40 The Futures of Others

Prospective Social Control, Moral Learning, and Prosocial Choice

40.1 Another agent is not another outcome

Chapter 39 ended with an extension of prospective control beyond the chooser’s own body. Another person’s anticipated pain, relief, fear, or disappointment can alter present affect and action before that person has experienced anything. The extension is real, but it changes the problem.

Consider a traveler who reaches a water cache left for a partner expected hours later. Taking the water would satisfy an immediate need. Leaving it requires control by a state that is absent twice over: the partner is not present, and the partner’s future thirst has not yet occurred. The relevant prediction includes more than a bodily consequence. It includes who the partner is, what the partner will know, what was promised, how much water remains elsewhere, how the partner is likely to respond, and what taking the cache would do to a continuing relationship.

An ordinary nonsocial outcome does not interpret the action taken toward it. Another agent does. The other person has goals, information, memories, and policies of their own. They can misunderstand, retaliate, forgive, reciprocate, warn others, or revise what they expect from the actor. Social action therefore enters a coupled control system. Each participant acts partly from a model of the other, and each action changes the evidence available to both.

Several routes allow another person’s future to influence present choice. Anticipated suffering can evoke vicarious affect. Concern for the person’s welfare can make their outcome worth improving. An estimate of what they know or intend can change the meaning of the same physical result. A promise, role, or norm can govern action even when no distress is visible. Expected gratitude, punishment, retaliation, or damage to reputation can alter the actor’s own future. Experiments that measure empathy and reasoning about mental states within the same task find separable behavioral and neural contributions, confirming that feeling with another person and understanding another person’s mind are related but distinct achievements [@kanskeetal2015empatorm].

The raw materials of this control problem did not begin with human moral philosophy. Social modulation of pain has been demonstrated in mice, especially among familiar animals: the state of one animal changes the response of another [@langfordetal2006miceempathy]. That is not human conscience. It shows that another organism’s condition can become a biologically effective variable without language or explicit rules. Human development adds perspective, causal explanation, remembered relationships, shared norms, promises, institutions, and concern for people who may be distant or entirely unknown.

Neuroscience can explain how these variables gain control over action. It cannot determine which act is morally right, how much blame a person deserves, or which principles a society should adopt. Those are normative judgments. The scientific problem is narrower and still formidable: how does a controller become governed by consequences that belong to another autonomous controller?

Another person’s possible future can govern present action through vicarious affect, inferred goals and beliefs, learned relationships, norms and commitments, expected reciprocity or sanction, and the anticipated consequences of one’s own conduct. Orbital and ventromedial systems help integrate these variables with value and action, but conscience is not one signal and morality is not one cortical function.

40.2 Distress, empathy, concern, norm knowledge, and control are not synonyms

The word empathy often absorbs several operations that need to be separated. A person can identify fear without sharing it, share distress without understanding its cause, understand another person’s belief without caring about the result, or know a social rule without allowing it to govern behavior. These dissociations are not exceptions around the edge of one faculty. They reveal the component processes from which social control is assembled.

Distress detection is the recognition of signals such as a cry, fearful posture, pain expression, withdrawal, or submission. Detection can redirect attention and prepare action without producing a matching conscious feeling. It also depends on sampling: a signal cannot influence the observer if gaze and attention never reach the informative feature.

Affective empathy is a change in the observer’s own affective state that is related to another person’s state. It need not be an exact copy. Another person’s pain may produce aversive arousal, protective readiness, anger at an aggressor, or a quieter state of concern. The observer’s response is constructed in the observer’s own body and brain, under the constraints of context and relationship.

Personal distress is a self-focused aversive response to another person’s condition. It can motivate helping, but it can also motivate escape from the scene. Empathic concern is other-oriented: the other person’s welfare becomes an outcome the observer is prepared to improve. The two can occur together, but they predict different actions when helping is costly or withdrawal is possible.

Perspective taking, or mentalizing, is the inference of what another agent sees, knows, believes, intends, or expects. This contribution is essential because identical outcomes can have different social meanings. A broken object produced accidentally is not equivalent to the same damage produced deliberately. A false statement spoken in honest error is not equivalent to a deception. Moral and social control often depend less on matching another person’s feeling than on estimating the hidden state that organized the other person’s action.

Norm knowledge represents what a group, relationship, or institution permits, requires, or condemns. Norms can be learned through emotional consequences, observation, reinforcement, explicit instruction, narrative, and language. They allow control by events never directly witnessed: a promise can remain binding after the person to whom it was made has left the room.

Moral judgment is an evaluation of permissibility, obligation, blame, or punishment. Moral control is the further achievement of allowing relevant welfare, intention, relationship, and norm information to reorganize present action. The distinction parallels the one developed in Chapter 38 and Chapter 39. Information may be available for report yet fail to become an effective control state.

Empathy is not one thing

Affective empathy is a change in one’s own affective state related to another person’s state.

Empathic concern is an other-oriented motivation to improve another person’s welfare.

Perspective taking is an estimate of what another person knows, believes, intends, or experiences.

Personal distress is aversive arousal centered on one’s own discomfort.

Social and moral control additionally use relationships, norms, commitments, expected responses, and regulation of one’s own action.

These processes interact, but none substitutes for all the others.

The ordinary word conscience can remain useful as a name for the experienced capacity to be governed by social and moral considerations. It should not be reified as one internal alarm. A complete account must explain how several kinds of information become effective together.

40.3 Distress can teach: Blair’s model

James Blair’s developmental account gave one part of this problem a clear mechanism. The original violence-inhibition mechanism began from a comparative observation: in many social species, distress or submission signals can interrupt aggression. A cry, fearful expression, or submissive posture is not merely information about the victim. It changes the aggressor’s current state and reduces the probability that attack will continue [@blair1995morality].

In human development, distress cues can also act as social reinforcers. A harmful action is followed by crying, fear, withdrawal, parental intervention, or loss of affiliation. Through repeated experience, perceptual features of distress and the actions that caused it acquire learned affective significance. The child does not have to calculate every future consequence from first principles. Contemplating a similar action can retrieve its likely social outcome and reduce the action’s current value before harm occurs.

This account places the amygdala in learning the significance of distress and other socially important cues. Orbital and ventromedial prefrontal territories relate expected outcomes to the current situation and to the action under consideration. Striatal systems update action policies from reinforcement. Insular, hypothalamic, and brainstem systems contribute affective and bodily preparation. The useful mechanism is therefore not a signal sent from an amygdala distress detector to a ventromedial conscience. It is recurrent learning in which another person’s response becomes part of the predicted consequence of one’s own action [@blair2007amygdalavmpfc].

Blair’s model generated a particularly sharp behavioral prediction. Young children ordinarily distinguish moral violations, such as hitting another child, from conventional violations, such as breaking a classroom rule about clothing or seating. Harm-based violations are often judged wrong even when an authority suspends the rule; conventional violations depend more visibly on the rule and setting. If victim distress is a major source of the special weight assigned to harmful acts, then people with severe deficits in distress-based learning should make less of this distinction. The original study found the predicted pattern in adults with high psychopathy scores [@blair1995morality].

That result was important because it connected moral development to ordinary learning mechanisms. It also identified a real route by which another person’s welfare gains motivational force. Distress is not merely recognized after a transgression. It can train the controller so that a predicted social consequence changes an action before the transgression occurs.

Distress learning is not the whole of moral development. Many obligations protect people whose distress is invisible or delayed. Intention can matter even when no harm occurs, as in an attempted poisoning with an inert substance. A promise can bind action when breaking it would never be discovered. Fairness can apply to anonymous people who display no emotion. Human social control therefore draws on attachment, caregiving, observation, reward and punishment, language, stories, explicit rules, perspective taking, reputation, reciprocity, and repeated practice applying a norm to one’s own conduct.

The anatomy is correspondingly broader than a single amygdala-to-vmPFC cable. The uncinate fasciculus is one major pathway between anterior temporal and orbital-frontal territories, but it carries information serving many interpersonal, semantic, affective, and state-representation functions. Even within psychopathy research, reduced right-uncinate integrity has been related especially to interpersonal traits such as manipulation and pathological lying rather than uniformly to affective, lifestyle, and antisocial features [@wolfetal2015uncinate]. A tract can be important without being dedicated to conscience.

An evolutionary-developmental account therefore avoids two false choices. Moral control is not an innate module that unfolds independently of experience, and the developing child is not a blank slate on which distress conditioning writes an entire moral code. Selection supplies social sensitivities, attachment systems, affective learning, action control, and capacities for representing other agents. Development organizes those capacities within particular relationships and cultures. Distress learning is one major route into that system because another person’s suffering is one of the most powerful consequences a social animal can encounter.

40.4 What focal lesions reveal about social control

Focal lesions provide causal evidence that orbital and ventromedial prefrontal territories are necessary for important parts of social control. After damage in adulthood, language, factual knowledge, and many conventional intellectual abilities can remain impressive while work, finances, relationships, judgment, and self-regulation deteriorate. Historical labels such as acquired sociopathy captured the severity of the personality change. They did not identify a single lost faculty.

One experiment isolated the difference between knowing a norm and applying it online. Patients with orbitofrontal damage participated in social interactions that created opportunities to violate norms of interpersonal distance and intimacy. Their behavior was more inappropriate, and during the interaction they did not recognize that it violated norms they could state. When they later watched recordings of themselves, they recognized the transgression and showed embarrassment [@beeretal2006selfmonitoring]. The social rule had not vanished. The failure occurred while the current situation, the self, and the rule had to be integrated quickly enough to alter behavior.

That result is the social counterpart of goal neglect. A control state can be present in declarative form yet fail to govern the moment in which it matters. Orbital and ventromedial cortex contributes to representing the current social state, relating it to learned outcomes, and updating behavior from the consequences of one’s own conduct. Damage weakens this online integration. It does not erase every feeling or every proposition about right and wrong.

Prosocial motivation provides a second causal test. In an effort-based decision task, patients with focal vmPFC lesions chose whether to exert physical force to earn rewards for themselves or for another person. The patients earned less for others, discounted others’ rewards more steeply by effort, and exerted less force when another person benefited [@lockwoodetal2024prosocial]. The lesion maps also exposed heterogeneity within vmPFC: more medial damage reduced prosocial motivation, whereas some lateral damage was associated with increased prosociality relative to other lesion locations. The result establishes a causal contribution without turning a large and architectonically varied territory into one prosocial organ.

Social judgment also depends on combining an outcome with the mental state that produced it. Patients with vmPFC damage can judge an unsuccessful attempt to harm as too permissible because the intended harm never occurred [@youngetal2010harmfulintent]. In a full intention-by-outcome design, they place too much weight on the visible result: attempted harm is undercondemned, while accidental harm is judged too harshly [@ciaramellietal2012outcome]. The deficit is not simple indifference to harm. It is a failure to integrate an inferred intention with a salient outcome when evaluating another agent.

Developmental timing changes the syndrome. Anderson and colleagues described two adults whose focal prefrontal injuries occurred before 16 months of age. Like adults with later damage, they showed severe failures of social conduct despite preserved basic cognitive abilities. Unlike typical adult-onset cases, they also showed defective acquisition of social conventions and moral reasoning [@andersonetal1999earlypfc]. These rare cases establish that an intact prefrontal contribution during development helps build the knowledge and policies later used for social control. They do not show that one frontal region contains a moral code. Early injury changes the development of a distributed system, including the experiences the child elicits from other people and the learning opportunities that follow.

The lesion evidence therefore supports a strong conclusion. Orbital and ventromedial systems help make social information behaviorally effective by relating the current interpersonal state to expected outcomes, learned values, effort, self-monitoring, and inferred intention. Adult injury can uncouple social knowledge from conduct. Very early injury can also derail the acquisition of social knowledge itself. Neither syndrome is adequately described as the destruction of a stored conscience.

40.5 Developmental psychopathy is not acquired frontal injury

The partial resemblance between focal frontal injury and psychopathic traits has motivated important experiments. It has also encouraged a misleading equation. Developmental psychopathy is not an adult vmPFC lesion moved backward in time.

Psychopathy, antisocial personality disorder, and callous-unemotional traits

Psychopathy is a multidimensional research and forensic construct involving varying interpersonal, affective, lifestyle, and antisocial features.

Antisocial personality disorder is a clinical diagnosis that places greater weight on persistent antisocial behavior.

Callous-unemotional traits describe a developmental dimension studied in children and adolescents, often in relation to conduct problems.

The constructs overlap, but they are not interchangeable. Children should not be treated as members of a fixed class of “psychopaths.”

Psychopathy scores vary along dimensions rather than separating people into a discrete natural kind [@waltersetal2007dimensionality]. Callous-unemotional traits in youth also vary dimensionally rather than forming a discrete class of young “psychopaths” [@murrieetal2007youth]. The common facets also matter. Interpersonal features include manipulation, grandiosity, and pathological lying. Affective features include shallow affect, reduced remorse, and callousness. Lifestyle features include irresponsibility and stimulation seeking. Antisocial features include persistent rule breaking and aggression. Two people with similar total scores can reach them through different combinations.

Impulsivity is therefore not universal. Lifestyle and antisocial facets are associated with rash action, whereas some interpersonal and affective features are associated with greater planning and persistence [@grayetal2019impulsivity]. Some individuals are disorganized and reactive; others are controlled and strategic. A model that defines the construct by poor behavioral control misses this heterogeneity.

The clearest recurring difference concerns affective responsiveness to other people. Across studies, psychopathic traits are more strongly related to reduced affective empathy than to impaired cognitive perspective taking, and the relationship differs across facets [@camposetal2022empathy]. A person can often identify what another person feels, predict what that person will do, and describe the rule being violated while assigning the information less motivational weight. That pattern is socially consequential precisely because mental-state understanding can remain useful for manipulation even when concern is weak.

Distress and fear processing remain central findings. Children and adolescents with disruptive behavior and high callous-unemotional traits have shown reduced amygdala responses to fearful expressions [@marshetal2008fear]. Yet the response depends partly on how the cue is sampled. Directing children to look at the eyes improves fear recognition, demonstrating that reduced attention to an informative feature can contribute to the apparent affective deficit [@daddsetal2008eyegaze]. In adults, attentional focus can also determine whether psychopathy-related fearlessness appears in defensive physiology [@newmanetal2010attention]. The social signal must enter the effective state before it can guide learning or action.

Developmental pathways differ as well. A low-anxiety or “primary” variant has been associated with reduced amygdala and insula responses to fear, whereas a high-anxiety or “secondary” variant did not show the same pattern and instead differed in other systems [@sethietal2018variants]. These are not two severities of one broken circuit. They are partly different organizations of affect, attention, learning, and control that can produce overlapping antisocial behavior.

The evidence from the moral–conventional task follows the same pattern. Blair’s original result supported the prediction that severe psychopathy would reduce the special status assigned to harm-based violations [@blair1995morality]. Later studies did not find a universal loss of the distinction. Adolescent offender groups have shown differences under some authority conditions but not a wholesale collapse of ordinary moral–conventional judgment [@dolanfullam2010moralconventional]. In a large offender sample, total psychopathy did not predict failure on every form of the task [@aharonietal2012moralwrong]. Adults with psychopathy can also give conventional answers to many sacrificial moral dilemmas [@cimaetal2010rightwrong]. The capacity to state a judgment is not the same as the affective learning, concern, or action control that supports conduct outside the laboratory.

Autonomic and structural findings require the same decomposition. Reduced electrodermal response to another person’s pain has been observed among violent offenders with both high and lower psychopathy scores, so the response is not a specific physiological signature of psychopathy [@pfabiganetal2015empathy]. Reduced integrity of the right uncinate fasciculus has been reported in relation to psychopathy, but in the largest cited diffusion study the association was concentrated in the interpersonal facet rather than the affective or antisocial facets [@wolfetal2015uncinate]. A group difference in one measure cannot stand for an absent capacity for conscience.

The strongest synthesis is positive. In important subgroups, atypical attention to distress, affective learning, amygdala and insula responsivity, reward processing, and frontotemporal interaction reduce the force with which another person’s welfare constrains action. Other individuals arrive at antisocial behavior through high negative affect, impulsivity, adversity, reinforcement history, or failures of context-sensitive control. Development unfolds through transactions among temperament, attachment, caregiving, peers, opportunity, punishment, language, and culture. No single developmental failure defines everyone with high psychopathy traits.

This distinction matters scientifically and ethically. Acquired lesions identify operations that depend causally on the damaged region. Developmental traits reflect a nervous system and social history that have organized together over years. Partial behavioral overlap is evidence that the systems share components, not that the two conditions are mechanistically identical.

40.6 Moral dilemmas decompose social judgment

Moral dilemmas became influential because a small change in causal structure can reverse a judgment while the final arithmetic remains constant. In the switch case, a runaway trolley will kill five people unless a bystander diverts it onto a track where it will kill one. In the footbridge case, the five can be saved only by pushing a person into the trolley’s path. Both cases exchange one death for five lives. Most participants are more willing to operate the switch than to push the person.

An early dual-process account interpreted the contrast as competition between an affective response to personal harm and a more controlled calculation of aggregate consequences [@greeneetal2001moral]. The model was valuable because it generated causal predictions. It was too coarse because the two scenarios differ along many dimensions at once: direct physical force, action versus redirection, using a person as a means, causal contact, vividness, responsibility, and the anticipated emotional and social consequences of acting.

Focal lesion studies showed that vmPFC contributes to the judgment, but the effect was selective. Patients with bilateral vmPFC damage were more willing than comparison participants to endorse harmful action in high-conflict personal dilemmas. Their judgments were not globally different across impersonal cases, low-conflict personal cases, and ordinary moral scenarios [@koenigsetal2007moraljudgment; @ciaramellietal2007personal]. The lesion shifted a restricted class of decisions in which instrumental harm evokes strong competing aversion.

Calling the shifted response “more rational” misstates the result. The lesion did not improve arithmetic, reveal a hidden moral truth, or strengthen impartial concern. It reduced the influence of information that normally opposes direct instrumental harm. The precise description is greater endorsement of instrumental harm to improve the aggregate outcome.

That wording matters because sacrificial dilemma responses are not a complete measure of utilitarian morality. A philosophical commitment to impartial welfare would also predict costly generosity, concern for distant strangers, and willingness to sacrifice one’s own interests. Endorsing harm in trolley cases shows little positive relation to those commitments and can instead track reduced aversion to harming [@kahaneetal2015utilitarian]. The laboratory response isolates one component of judgment, not a moral worldview.

Intention reveals another component. Moral evaluation combines what happened with what the actor believed and tried to do:

	Harmless outcome	Harmful outcome
Neutral belief or intention	No harm	Accidental harm
Harmful belief or intention	Attempted harm	Intentional harm

A controller that used outcome alone would condemn accidental harm and excuse a failed attempt. Typical adults instead assign substantial weight to belief and intention. Disrupting right temporoparietal junction activity with transcranial magnetic stimulation reduces the weight of beliefs in moral judgment [@youngetal2010rtpj]. Focal vmPFC damage produces a complementary integration failure: attempted harms are judged too permissible and accidental harms too blameworthy [@youngetal2010harmfulintent; @ciaramellietal2012outcome]. Temporoparietal systems contribute information about another person’s mental state; vmPFC helps relate that inferred state to outcome value and the current judgment.

These experiments replace the image of a hot moral system fighting cold reason with a more informative decomposition. A moral judgment can require representation of agency, belief, intention, causal role, outcome, norm, relationship, anticipated feeling, and available action. Different experimental scenarios change several of these variables together. Lesions and stimulation reveal which information loses influence; they do not expose a single moral faculty.

40.8 Prospective social control belongs to a distributed loop

The chapter’s positive architecture is a recurrent loop among specialized systems. It is not the claim that every brain region does everything, and it is not a serial pathway in which one region detects distress, another computes morality, and a third executes the verdict.

Perceptual systems identify agents and social signals. Faces, voices, bodies, gaze, biological motion, and action kinematics provide evidence about what another organism is doing and feeling. Sampling and attention determine which evidence enters the control state.
Posterior temporal, temporoparietal, and medial systems represent agency and mental state. They help estimate who acted, what that person could see, what they believed, and whether the outcome was intended. These variables change the meaning of the same physical event.
Hippocampal, temporal, and cortical memory systems supply relationship history. A stranger, child, rival, friend, debtor, and trusted partner do not enter valuation with the same prior. Memory provides episodes, person knowledge, promises, debts, and the context in which a current action must be interpreted.
Amygdalar and insular systems contribute learned affective significance and bodily relevance. Distress, threat, affiliation, rejection, and uncertainty can alter attention, autonomic state, and readiness to approach, help, withdraw, appease, or defend. These contributions make another person’s state consequential without defining the correct action by themselves.
Hypothalamic and brainstem systems organize bodily preparation. Social encounters change circulation, respiration, endocrine state, defensive readiness, vocalization, posture, and orienting. The body is part of the interpersonal controller, not an audience waiting for a cortical decision.
OFC and vmPFC relate inferred states and outcomes to current value. They help represent what situation the interaction is in, which learned schema applies, how self and other outcomes are weighted, how much effort an outcome warrants, and how intention changes the value of a result. Focal lesions show that these contributions are causally important and regionally heterogeneous.
Striatal and dopamine-related systems update social policies. Approval, rejection, reciprocity, punishment, cooperation, and betrayal alter which actions are likely to be selected in later interactions. Social learning uses the same general reinforcement machinery as nonsocial learning while operating over outcomes that depend on another agent.
Lateral and medial frontal systems maintain commitments and regulate action. A promise, role, instruction, or long-term relationship must remain effective when an immediate temptation is stronger. These systems contribute rules, conflict, effort, sequencing, and the maintenance or revision of a selected policy.
Language and culture extend control across people and time. Words preserve norms, contracts, reputations, laws, and narratives beyond the episode in which they were learned. Institutions distribute social memory across records and other people, allowing action to be governed by absent beneficiaries and consequences that may occur years later.
The action changes the other controller. The other person learns, responds, and changes the shared environment. Their response becomes new evidence about intention, trust, power, and future opportunity. Social control is therefore closed-loop control between agents, not a private moral calculation followed by an output.

This architecture explains why no single deficit produces every form of antisocial behavior. A person can fail to notice distress, understand intention, retrieve a promise, value another’s outcome, maintain a norm, control an impulse, or learn from the partner’s response. The resulting behaviors can look similar while arising at different points in the loop. Conversely, an abnormality in one component need not abolish the others.

Regional specialization remains real. Amygdala damage is not equivalent to temporoparietal disruption; early frontal injury is not equivalent to adult injury; reduced affective empathy is not equivalent to deficient perspective taking; and knowledge of a rule is not its successful application. The distributed account gains explanatory power by preserving these distinctions and showing how their outputs become mutually constraining.

The control-theoretic core is straightforward. Another person’s possible future enters the actor’s current state through prediction. Affect, memory, intention, norm, relationship, and expected response alter the value of available actions. Action changes both agents and the relationship between them. Feedback updates the model. What ordinary language calls conscience emerges when this recurrent system reliably allows the futures of others to govern present conduct.

40.9 Coda: information for futures shared with others

Social control is limited by what the organism knows. A partner’s reliability, a rival’s likely response, the location of shared resources, the meaning of a local signal, and the norm governing an unfamiliar group are not given in advance. A controller with an inaccurate model of other agents will misjudge both danger and opportunity.

Information about others must therefore be acquired before every need becomes urgent. Social animals watch, approach, test, play, eavesdrop, follow, and sample. Humans ask questions, tell stories, inspect reputations, enter new groups, and explore institutions. These actions cost time and energy and can expose the organism to rejection, deception, or threat. They also enlarge the set of social futures that can later be predicted and controlled.

The next chapter, Wandering When Full, treats exploration as an allostatic investment in future controllability. The extension is now clear. Exploration does not merely discover food, shelter, and routes. It also discovers agents: who knows what, who can be trusted, which signals matter, and which relationships can support action when circumstances change.

Established findings, working synthesis, and open questions

Established findings. Distress cues shape social learning. Focal orbital and ventromedial injury can impair social conduct, online self-monitoring, prosocial motivation, and the integration of intention with outcome while sparing many conventional cognitive abilities. Very early prefrontal injury can profoundly disrupt later social and moral development. Psychopathy is multidimensional; higher affective-interpersonal traits are often associated with reduced affective empathy and atypical processing of distress. vmPFC injury selectively alters some high-conflict moral judgments rather than moral judgment in general.

Working synthesis. Another person’s future becomes behaviorally effective through the integration of affective significance, mental-state inference, remembered relationships, learned norms, expected reciprocity or sanction, and control of one’s own action. OFC and vmPFC are important nodes, especially where social outcomes must be related to value, effort, context, and policy. They do not constitute a conscience module. Social control closes a recurrent loop between autonomous agents.

Open questions. Experiments must determine when vicarious bodily response causally drives prosocial action, how distinct psychopathy dimensions arise developmentally, how attentional and affective mechanisms interact, how vmPFC subregions make different social contributions, and how well hypothetical dilemmas and economic games predict conduct in enduring relationships and institutions.