1185.Allman, J., Hakeem, A., and Watson, K. (2002). Two phylogenetic specializations in the human brain. Neuroscientist 8, 335–346. Kohlberg, L. (1969). Stage and sequence: The cognitive-develop-

mental approach to socialization. In Handbook of Socialization The-Bargh, J.A., and Chartrand, T.L. (1999). The unberable automaticity ory and Research, D.A. Goslin, ed. (Chicago: Rand McNally), pp.of being. Am. Psychol. 54, 462–479. 347–480.Botvinick, M., Nystrom, L.E., Fissell, K., Carter, C.S., and Cohen, J.D.

(1999). Conflict monitoring versus selection-for-action in anterior MacDonald, A.W., 3rd, Cohen, J.D., Stenger, V.A., and Carter, C.S. cingulate cortex. Nature 402, 179–181. (2000). Dissociating the role of the dorsolateral prefrontal and ante-

rior cingulate cortex in cognitive control. Science 288, 1835–1838.Botvinick, M.M., Braver, T.S., Barch, D.M., Carter, C.S., and Cohen, J.D. (2001). Conflict monitoring and cognitive control. Psychol. Rev. MacLeod, C.M. (1991). Half a century of research on the stroop 108, 624–652. effect: An integrative review. Psychol. Bull. 109, 163–203. Calder, A.J., Lawrence, A.D., and Young, A.W. (2001). Neuropsychol- Maddock, R.J. (1999). The retrosplenial cortex and emotion: New ogy of fear and loathing. Nat. Rev. Neurosci. 2, 352–363. insights from functional neuroimaging of the human brain. Trends

Neurosci. 22, 310–316.Carter, C.S., Braver, T.S., Barch, D.M., Botvinick, M.M., Noll, D., and Cohen, J.D. (1998). Anterior cingulate cortex, error detection, and McClure, S.M., Laibson, D.I., Loewenstein, G., and Cohen, J.D. the online monitoring of performance. Science 280, 747–749. (2004). Separate neural systems value immediate and delayed mon-

etary rewards. Science, in press.Cohen, J.D., Dunbar, K., and McClelland, J.L. (1990). On the control

Neuron 400

Mill, J.S. (1998). Utilitarianism, R. Crisp, ed. (New York: Oxford Uni- versity Press).

Miller, E.K., and Cohen, J.D. (2001). An integrative theory of prefron- tal cortex function. Annu. Rev. Neurosci. 24, 167–202.

Paulus, M.P., Rogalsky, C., Simmons, A., Feinstein, J.S., and Stein, M.B. (2003). Increased activation in the right insula during risk-taking decision making is related to harm avoidance and neuroticism. Neu- roimage 19, 1439–1448.

Posner, M.I., and Snyder, C.R.R. (1975). Attention and cognitive control. In Information Processing and Cognition, R.L. Solso, ed. (Hillsdale, NJ: Erlbaum), pp. 55–85.

Posner, M.I., Petersen, S.E., Fox, P.T., and Raichle, M.E. (1988). Localization of operations in the human brain. Science 240, 1627– 1631.

Ramnani, N., and Owen, A.M. (2004). Anterior prefrontal cortex: In- sights into function from anatomy and neuroimaging. Nat. Rev. Neu- rosci. 5, 184–194.

Rozin, P., Lowery, L., Imada, S., and Haidt, J. (1999). The cad triad hypothesis: A mapping between three moral emotions (contempt, anger, disgust) and three moral codes (community, autonomy, divin- ity). J. Pers. Soc. Psychol. 76, 574–586.

Sanfey, A.G., Rilling, J.K., Aronson, J.A., Nystrom, L.E., and Cohen, J.D. (2003). The neural basis of economic decision-making in the ultimatum game. Science 300, 1755–1758.

Shiffrin, R.M., and Schneider, W. (1977). Controlled and automaitc information processing: Ii. Perceptual learning, automatic attending, and a general theory. Psychol. Rev. 84, 127–190.

Small, D.M., Gitelman, D.R., Gregory, M.D., Nobre, A.C., Parrish, T.B., and Mesulam, M.M. (2003). The posterior cingulate and medial prefrontal cortex mediate the anticipatory allocation of spatial atten- tion. Neuroimage 18, 633–641.

Stroop, J.R. (1935). Studies of interference in serial verbal reactions. J. Exp. Psychol. 12, 643–662.

Talairach, J., and Tournoux, P. (1988). A Co-Planar Stereotaxic Atlas of the Human Brain (New York: Thieme).

Thomson, J.J. (1986). Rights, Restitution, and Risk: Essays, in Moral Theory (Cambridge, MA: Harvard University Press).

Wager, T.D., and Smith, E.E. (2003). Neuroimaging studies of work- ing memory: A meta-analysis. Cogn. Affect. Behav. Neurosci. 3, 255–274.

Wager, T.D., Rilling, J.K., Smith, E.E., Sokolik, A., Casey, K.L., David- son, R.J., Kosslyn, S.M., Rose, R.M., and Cohen, J.D. (2004). Pla- cebo-induced changes in fmri in the anticipation and experience of pain. Science 303, 1162–1167.

Woods, R.P., Cherry, S.R., and Mazziotta, J.C. (1992). Rapid auto- mated algorithm for aligning and reslicing pet images. J. Comput. Assist. Tomogr. 16, 620-633.