Why People Pay for Justice: Science Reveals the Dark and Light Sides of Altruistic Punishment

The article analyzes the phenomenon of altruistic punishment — people's willingness to bear personal costs to punish norm violators even without personal benefit. It summarizes key research from behavioral economics, evolutionary biology, and neuroscience since the groundbreaking study by Fehr and Gächter (2002), including public goods game experiments, cross-cultural comparisons, and neuroimaging studies revealing the activation of the brain's reward centers during punishment.

People willingly incur personal costs to punish those who violate cooperation norms — even in situations where they can derive no personal benefit from the punishment. This phenomenon, known as altruistic punishment (in the scholarly literature, "altruistic punishment"), represents one of the most provocative discoveries in behavioral economics of the past two decades. Since the groundbreaking 2002 study by Ernst Fehr and Simon Gächter, an extensive debate has grown up around this concept, reaching into evolutionary biology, neuroscience, cultural anthropology, and legal theory. Research has revealed that punishment activates the brain's reward centers, that it differs fundamentally across cultures, and that its "altruistic" nature is contested. Understanding this phenomenon has direct implications for how we conceive of legal systems, internet platforms, and organizational cultures.

The concept of altruistic punishment was formally introduced by Ernst Fehr and Simon Gächter in the article "Altruistic punishment in humans," published in the journal Nature (2002, volume 415, pages 137–140). They defined it as a situation in which "individuals punish, although the punishment is costly for them and yields no material gain." Crucially, behavior meeting this definition was observed even in anonymous one-shot interactions, where the punisher can expect no future reciprocity, no reputational benefit, and no repeated encounter with the person punished.

The act is "altruistic" in this context in the behavioral sense — it is costly for the actor and beneficial for the other members of the group, because it deters free-riding. The definition does not require an altruistic motivation; it suffices that the behavior meets the formal criteria of costliness and group benefit.

Distinguishing it from other forms of punishment is essential to grasping the specificity of this concept. Reciprocal punishment is grounded in repeated interactions — the punisher refuses future cooperation with a cheater, but in doing so also penalizes the other cooperators in the group (Boyd and Richerson, 1992, Ethology and Sociobiology, volume 13, pages 171–195). Institutional punishment is administered by formal institutions (courts, police), with the costs shared collectively through taxes. Retaliatory (vengeful) punishment is motivated by personal revenge — the punisher was directly harmed and seeks retribution. Second-party punishment comes from the direct victim, whereas third-party punishment, in which an uninvolved observer punishes a norm violator at their own expense, represents the "purest" form of altruistic punishment (Fehr and Fischbacher, 2004, Evolution and Human Behavior, volume 25, pages 63–87).

The free-rider problem and its solution. The central evolutionary puzzle is this: why would natural selection favor individuals who bear the costs of enforcing cooperation, when they could "ride free" on the efforts of others? In public-good situations, every individual has an incentive to free-ride — to exploit the contributions of others without contributing themselves. Without enforcement mechanisms, cooperation collapses into the classic "tragedy of the commons." The standard evolutionary mechanisms — kin selection (Hamilton, 1964, Journal of Theoretical Biology, volume 7, pages 1–52), reciprocal altruism (Trivers, 1971, Quarterly Review of Biology, volume 46, pages 35–57), and indirect reciprocity — can explain cooperation in small kin groups or pairwise relationships, but they fail to explain large-scale cooperation among unrelated strangers.

Moreover, there is the so-called second-order free-rider problem (Heckathorn, 1989, Rationality and Society, volume 1, pages 78–100): even if punishment sustains cooperation, punishers are at a disadvantage relative to cooperators who do not punish — the latter enjoy the benefits of cooperation without bearing the costs of enforcement.

Strong reciprocity. The concept of strong reciprocity was introduced by Herbert Gintis (2000, Journal of Theoretical Biology, volume 206, pages 169–179). He defined it as a predisposition to cooperate with others and to punish non-cooperators, even in anonymous one-shot interactions where no future benefit can be expected. Strong reciprocity thus transcends the bounds of both kin selection and reciprocal altruism.

Empirical support was provided by Fehr, Fischbacher, and Gächter (2002, Human Nature, volume 13, pages 1–25), who showed that strongly reciprocal individuals can enforce near-universal cooperation even under conditions where pure self-interest would lead to a complete collapse of cooperation.

The key asymmetry: why selection tolerates punishers. A groundbreaking theoretical model was put forward by Boyd, Gintis, Bowles, and Richerson (2003, PNAS, volume 100, pages 3531–3535). They identified a fundamental asymmetry between altruistic cooperation and altruistic punishment: whereas the disadvantage of cooperators relative to cheaters is constant (equal to the cost of cooperation), the disadvantage of punishers relative to non-punishing cooperators declines as cheaters become rare — because there are fewer people to punish. When punishers are common, individual selection against them is weak, which allows group selection — competition between groups, in which more cooperative groups outcompete less cooperative ones — to sustain punishment even in large groups.

Cultural evolution and gene-culture coevolution. Boyd and Richerson (1992) showed that "retribution" — punishment directed exclusively at non-cooperators — can sustain cooperation in large groups in two ways: if the benefits of cooperation exceed the costs of enforcement, and if the cost of punishment is sufficiently high, then "moralistic strategies" (cooperate, punish non-cooperators, and punish those who do not punish) can be evolutionarily stable. The crucial caveat is that moralistic strategies can stabilize any costly behavior — whether group-beneficial or not — which explains the existence of costly cultural practices such as rituals or taboos.

Henrich and Boyd (2001, Journal of Theoretical Biology, volume 208, pages 79–89) developed a model in which cooperation and punishment norms are acquired through conformist transmission — the tendency to copy the most common behavior. Even a small degree of conformism stabilizes punishment, and once cooperation is stabilized in one group, it can spread through populations via cultural group selection.

A comprehensive synthesis of these findings was provided by Bowles and Gintis (2011, A Cooperative Species: Human Reciprocity and Its Evolution, Princeton University Press). They used the Price equation (Price, 1970, Nature, volume 227, pages 520–521) and multilevel selection models to analyze the evolution of altruism, showing how genetic differentiation between groups, maintained by cultural conformity, allows group selection to overcome within-group selection. A later model of coordinated punishment (Boyd, Gintis, and Bowles, 2010, Science, volume 328, pages 617–620) addresses the problem of how punishment arises: in real human societies people "gang up" on cheaters — punishment is coordinated, the costs are shared, and therefore it can emerge even when it is initially rare.

Richerson et al. (2016, Behavioral and Brain Sciences, volume 39, e30) provided a comprehensive review of evidence from five domains supporting cultural group selection as necessary to explain human cooperation.

The public goods game with punishment (Fehr and Gächter, 2000, 2002). The key experiment was conducted by Fehr and Gächter first in 2000 (American Economic Review, volume 90, pages 980–994) and definitively in 2002 (Nature, volume 415, pages 137–140). In the public goods game, groups of four players received an initial endowment of 20 monetary units per round. Players could contribute any amount (0 to 20) to a group project with a marginal per-capita return of 0.4 — total contributions were multiplied by 1.6 and divided equally. Contributing is therefore collectively beneficial but individually disadvantageous.

In the punishment condition, each player, after learning the others' contributions, could assign 0 to 10 punishment points to anyone in the group. Each punishment point cost the punisher 1 monetary unit and reduced the recipient's income by 3 units (a cost ratio of 1:3). The experiment included two key variants: "stranger" (random reshuffling each round) and "perfect stranger" (a guaranteed one-shot encounter), thereby ruling out the effects of reciprocity and reputation.

The results were dramatic. Without punishment, average contributions fell from an initial 8–10 units to roughly 2 units in the final round; in the earlier experiment (Fehr and Gächter, 2000), 73% of players contributed zero in the final round. With the option to punish, contributions instead rose — in the "partner" condition (repeated interactions) they reached nearly the maximum of 19–20 units. Crucially, punishment also occurred in the "perfect stranger" condition, where it was genuinely one-shot and anonymous. Punishment was, moreover, directed above all at players contributing below the group average — the larger the negative deviation, the more punishment.

The ultimatum game and third-party punishment. The ultimatum game (Güth, Schmittberger, and Schwarze, 1982, Journal of Economic Behavior and Organization, volume 3, pages 367–388) represents the simplest framework for studying the costly punishment of unfairness. A proposer divides a sum of money; the responder either accepts or rejects the offer — if it is rejected, both get nothing. Game theory predicts that the responder will accept any positive offer, but empirically offers below 20% are rejected in roughly half of cases. Typical offers reach 40–50% of the total sum (a meta-analysis by Oosterbeek et al., 2004, confirms an average of about 40% across 37 papers comprising 75 experiments). Rejection is a clear form of costly punishment — the responder sacrifices real money to punish an unfair division.

Third-party punishment was studied by Fehr and Fischbacher (2004, Evolution and Human Behavior, volume 25, pages 63–87). In their experiment, a third player, whose payoff was unaffected by the decisions of the others, observed behavior in a dictator game or prisoner's dilemma and could punish the norm violator at their own expense. Approximately 60–65% of third parties punished violations of both distribution and cooperation norms, with the strength of punishment increasing with the severity of the violation. Even uninvolved observers, then, pay to enforce norms.

Cross-cultural study: fifteen diverse populations. The largest cross-cultural study was conducted by Henrich et al. (2006, Science, volume 312, pages 1767–1770) across fifteen diverse populations around the world — hunter-gatherers, horticulturalists, pastoralists, and smallholder farmers from Africa, South America, Oceania, and Asia. All fifteen populations showed some willingness to punish at a cost as inequitable behavior increased. At the same time, however, there was enormous cross-cultural variability in the degree of punishment. Costly punishment was positively associated with altruistic behavior — societies with higher offers in the dictator game also showed a greater willingness to punish. An earlier study by Henrich et al. (2005, Behavioral and Brain Sciences, volume 28, pages 795–855) showed that average offers in the ultimatum game ranged from 26% (the Machiguenga, Peru) to 58% (the Lamalera, Indonesia), with the degree of market integration and the need for cooperation in everyday life being the best predictors of prosocial behavior.

Punishment versus reward. The effectiveness of punishment versus positive incentives was examined by Rand, Dreber, Ellingsen, Fudenberg, and Nowak (2009, Science, volume 325, pages 1272–1275) in a public goods game with 50 rounds. Reward proved to be just as effective as punishment for sustaining cooperation, but it led to higher overall payoffs, because punishment destroys value. When both options were available, reward increased contributions and payoffs, whereas punishment had no effect on contributions and reduced payoffs. The authors concluded that human cooperation in repeated interactions is best supported by positive interactions.

The striatum and the "sweet taste of revenge." A groundbreaking neuroimaging study was published by de Quervain, Fischbacher, Treyer, Schellhammer, Schnyder, Buck, and Fehr (2004, Science, volume 305, pages 1254–1258). Using positron emission tomography (PET), they scanned the brains of fifteen men while they decided whether to punish cheaters in a trust game. The key element consisted of two conditions: real punishment (which actually reduced the cheater's payoff) versus symbolic punishment (with no financial consequences).

Real punishment, compared with symbolic punishment, activated the dorsal striatum (the dorsal caudate nucleus), a brain region involved in processing the rewards of goal-directed actions. Participants with stronger striatal activation were willing to bear greater costs to punish cheaters — the positive association between striatal activity and willingness to pay for punishment is the study's key finding. The ventromedial prefrontal cortex and the medial orbitofrontal cortex were activated when participants strongly desired to punish but punishment was costly, suggesting that these regions weigh the benefits and costs of punishment.

These findings support the hypothesis that people derive intrinsic satisfaction from punishing norm violators — the anticipation of "justice" motivates costly punishment. As Brian Knutson of Stanford noted in a commentary in the same issue of Science: instead of cold, calculating rationality, the driving force behind punishment may be passion, not reason.

The clash of emotion and reason: the insula versus the prefrontal cortex. Sanfey, Rilling, Aronson, Nystrom, and Cohen (2003, Science, volume 300, pages 1755–1758) scanned the brains of responders in the ultimatum game using functional magnetic resonance imaging (fMRI). Unfair offers activated both emotional and cognitive regions: the anterior insula — associated with negative emotional states and feelings of disgust — and the dorsolateral prefrontal cortex (DLPFC) — associated with cognitive control and reasoning. What proved decisive was whether the insula or the DLPFC dominated: when insular activity exceeded DLPFC activity, participants rejected the unfair offer; when the DLPFC dominated, they accepted it. This was a founding study of neuroeconomics, showing that emotions play a key role in economic decisions about punishment.

Knoch, Pascual-Leone, Meyer, Treyer, and Fehr (2006, Science, volume 314, pages 829–832) used repetitive transcranial magnetic stimulation (rTMS) to temporarily disrupt the right DLPFC. The disruption significantly increased acceptance of unfair offers — participants were less willing to punish unfairness. Crucially, disrupting the DLPFC did not change the assessment of fairness — participants still judged unfair offers to be unfair; only their behavior changed.

Buckholtz et al. (2008, Neuron, volume 60, pages 930–940; 2015, Neuron, volume 87, pages 1369–1380) investigated the neural correlates of third-party punishment in criminal scenarios. The amygdala, medial prefrontal cortex, and posterior cingulate cortex predicted the magnitude of punishment (emotional processing), whereas the right DLPFC distinguished between scenarios on the basis of criminal responsibility. Inhibitory TMS of the right DLPFC reduced punishment without affecting the assessment of guilt, showing that the assessment of wrongdoing and the decision about punishment are cognitively and neurally distinct processes. Buckholtz and Marois (2012, Nature Neuroscience, volume 15, pages 655–661) proposed a comprehensive neurobiological model of third-party punishment involving a cascade of five regions: the temporoparietal junction (encodes intent), the amygdala (generates emotional arousal), the medial prefrontal cortex (integrates the signals), the DLPFC (decides on punishment), and the intraparietal sulcus (response selection).

Hedonic reward and reputational signaling. Paradoxically, the very neuroscientific findings that confirmed the significance of altruistic punishment simultaneously cast doubt on its "altruistic" character. If punishers derive hedonic satisfaction from the activation of the striatal reward system (de Quervain et al., 2004), then punishment is not purely altruistic but also serves personal emotional gratification.

Brethel-Haurwitz, Stoycos, Cardinale, and Marsh (2016, Scientific Reports, volume 6, 18974) tested whether extraordinary real-life altruists (for example, kidney donors) punish more in the ultimatum game. Extraordinary altruists did not punish more than the control group. The authors concluded that what is often called altruistic punishment should more accurately be called costly punishment.

Pedersen, Kurzban, and McCullough (2013, Proceedings of the Royal Society B, volume 280, 20122723) demonstrated that while victims of unfairness punished, witnesses of unfairness did not. The emotional reactions of witnesses were characterized more by envy of the selfish gain than by moral outrage. In a follow-up study (Pedersen, McAuliffe, and McCullough, 2018, Journal of Experimental Psychology: General, volume 147, pages 514–544), the authors confirmed that uninvolved witnesses do not punish altruistically.

The theory of costly signaling (Gintis, Smith, and Bowles, 2001, Journal of Theoretical Biology, volume 213, pages 103–119) proposes that punishment may function as a costly signal of an individual's quality — much like the peacock and its tail — signaling reliability and trustworthiness. Nelissen (2008, Evolution and Human Behavior) showed that only altruists who invested the most in punishment were preferred as partners in subsequent trust games, confirming the signaling function.

Antisocial punishment: when cooperators get punished. One of the most important findings in the entire scholarly literature is the study by Herrmann, Thöni, and Gächter (2008, Science, volume 319, pages 1362–1367). In public goods game experiments with participant groups from various cities around the world, they documented antisocial punishment — the sanctioning of prosocial individuals, that is, the punishment of those who contributed the most.

The cross-cultural variability was massive. In some groups (Boston, Copenhagen, Zurich, Melbourne) people punished free-riders above all. In others (Muscat, Athens, Riyadh, Dnipropetrovsk, Minsk, Samara), antisocial punishment was so strong that it completely eliminated the cooperation-enhancing effect of punishment. The predictors of antisocial punishment included weak norms of civic cooperation (measured by the World Values Survey) and a weak rule of law. The conclusion is fundamental: punishment options are socially beneficial only when they are complemented by strong social norms of cooperation.

Winners don't punish: the cost to the group. Dreber, Rand, Fudenberg, and Nowak (2008, Nature, volume 452, pages 348–351) conducted an experiment in which 104 participants played an extended prisoner's dilemma with the options to cooperate, defect, or punish at a cost. They found a strong negative association between total earnings and the use of costly punishment. The five highest-earning players never used costly punishment. Punishment increased the frequency of cooperation but did not increase the group's average payoffs — the costs of punishment outweighed the gains from increased cooperation. Martin Nowak summed it up: "In an extremely competitive setting, the winners are those who resist the temptation to escalate conflicts, while the losers punish and perish."

Egas and Riedl (2008, Proceedings of the Royal Society B, volume 275, pages 871–878) added that cooperation is sustained only when the cost-to-impact ratio of punishment is favorable — low cost to the punisher and high impact on the punished. With an unfavorable ratio, punishment fails. Nikiforakis (2008, Journal of Public Economics, volume 92, pages 91–112) showed that when retaliatory punishment is possible, punishment fails to sustain cooperation and can lead to destructive punishment wars.

Alternatives: exclusion, positive incentives, institutions. A meta-analysis by Balliet, Mulder, and Van Lange (2011, Psychological Bulletin, volume 137, pages 594–615), comprising 187 effect sizes, found that rewards and punishments have a statistically comparable positive effect on cooperation (d = 0.51 for rewards, d = 0.70 for punishments). Hilbe and Sigmund (2010, Proceedings of the Royal Society B) proposed an optimal strategy: first build cooperation with rewards and only then move on to punishing the remaining cheaters.

Exclusion from the group is, according to anthropological evidence, a far more typical form of punishment in early human societies than direct sanctions (Baumard, André, and Sperber, 2013). Gürerk, Irlenbusch, and Rockenbach (2006, Science, volume 312, pages 108–111) showed that when participants could choose between an institution with sanctions and one without, the entire population eventually migrated to the sanctioning institution and cooperated strongly. This finding was successfully replicated in a large inter-laboratory study (Przepiorka et al., 2023, PNAS, N = 1,008, 7 laboratories).

Retributive intuitions in legal systems. Research on altruistic punishment provides an empirical, evolutionary basis for understanding retributive intuitions in the legal context. Carlsmith, Darley, and Robinson (2002, Journal of Personality and Social Psychology, volume 83, pages 284–299), in three studies, found that despite a stated preference for deterrence, the actual punishment judgments handed down by laypeople were governed exclusively by the principle of just deserts — high sensitivity to the severity of the offense, but insensitivity to deterrence factors such as the probability of detection. Balafoutas, Nikiforakis, and Rockenbach (2016, Nature Communications, volume 7, 13327), however, showed that in field experiments altruistic punishment does not increase with the severity of the violation — fear of counterattack raises the marginal cost of punishing serious violations. This provides a rational justification for the existence of formal legal institutions that replace decentralized peer punishment.

Online communities and social media. Digital platforms have created new arenas for altruistic punishment: negative ratings, flagging, content moderation, and public shaming represent low-cost punishment mechanisms. Jordan (2019, Scientific Reports, volume 9, 2435) showed that cooperative behavior spreads through social networks from person to person, and that low-cost but high-impact punishment prolongs and broadens this cooperative wave. For how platforms are designed, this means that moderation tools that are easy to use (low cost) but carry real consequences (high impact) best sustain cooperation. Grimalda, Pondorfer, and Tracer (2016, Nature Communications, volume 7, 12288), however, found that building one's social image was a stronger driver of cooperation than altruistic punishment — reputation systems (ratings, follower counts) may be more effective than punishment mechanisms.

The phenomenon of "cancel culture" can be analyzed as a large-scale manifestation of altruistic punishment: individuals bear personal costs (the risk of backlash) to punish perceived norm violators. Herrmann et al. (2008) are especially instructive here: internet platforms, which often lack strong moderation, may be especially prone to antisocial punishment.

Organizations and corporate culture. Frey, Adams, Pfeffer, and Belmi (2023, Journal of Management, volume 49, pages 196–236), in an extensive review, showed that punishment in modern organizations (remote work, the platform economy) involves different actors, situations, and methods than in traditional organizations. Whistleblowing represents a direct analogue of altruistic punishment — the whistleblower bears personal costs (retaliation, career damage) to punish an organizational norm violation. Research on organizational justice shows that the perception of punishment as fair is a key predictor of subordinates' cooperative behavior; unfair punishments, by contrast, foster counterproductive behavior and revenge — an analogue of antisocial punishment.

The "confused learners" challenge. The most significant methodological challenge comes from the line of research by Burton-Chellew and West. Burton-Chellew, El Mouden, and West (2016, PNAS, volume 113, pages 1291–1296) showed that variability in behavior in the public goods game is better explained by variability in understanding of the game than by social preferences — when players played against computers (where no one can benefit from cooperation), they exhibited similar patterns. In a meta-analysis of 237 public goods games (Burton-Chellew and West, 2021, Nature Human Behaviour, volume 5, pages 1330–1338), they found that contributions declined faster when individuals had greater influence over their own payoffs — consistent with the "confused learners" hypothesis.

Burton-Chellew and Guérin (2021, Proceedings of the Royal Society B, volume 288, 20211611) experimentally separated cooperation and punishment: individuals immune to punishment reduced their cooperation, even though they were surrounded by cooperators, but continued to punish without cooperating themselves. Cooperation and punishment therefore do not form a single linked altruistic trait. West and Burton-Chellew (2025, Evolution and Human Behavior, volume 46, 106749) argued that replicating poorly designed experiments canonizes incorrect conclusions — the basic public goods game has been replicated more than a hundred times with consistent results, but the experimental design cannot distinguish between prosocial preferences and confusion.

A direct rebuttal was provided by Wang et al. (2024, PNAS, volume 121, issue 10), who argued that confusion alone cannot explain the observed patterns of cooperative behavior in public goods games. The debate between the "strong reciprocity school" (Fehr, Gächter, Gintis, Boyd, Richerson) and the "skeptical school" (Burton-Chellew, West, Pedersen, Kurzban) remains open.

What is supported and what is not. The basic behavioral phenomenon — that people punish in anonymous one-shot interactions — has been replicated hundreds of times across diverse populations and cultures. That punishment increases the degree of cooperation in standard experimental setups is consistently confirmed. What remains contested is: whether the motivation is genuinely altruistic versus confused, competitive, or selfish; whether punishment increases net group welfare; and whether laboratory results hold under field conditions.

The most recent studies (2023–2025). Among the latest contributions is research on the cross-cultural reinforcement learning of altruistic punishment (2024, PLOS Computational Biology): Chinese participants showed higher learning rates than Americans, with the development of ingroup favoritism from adolescence into adulthood. Alam and Rai (2025, PNAS, volume 122, issue 34, e2508479122) brought the surprising finding that when third-party punishment is profitable, cooperation actually decreases — observers distrust the motives of a paid punisher and view social norms through the lens of self-interest. Yang, Hoffman, and Krueger (2024, Neuroscience & Biobehavioral Reviews, volume 157, 105525) provided a comprehensive review of the neuropsychological foundations of social punishment.

Research on altruistic punishment has followed a remarkable trajectory over two decades — from elegant experimental demonstration (Fehr and Gächter, 2002), through the neural identification of the "pleasure of justice" (de Quervain et al., 2004), to a profound critique that questions both the "altruistic" nature of this behavior and its universal functionality.

Three key insights rise above the original narrative. First, context is decisive: antisocial punishment (Herrmann et al., 2008) and the failure of punishment in the field for serious violations (Balafoutas et al., 2016) show that punishment is not a universal cure for the cooperation problem — it works only in the presence of strong social norms and institutional support. Second, alternatives exist: positive incentives, reputation systems, and institutional solutions can be equal or superior to peer punishment, especially in repeated interactions. Third, the motivation is complex: it is not pure altruism, but a mixture of moral outrage, hedonic reward, reputational signaling, and cognitive biases — and it is precisely this complexity that makes the phenomenon so authentically human.

The field stands at a genuine inflection point. The basic experimental phenomenon is well-supported, but its interpretation — and hence its implications for law, public policy, organizations, and the design of digital platforms — remains deeply contested. Future research will have to bridge the gap between the laboratory and the field, disentangle the motivational components of punishment, and explain how decentralized punishment gave rise to the central institutions of justice that define modern societies.

Transparency of creation

The conception, structure, and editorial line of the article are the work of the author, who prepared the content outline, set the key theses, and directed the entire creative process. Generative AI (Claude Opus 4.6, Anthropic) was used as a tool for research, fact-checking, and fleshing out the author's draft.

The author verified the key findings and approved the final wording. No part of the text was published without conscious authorial oversight. The factual data were verified against the publicly available sources cited in the text.

This procedure complies with the transparency principles of EU Regulation 2024/1689 (the AI Act). #poweredByAI