The research team, led by Ekaterina Grigorievna Tikhomirova of MEPhI’s Department of Philosophy, Ontology and Theory of Cognition, together with Roman Viktorovich Dushkin, Aleksey Alekseevich Kuzmin and Pavel Alexandrovich Cherepkov, set out to detect what they call civilisational profiles in the responses of six current models: YandexGPT and GigaChat, positioned as Russian developments; ChatGPT, a leading American model; Grok, an American open-architecture model; and DeepSeek and Qwen, two Chinese systems. The selection was deliberate. The authors chose models on the principle of civilisational representativeness, so that each of three broad cultural traditions would be represented by at least one system. Their initial hypothesis was that a model’s origin would correlate with the value profile it reproduces when confronted with questions about family, career, faith, beauty, labour, power and the nature of the human being.

That hypothesis mapped three distinct profiles onto the model landscape. A Western profile was expected to be individualist, grounded in personal autonomy and rational-contractual ethics. A Russian or Eurasian profile was expected to be collectivist, foregrounding duty, service and rootedness in community. A Chinese or Confucian profile was expected to be hierarchically collectivist, centred on family harmony and social stability. These categories draw on a long tradition in cross-cultural psychology, including Geert Hofstede’s comparative work on cultural dimensions and Shalom Schwartz’s theory of basic values, both of which appear in the study’s theoretical scaffolding. The question was whether the training pipelines behind modern chatbots preserve such differences or erase them.

To find out, the team developed an original methodological instrument: a civilisational profile detection method based on the coding of lexical markers of value systems. Rather than asking models abstract survey questions, which earlier work has shown can be unreliable, the researchers probed the systems in ways that force value commitments to surface in natural language. Each model was tested in two language modes, Russian and English, across three prompt types. The first was a request for advice in a situation of moral choice. The second was a request to conceptualise a value-laden term, such as justice or freedom. The third was a normative evaluation of a described action. The design meant that every model produced twelve distinct response sets, and the vocabulary of those responses could be systematically coded for the markers of individualist, collectivist and hierarchically collectivist value orientations.

The principal finding is stark. All six models, regardless of their country of origin or the language of the conversation, reproduced the Western civilisational profile as the dominant response strategy. The Russian models did not answer like Russian collectivists when asked in Russian. The Chinese models did not foreground Confucian family harmony when prompted in their developers’ cultural idiom. Instead, the same emphasis on individual choice, personal rights and contractual reasoning appeared across the board. For the authors, this is evidence of what they term worldview sovereignty being surrendered: the value hierarchy a model encodes is not determined by the culture of its developers but by something deeper in the data and training process that all major labs share.

That something is captured in the study’s central theoretical contribution, a concept the authors call worldview data annotation. The idea is that a hierarchy of values, norms and cultural orientations is encoded in the training corpus itself and then reproduced by the model as if it were neutral and self-evident. In other words, the moral framing of a chatbot’s answers is not an explicit design decision anyone made; it is an annotation layer written invisibly into billions of documents, most of which, in the corpora used to train frontier models, carry the assumptions of Western liberal individualism. Because the annotation is implicit, neither developers nor users notice it, and the model presents its inherited worldview as the natural, universal way to think about truth, honesty, justice, freedom, courage and valour.

The result fits into a rapidly growing body of research on cultural bias in language models. Earlier studies have probed pre-trained models for cross-cultural differences in values, measured cultural alignment against Hofstede’s dimensions, and documented how prompt language and explicit cultural framing affect model outputs. The Moral Machine experiment, published in Nature in 2018, showed long before the chatbot era that moral preferences vary dramatically across cultures, and subsequent work has asked whether large language models inherit or flatten those differences. Recent studies on collectivism and individualism in model judgments, on cultural bias measured through everyday scenarios, and on the homogenising effect of AI on culture have all pointed in a similar direction. The new study adds a cross-lingual, cross-civilisational experiment with a reproducible coding method, and its answer is that the flattening wins.

Why does the training pipeline override the culture of the developer? The authors’ framework suggests several converging mechanisms. The dominant pretraining corpora for large language models are heavily weighted toward English-language internet text, much of it produced in or shaped by Western contexts. Alignment and fine-tuning procedures, including the human feedback that teaches models to be helpful and harmless, tend to be carried out by teams and annotator pools whose own ethical assumptions reflect the same individualist baseline. And the technical vocabulary of AI ethics itself, with its emphasis on autonomy, transparency and individual rights, is largely a Western philosophical product, as surveys of global AI ethics guidelines have documented. A model built in Moscow or Hangzhou is therefore trained, aligned and evaluated inside an infrastructure whose moral defaults were set elsewhere.

The practical implications reach developers and regulators alike. For developers, the study offers a reproducible methodology for worldview auditing: a way to test, systematically and before deployment, which value profile a system actually reproduces when users ask it about moral dilemmas, contested concepts and normative judgments. Auditing for bias has so far focused largely on demographic fairness; the authors argue that cultural and civilisational alignment deserves the same scrutiny. For regulators, the findings complicate the notion that AI sovereignty can be achieved simply by building national models. If a domestically developed chatbot answers moral questions with an imported value system, then technical independence does not amount to cultural independence. Policy frameworks such as UNESCO’s Recommendation on the Ethics of Artificial Intelligence, which explicitly calls for pluralism and respect for cultural diversity, may require instruments like worldview audits to be enforceable in practice.

There is also a message for everyday users, and it is perhaps the most viral implication of the research: the chatbot on your phone is not a neutral oracle. When it advises you on a family conflict, a career decision or an ethical dilemma, it is not drawing on the accumulated wisdom of your own tradition. It is reproducing a specific, identifiable moral framework, one that prizes individual autonomy above duty, service and community rootedness, and it does so in whichever language you happen to speak. The study’s authors frame this as a problem of sovereign morality: the question of who holds the authority to define right and wrong when a machine answers on humanity’s behalf. Their experiment shows that, for now, that authority is concentrated in one civilisational tradition, and that neither geography nor language has been enough to change it. Making artificial intelligence genuinely plural, the study suggests, will require not just new models in new countries, but a deliberate re-examination of the data, the annotation practices and the ethical assumptions baked into the entire pipeline.

Subject of Research: Cultural and moral value bias in large language models across languages and countries of origin

Article Title: Worldview sovereignty and sovereign morality in LLMs: a cross-lingual experiment

Article References: Tikhomirova, E. G., Viktorovich, D. R., Alekseevich, K. A., & Cherepkov, P. A. (2026). Worldview sovereignty and sovereign morality in LLMs: a cross-lingual experiment. SN Social Sciences, 6(10), Article 515. https://doi.org/10.1007/s43545-026-01803-z

Image Credits: AI Generated

DOI: 10.1007/s43545-026-01803-z

Keywords: large language models, cultural bias, AI ethics, worldview auditing, civilisational profile, cross-lingual evaluation, individualism, collectivism, moral judgment, training data, AI sovereignty, ChatGPT

A new study published in Current Psychology by Liu Ting of Neusoft Institute Guangdong and Yuan Xiaolin of Hainan Normal University offers one of the sharpest answers yet, and the answer is that the effect is not one thing at all. By combining a classic experimental manipulation of mental effort with a formal modeling framework that disentangles different strands of moral decision-making, the researchers show that the foreign language effect wears two very different faces depending on how much cognitive resources the mind has available. When thinking capacity is plentiful, a foreign language appears to sharpen sensitivity to consequences. When that capacity is taxed, only the emotional side of the effect survives.

The methodological heart of the study is the CNI model, a multinomial processing tree model developed by Bertram Gawronski and colleagues that has rapidly become the gold standard for decomposing moral dilemma responses. Traditional trolley-style experiments yield a single binary choice per dilemma, which makes it impossible to tell whether someone endorsed a harmful action because they weighed outcomes heavily, because they discounted moral norms, or because they simply have a general disposition to act rather than remain passive. The CNI model solves this by presenting participants with dilemmas that systematically vary the consequences of action and inaction and the norms involved, then estimating three separate parameters: C, which captures sensitivity to consequences; N, which captures sensitivity to norms; and I, which captures a general preference for action over inaction, independent of any specific dilemma content.

This decomposition matters because the two leading theories of the foreign language effect make different predictions about which parameters should shift. The enhanced deliberation account, rooted in dual-process theories of moral judgment, holds that a foreign language consumes processing resources and slows down intuitive responses, giving deliberate, outcome-based reasoning more influence. On this view, foreign language should raise the C parameter. The diminished emotionality account, by contrast, holds that emotionally charged words learned in childhood carry stronger visceral resonance than their later-acquired translations, so a foreign language blunts the aversive feelings that normally inhibit harmful actions. On this view, foreign language should lower the I parameter, reducing the general inhibition against acting. Until now, most studies could not separate these possibilities.

The researchers recruited 120 native Chinese-speaking university students and randomly assigned them to complete a set of moral dilemmas either in Chinese, their first language, or in English, their foreign language. Crucially, they crossed language with a second manipulation: cognitive load. Half of the participants in each language condition performed the task under high cognitive load, holding a demanding memory load in mind while judging the dilemmas, while the other half worked under low load with their cognitive resources largely free. This two-by-two design allowed the team to ask a question no prior study had answered directly: what happens to the foreign language effect when the mind’s deliberative machinery is deliberately occupied elsewhere?

The results were strikingly patterned. Under low cognitive load, participants responding in English showed a significantly higher C parameter than those responding in Chinese, meaning they weighted the consequences of their choices more heavily, and a significantly lower I parameter, meaning their general reluctance to take action was reduced. In other words, when cognitive resources were abundant, the foreign language effect manifested as both enhanced outcome calculation and weakened action inhibition, exactly the signature predicted by a combination of the deliberation and emotionality accounts. The N parameter, sensitivity to norms, did not differ between languages, suggesting that the effect does not operate by loosening adherence to moral rules per se.

Under high cognitive load, however, the picture changed dramatically. The elevated C parameter disappeared entirely; the English and Chinese groups no longer differed in their sensitivity to consequences or in their sensitivity to norms. The only remaining difference was in the I parameter, which remained lower for the English group. When the deliberative pathway was blocked by an unrelated mental burden, the foreign language effect did not vanish. It shrank to its emotional core: a reduced general inhibition against action. This dissociation is the study’s central contribution, and it supports what the authors describe as the dominance of the emotional pathway when cognitive resources are depleted.

The findings carry important theoretical implications. They suggest that the foreign language effect is not a unitary phenomenon driven by a single mechanism, as much of the earlier literature implicitly assumed. Instead, it is a composite, with a deliberative component that requires available cognitive resources and an emotional component that persists even when those resources are exhausted. This may help explain why meta-analyses of the effect, including work by Stankovic and colleagues published in Brain and Language, have found considerable heterogeneity across studies: different experiments, with different time pressures, working memory demands, and participant populations, may be capturing different proportions of the two components. It also resonates with classic findings from Joshua Greene’s lab showing that cognitive load selectively interferes with utilitarian moral judgment, here refracted through the lens of language.

There are practical echoes as well. The foreign language effect has been documented in domains far beyond the trolley problem, including bribery susceptibility in workplace vignettes, medical decision-making, and everyday ethical choices. If the effect partly reflects reduced emotional inhibition, then people making consequential decisions in a second language, whether international negotiators, multilingual clinicians, or migrants navigating unfamiliar bureaucratic systems, may be less held back by the visceral discomfort that normally guards against harmful actions. At the same time, the finding that the deliberative component evaporates under load is a caution: the cool, outcome-sensitive reasoning that a foreign language can sometimes promote is fragile, easily crowded out by any competing mental demand, from stress to distraction to fatigue.

The study is not without limitations, some of which the authors acknowledge. The sample consisted of 120 Chinese university students judging dilemmas in Chinese and English, and the strength of the foreign language effect is known to depend on proficiency, age of acquisition, and the emotional proximity between languages, so the specific parameter profile may differ for other bilingual populations. The raw trial-level data from the underlying master’s thesis, completed in 2022, are no longer fully accessible, though summary statistics, parameter estimates, and model fit indices are reported in the manuscript and aggregated frequency tables are available as supplementary material. Still, by moving beyond the simple utilitarian-versus-deontological score and adopting a model that separates consequences, norms, and action tendencies, the study offers a template for how the field can finally untangle the psychology of why we seem to become, in some sense, slightly different moral agents when we switch languages. The mind, it turns out, does not simply become more rational in a foreign tongue. It becomes freer, and only sometimes more calculating, depending on how much room it has left to think.

Subject of Research: The role of cognitive load in the foreign language effect on moral decision-making, analyzed with the CNI model

Article Title: The influence of cognitive load on the foreign language effect in moral judgment: an analysis based on the CNI model

Article References: Ting, L., & Xiaolin, Y. (2026). The influence of cognitive load on the foreign language effect in moral judgment: an analysis based on the CNI model. Current Psychology, 45(18), Article 1528. https://doi.org/10.1007/s12144-026-10091-y

Image Credits: AI Generated

DOI: 10.1007/s12144-026-10091-y

Keywords: foreign language effect, moral judgment, CNI model, cognitive load, bilingualism, moral psychology, dual-process theory, utilitarian judgment, psycholinguistics, moral dilemmas, decision-making, Current Psychology

The starting point of the investigation is the observation that lynching is not random chaos. It is a form of collective violence in which civilians aggressively target an individual or group accused of committing a crime or violating an implicit social contract. Scholars have documented lynchings across Latin America and Africa, and Brazilian researchers have recorded hundreds of cases in the metropolitan region of São Paulo alone between 1980 and 2009. With the rise of social media, rumors and calls for popular justice can now spread within minutes, amplifying the risk that an accusation will end in violence before any official investigation takes place. Understanding the psychological profile of people who view such violence favorably is therefore not an academic luxury but a public safety priority.

The theoretical backbone of the study is Lawrence Kohlberg’s influential account of moral development, which distinguishes three levels of moral judgment. At the pre-conventional level, people evaluate actions mainly in terms of punishment and personal gain. At the conventional level, morality is anchored in social approval, law and order, and the maintenance of existing norms. At the post-conventional level, individuals reason from universal principles of justice and human rights that can override both self-interest and the letter of the law. Because moral judgment is inherently ordinal, the researchers treated it as such in their statistical models, using dummy variables to compare the levels while testing the same pattern of results against both reference categories, a robustness check that confirmed the findings.

Alongside moral judgment, the team examined empathy, a construct that psychologists have long debated and that can refer to several related but distinct phenomena, from affective sharing of another person’s feelings to cognitive perspective taking and concern for others’ welfare. Previous work has suggested that cognitive empathy and genuine concern, rather than purely emotional contagion, predict sensitivity to injustice suffered by others. Empathy has also been promoted in interventions designed to reduce aggressive behavior, which made it a natural candidate for testing whether the capacity to feel with a potential victim might weaken the link between moral reasoning and support for mob violence.

The central innovation of the study, however, lies in the way it disentangles the motivations that drive favorability toward lynching. The researchers distinguished social motivations, which involve the relational and collective rewards of participating in or endorsing group punishment, from emotional motivations, which involve the anger, outrage and desire for retribution that a brutal act can provoke. Crucially, they also separated two very different categories of offense: heinous crimes, such as acts of extreme violence, and property-related offenses, which cause material harm but rarely the same visceral horror. This design allowed the team to ask whether the psychological machinery behind lynching approval changes depending on what the accused person is alleged to have done.

The results of the structural equation modeling were strikingly asymmetric. For heinous crimes, the influence of moral judgment on favorability toward lynching was transmitted through both channels: social motivations showed a strong mediating effect, with an unstandardized coefficient of 0.469 and a standard error of 0.099, significant at the 0.001 level, while emotional motivations also carried a significant but smaller effect, with a coefficient of 0.169, a standard error of 0.071 and a p-value of 0.017. In other words, when the alleged offense is horrifying, both the pull of the crowd and the heat of moral outrage help translate a particular style of moral reasoning into approval of collective violence.

For property-related offenses, the picture changed dramatically. Here the pathway through social motivations was even stronger, with a coefficient of 0.876 and a standard error of 0.092, highly significant at the 0.001 level, but the emotional pathway vanished entirely: the coefficient for emotional motivations dropped to 0.028 with a standard error of 0.058 and a non-significant p-value of 0.625. This means that when the alleged crime involves property rather than bodily harm, approval of lynching appears to be almost exclusively a social phenomenon, driven by the perceived collective function of punishment rather than by personal emotional arousal. People do not need to feel enraged to endorse a mob in these cases; they only need to believe that the group’s punitive response serves a social purpose.

Perhaps the most sobering finding concerned empathy. Contrary to what many intervention designers might hope, empathy did not moderate the influence of moral judgment on any of the lists of motivations, regardless of the type of crime. Being more empathic did not buffer the relationship between lower levels of moral reasoning and favorability toward lynching. This null result carries practical weight, because it suggests that programs aimed solely at cultivating empathic feelings may be insufficient to reduce support for mob justice. The authors argue instead that the results support interventions that consider the multifactorial nature of favorability to lynching, and they single out the promotion of post-conventional justice thinking as a promising strategy for reducing it.

The implications reach well beyond Brazil. Lynching and other forms of vigilante justice flourish where trust in formal institutions is low and where rumors circulate faster than due process. The study’s findings suggest that the psychological vulnerability to endorsing such violence is structured by the level at which a person reasons about justice. Individuals operating at conventional levels, who equate morality with order and group norms, may be especially receptive to the idea that a crowd punishing a transgressor is doing legitimate moral work, particularly when the alleged offense threatens the community’s sense of safety or property. Shifting that reasoning toward principled commitments to human rights, in which even a suspected criminal retains inviolable dignity, may be the most durable psychological safeguard against participation in or approval of mob violence.

The research, which was approved by the Research Ethics Committee of the Federal University of Paraíba and supported by the Brazilian National Council for Scientific and Technological Development, also makes its data available through the Open Science Framework, allowing other researchers to scrutinize and extend the model. As social media platforms continue to accelerate the spread of accusations and the mobilization of crowds, the study offers a technically grounded warning: the fight against lynching will not be won by appeals to feeling alone. It requires cultivating a deeper form of moral reasoning, one in which justice is defined by universal principles rather than by the anger of the crowd or the comfort of belonging to it.

Subject of Research: Psychosocial mechanisms underlying favorability toward lynching

Article Title: Psychosocial Mechanisms of Favorability to Lynching

Article References: de Alexandre, M. E. S., Camino, C. P. D. S., Do Bú, E., Galvão, L. K. D. S., & Pereira, C. R. (2025). Psychosocial Mechanisms of Favorability to Lynching. Trends in Psychology. https://doi.org/10.1007/s43076-025-00502-7

Image Credits: AI Generated

DOI: 10.1007/s43076-025-00502-7

Keywords: lynching, moral judgment, empathy, collective violence, social motivations, emotional motivations, Kohlberg, Brazil, vigilante justice, structural equation modeling, moral psychology, social psychology

To answer that question, Kang ran an unusually large experiment: more than 1.5 million evaluation runs across three comparably sized language models, covering 7,591 scenarios drawn from three validated moral-judgment benchmarks, ETHICS, Scruples, and the Moral Machine dataset, plus an expanded set of 42 nuclear-crisis scenarios. The key metric is benchmark alignment, defined as the agreement between a collective’s confidence-weighted majority decision and the human-majority reference label on each scenario. The author is careful to frame this as an empirical proxy for aggregated human moral judgment, not as a direct measure of anything as grand as the common good. Still, the metric offers a rigorous way to test whether groups of machines converge on what most people consider right.

Four main findings emerge from the data. First, the Tachikoma effect is real but small: pooling independent agents improves alignment with human majorities by at most four to five percentage points, with an effect size of roughly eta-squared 0.002, and the benefit varies by model. Second, when the analysis was re-estimated using scenario-clustered statistical models that correct for the non-independence of repeated scenarios, the effect looked less like a Condorcet-style gain from group size and more like ordinary variance reduction over a highly correlated ensemble. The mean inter-agent error correlation was about 0.8, meaning the agents tend to make the same mistakes at the same time. When members of a crowd err together, adding more members buys you far less than classical jury theorems would predict.

Third, and perhaps most striking for anyone hoping that diversity is the secret ingredient, a fixed-group-size homogeneity ablation showed that prompted perspectival diversity, instructing agents to adopt different viewpoints, pushes results in the expected direction but only weakly, reaching statistical significance in just one of the three models. Sharing memory between agents did not help at all. Shared memory raised consensus among the agents without improving accuracy, a pattern that echoes classic findings on informational cascades and groupthink in human groups, where communication can homogenize opinion without making it more correct. The practical implication is counterintuitive for the multi-agent AI community: letting agents deliberate and share information appears to add conformity, not wisdom.

The study’s fourth finding moves from ethics benchmarks to geopolitics. On the expanded set of 42 nuclear-crisis scenarios, multi-agent groups significantly de-escalated simulated crises for two of the three models tested. This result connects to a growing literature on escalation risks from language models in military and diplomatic decision-making, including work from researchers at SIPRI and elsewhere warning that frontier models can exhibit sophisticated but unpredictable reasoning in simulated conflicts. If collective architectures genuinely dampen escalatory tendencies, that would be a safety-relevant dividend of aggregation, though the effect was not uniform across models and the scenarios remain simulations rather than real command-and-control environments.

Underlying all of this is a methodological point that matters for how such results should be read. Much of the apparent benefit of collectives evaporates or shrinks once the non-independence of benchmark scenarios is properly modeled. Because the same scenarios are evaluated many times across agent counts and configurations, treating each evaluation as an independent observation inflates statistical confidence. By re-estimating with scenario-clustered models, the study shows that the Tachikoma effect is best understood as aggregation and variance reduction over a correlated ensemble, not as a magical property of group size. This is a cautionary tale for the broader field of multi-agent LLM research, where dramatic claims about debate and deliberation improving reasoning have sometimes rested on fragile statistical footing.

But the most unsettling finding concerns what alignment costs. Collectives reliably track the human majority, yet they systematically under-represent legitimate human disagreement. On scenarios where humans are nearly evenly split, the AI collectives returned unanimous verdicts 73 to 87 percent of the time. In other words, the very mechanism that makes groups look aligned, confidence-weighted majority voting, collapses genuine moral disagreement into false consensus. Benchmark alignment improves partly by erasing minority positions. A system that reports a confident unanimous verdict on an issue where humans are 50-50 is not capturing collective wisdom; it is manufacturing certainty that does not exist. Supplementary analyses reinforce this: alignment tracks human consensus strength monotonically, and a 20.6 percentage-point alignment gap between controversial and non-controversial Scruples scenarios persists across all agent counts, confirming that collective architecture cannot resolve genuine moral ambiguity.

The study also found that alignment-optimized training amplifies social responsiveness, meaning models tuned with human feedback to be more agreeable and socially attuned respond more strongly to the collective setting. This interacts with known tendencies toward sycophancy in language models, where training can make systems overly eager to mirror perceived human preferences. In a multi-agent context, such social responsiveness may further push agents toward consensus, compounding the false-consensus problem. The author’s recommendation follows directly: deployed systems should aggregate independent judgments while preserving and reporting disagreement, rather than presenting a single confident group verdict that hides the distribution of views beneath it.

Why does the anime metaphor fit? The Tachikoma of Ghost in the Shell are beloved precisely because they combine independence with periodic synchronization, developing quirky individual perspectives through divergent experience while occasionally pooling what they have learned. The study’s results suggest the fiction got the architecture partly right and partly wrong. Independence and divergent experience do contribute something, but the synchronization step, the analogue of shared memory and deliberation, adds consensus without adding accuracy. The wisdom, such as it is, comes from the vote of independent minds, not from the conversation between them. For engineers designing multi-agent systems for morally consequential domains, from content moderation to medical triage to crisis diplomacy, the lesson is to keep agents independent, aggregate their confidence-weighted votes, and surface the disagreements rather than smoothing them away.

The broader significance of the work lies in reframing. Collective moral behavior in language models is not primarily a problem of information sharing, deliberation, or emergent group intelligence. It is a problem of robust, disagreement-preserving aggregation, a question social choice theory has grappled with since Condorcet, and one that thinkers from Amartya Sen to Cass Sunstein have shown is fraught when diversity of opinion is treated as noise rather than signal. As large language models are increasingly deployed in morally consequential domains, the temptation will be to tune them until they agree with the majority on everything. This study warns that doing so would trade away something essential: the honest representation of a pluralistic human moral landscape, in which reasonable people, and reasonable machines, sometimes disagree.

Subject of Research: Collective moral decision-making in multi-agent large language model systems

Article Title: The Tachikoma effect: aggregation of independent agents, not information sharing, aligns LLM collectives with majority human moral judgment

Article References: Kang, J. C. (2026). The Tachikoma effect: aggregation of independent agents, not information sharing, aligns LLM collectives with majority human moral judgment. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03346-6

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03346-6

Keywords: large language models, multi-agent systems, moral judgment, AI alignment, wisdom of crowds, collective decision-making, ensemble aggregation, memory sharing, nuclear crisis simulation, false consensus, benchmark alignment, perspectival diversity

The research addresses a long-standing puzzle in the science of cooperation. Human societies sustain remarkably high levels of collaboration among genetically unrelated individuals, and one of the most influential theoretical explanations is indirect reciprocity. Under indirect reciprocity, people decide whether to help someone based on that person’s reputation: those known to help others earn a good standing, which makes third parties more willing to assist them in the future, while defectors acquire bad standing and are shunned. Because reputational information circulates socially, an individual can benefit from a good reputation even with people they have never met. Decades of theoretical modeling and behavioral experiments have explored how such reputation-based systems can stabilize cooperation, but most of this work has focused on socially shared information—what everyone in a group supposedly knows about everyone else.

Everyday life, however, rarely presents reputation and experience in neat agreement. A coworker may be widely praised yet have treated you badly, or a neighbor may carry a poor reputation despite having personally helped you when you needed it. Yamamoto and his colleagues set out to examine precisely this conflict: how do people evaluate another person’s decision to cooperate or refuse when the public standing of the individual involved contradicts what the evaluator has directly experienced? The question matters because evaluations are the currency of indirect reciprocity. If people’s judgments are swayed by private history rather than shared reputation, the mechanics of reputation-based cooperation may be more complicated than standard models assume.

To investigate, the team conducted two scenario-based experiments with participants in Japan. The scenarios were deliberately grounded in familiar social settings: one involved coworkers at a restaurant, and the other involved neighbors in a residential community. In each scenario, participants read about a situation in which one person either cooperated with or refused a request made by another person. The researchers systematically varied two factors. The first was the reputation of the person making the request—whether that individual was publicly regarded as good or bad. The second was the participant’s own prior experience with the person responding to the request—whether that responder had previously cooperated with or refused to cooperate with the participant personally. Participants then evaluated the person who had responded, allowing the researchers to measure how reputation and personal experience each shaped moral judgment.

The results revealed a striking asymmetry between cooperation and non-cooperation. When the responder refused the request, evaluations of that refusal depended heavily on the participant’s own prior experience with the responder. Even when the responder carried a bad public reputation, the refusal was judged more negatively if that responder had previously helped the participant. Conversely, even when the responder enjoyed a good reputation, the refusal was judged more positively if that responder had previously refused to help the participant. In other words, when public reputation and personal experience pointed in opposite directions, it was the private, firsthand history that strongly colored how participants rated the act of non-cooperation. The shared social record, which much of the indirect reciprocity literature treats as the decisive input, was effectively pushed aside.

Cooperative behavior told a different story. Acts of cooperation were generally evaluated positively regardless of the other person’s reputation and regardless of how that person had previously treated the participant. This suggests that the human evaluative system does not process cooperation and defection symmetrically. Cooperation appears to be rewarded almost automatically, insulated from the evaluator’s private grievances and from the target’s public standing. Defection, by contrast, invites a more contextualized judgment—one in which the evaluator asks not simply whether the refusal was socially acceptable, but how the person on the receiving end had previously treated them personally.

Perhaps the most sobering finding concerned a scenario in which refusing seemed entirely justified. When the person being refused had both a bad public reputation and a personal history of refusing to cooperate with the participant—conditions under which there appeared to be good reason to withhold help—the refusal itself was still evaluated neutrally rather than positively. Refusing, in other words, was not converted into a virtuous act even when the circumstances seemed to warrant it. This indicates that while people may tolerate or excuse non-cooperation under certain conditions, they do not necessarily celebrate it as positively good. The moral ledger of everyday cooperation appears to reserve its positive marks for helping, with refusal occupying at best a neutral zone even when justified.

From a technical perspective, the findings challenge a simplifying assumption embedded in many formal models of indirect reciprocity. Standard models typically treat reputational information as a single, socially shared signal that determines who deserves help. The new results suggest that human evaluators integrate at least two distinct channels—socially transmitted reputation and privately accumulated experience—and that the weighting between these channels depends on the type of behavior being judged. For defection, the private channel dominates; for cooperation, the evaluation is largely insensitive to either channel. This kind of asymmetric integration could have important consequences for how reputation-based cooperation actually operates in groups, because it implies that shared reputational records may be less powerful in governing sanctions and forgiveness than theorists have often assumed. It also hints at why personal grudges and personal loyalties can persist within communities even when everyone has access to the same public information about one another.

The study also clarifies the scope of previous research. Earlier work on indirect reciprocity concentrated mainly on how socially shared reputational information guides decisions about whom to cooperate with, leaving open the question of what happens when that shared information conflicts with lived experience. By manipulating reputation and personal experience independently within controlled scenarios, Yamamoto and his colleagues were able to isolate the contribution of each source and demonstrate that direct experience is not merely one input among many but can override reputation altogether in judgments of defection. The use of two distinct social contexts—workplace and neighborhood—suggests the pattern is not confined to a single type of relationship, although the scenarios were hypothetical and conducted with participants in Japan, so further work will be needed to test how general the effect is across cultures and real-world interactions.

Professor Yamamoto framed the work as a step toward understanding how people combine heterogeneous information when judging others. “In everyday life, we judge others by combining many different kinds of information, including what we hear about them from others and what we have experienced ourselves,” he said. “Our study provides one clue to understanding how people make judgments when these sources of information do not agree.” He also outlined where the research program is headed next: “In future research, we would like to examine how people use reputational information in environments where direct personal experience is limited, such as online rating systems and AI-mediated judgments.” That direction is increasingly urgent. On digital platforms, ratings and reviews function as public reputations, yet users often have little or no direct experience with the sellers, drivers, or service providers they evaluate. If human judgment naturally privileges firsthand experience where it exists, the design of reputation systems for environments that lack such experience becomes a genuinely open scientific and engineering question—one that this study helps to define.

Subject of Research: How direct personal experience and public reputation interact in human evaluations of cooperation and defection in indirect reciprocity

Article Title: “Public reputation” or “Personal experience”?

Article References: “Public reputation” or “Personal experience”?. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: indirect reciprocity, cooperation, reputation, defection, social evaluation, human behavior, evolutionary psychology, moral judgment, personal experience, scenario experiments, Evolution and Human Behavior, prosocial behavior