In 2023, a team led by Molly Crockett at Princeton published neuroimaging data showing that participants evaluated identical harmful actions as significantly less morally wrong when perpetrated by in-group members—and that this bias appeared in neural activity within 170 milliseconds of stimulus onset, well before conscious deliberation could intervene. The finding crystallized something experimental philosophers have long suspected: our moral perception is not a neutral lens applied equally to all agents, but a deeply partisan faculty shaped by group allegiance before reasoning even begins.

This should unsettle anyone who takes moral philosophy seriously. The dominant frameworks in normative ethics—Kantian universalism, utilitarian impartiality, Rawlsian fairness behind the veil of ignorance—all presuppose that moral agents can, at least in principle, evaluate actions and outcomes without systematic distortion tied to tribal membership. Yet the empirical evidence increasingly suggests that impartiality is not the default we override with bias, but rather an achievement built against the grain of automatic moral cognition.

What follows is an examination of three interlocking bodies of research: the neuroscience and social psychology of automatic group bias in moral evaluation, the psychological mechanisms that enable moral circle expansion beyond parochial boundaries, and the philosophical cost-benefit analysis of whether some degree of in-group partiality is not merely inevitable but defensible. The implications reach from metaethics to AI alignment, from refugee policy to the design of institutions meant to adjudicate between competing moral claims.

The evidence for rapid, automatic in-group favoritism in moral judgment is now substantial and converges across methodologies. Implicit Association Tests consistently reveal that participants associate in-group members with moral virtue and out-group members with moral vice, with effect sizes that remain robust across cultures. But IAT data only scratches the surface. EEG studies by Bernhard and colleagues have demonstrated that the error-related negativity—a neural signature of norm violation detection—is significantly attenuated when in-group members commit transgressions, suggesting that the very mechanism for recognizing moral wrongs is modulated by group identity at the level of automatic processing.

Functional neuroimaging adds further resolution. The medial prefrontal cortex, a region critical for mentalizing and moral evaluation, shows differential activation patterns depending on whether a moral agent is categorized as in-group or out-group. Crucially, this categorization happens rapidly and influences downstream processing in the temporoparietal junction and anterior insula—regions associated with perspective-taking and empathic distress, respectively. The architecture of moral cognition, it appears, is designed to treat group membership as morally relevant information.

Joshua Greene's dual-process model offers a useful interpretive framework here. The automatic, emotion-driven system—what Greene calls the "camera's automatic settings"—encodes group loyalty as a default moral heuristic. This makes evolutionary sense: in ancestral environments, coalition membership was among the strongest predictors of cooperative intent. The problem is that these automatic settings now operate in contexts radically different from those in which they evolved, producing systematic moral distortions when applied to strangers, immigrants, or members of different political parties.

What makes this particularly troubling for moral philosophy is the phenomenological invisibility of the bias. Participants in these studies do not experience themselves as applying different moral standards to in-group and out-group members. They experience themselves as perceiving moral reality accurately. The bias operates not at the level of explicit reasoning—where it might be identified and corrected—but at the level of moral perception itself, shaping which actions register as harmful, which suffering registers as salient, and which agents register as morally considerable.

Recent work by Jay Van Bavel's team at NYU has shown that even minimal group assignments—arbitrary team memberships created in the lab minutes before testing—are sufficient to produce measurable biases in moral evaluation. This rules out the hypothesis that in-group moral favoritism requires deep cultural conditioning or shared history. The machinery of moral parochialism is readily activated, alarmingly flexible in what counts as an in-group, and largely opaque to introspection.

Takeaway

Moral perception is not a neutral faculty that bias occasionally distorts—it is a partisan faculty that impartiality must be actively engineered to override, and recognizing this changes what we should expect from moral agents and the institutions we design for them.

If automatic group bias is the default, the historically remarkable fact is that moral circles have expanded at all. Peter Singer's famous metaphor of the "expanding circle" describes a trajectory from kin to tribe to nation to species to sentient beings—but the metaphor obscures the cognitive mechanisms that make expansion possible and the psychological costs it imposes. Recent experimental work has begun to identify these mechanisms with some precision.

One primary pathway is what psychologists call individuation—the process of perceiving out-group members as distinct individuals rather than interchangeable exemplars of a category. Neuroimaging studies by Susan Fiske and colleagues demonstrate that when participants are prompted to individuate out-group members (by asking about their food preferences, for instance), activation in the medial prefrontal cortex increases to levels normally reserved for in-group processing. The moral perception system can be "tricked" into treating out-group members as morally relevant—but it requires deliberate cognitive effort that disrupts the default categorical processing.

A second mechanism involves narrative transportation—the capacity of stories to simulate the phenomenological experience of out-group members. Dan Batson's extensive program of research on empathic concern has shown that perspective-taking exercises produce measurable increases in moral concern for out-group members, with effects that persist over weeks. However, these effects are fragile and context-dependent. Keith Payne's research demonstrates that cognitive load, stress, and time pressure all collapse expanded moral circles back to parochial defaults, suggesting that moral circle expansion depends on cognitive resources that are not always available.

The third and perhaps most significant mechanism is institutional scaffolding. Historical moral circle expansion correlates less with changes in individual psychology than with the development of institutions—legal systems, international organizations, human rights frameworks—that enforce impartial treatment regardless of individual moral motivation. This aligns with Greene's argument that utilitarian reasoning functions as a "manual mode" that can override automatic parochial settings, but it adds the crucial insight that manual mode cannot be sustained by individual willpower alone. It requires structural support.

Importantly, recent work by Adam Waytz and Liane Young suggests that moral circle expansion is not a monotonic process but involves genuine trade-offs. Using their Moral Expansiveness Scale, they found that individuals who score high on concern for distant others—animals, the environment, future generations—tend to score lower on measures of loyalty and care toward close others. The moral circle, in other words, may not simply expand; it may redistribute finite empathic resources, raising hard questions about whether universal moral concern is psychologically sustainable without institutional augmentation.

Takeaway

Moral circle expansion is real but cognitively expensive and structurally dependent—it requires not just better individuals but better institutions, because the psychological machinery of impartiality runs on resources that deplete under pressure.

The empirical findings above set the stage for a normative question that experimental philosophy is uniquely positioned to inform: Is impartiality the correct ideal, or does the psychology of moral cognition reveal something defensible about partiality? The philosophical debate between impartialists (who hold that equal moral consideration is owed to all) and partialists (who defend special obligations to those with whom we share relationships or group membership) has traditionally proceeded through thought experiments. Empirical data now allows us to reframe it.

Bernard Williams's classic objection to utilitarianism—that a man who must deliberate impartially before saving his drowning wife has "one thought too many"—gains neurological grounding from recent research. Studies by Pascual-Leone and colleagues show that participants forced to deliberate impartially in contexts involving close others exhibit increased activation in the dorsolateral prefrontal cortex (associated with cognitive control and override of automatic responses) and report significantly higher subjective distress. Impartiality in personal contexts, the data suggest, is not merely counterintuitive but psychologically costly in ways that may undermine the relational goods—trust, attachment, commitment—that make moral communities possible.

Yet the partialist position faces its own empirical challenges. The same automatic mechanisms that produce defensible loyalty to family and friends also produce indefensible discrimination against racial out-groups, refugees, and future generations who bear the costs of present-day climate policy. The psychological machinery does not distinguish between partiality that sustains valuable relationships and partiality that perpetuates injustice. This is what we might call the partiality indiscriminacy problem: the cognitive systems underlying group loyalty are blunt instruments incapable of the fine-grained moral distinctions that any defensible partialist theory requires.

Greene's resolution—that automatic parochial morality is appropriate for navigating personal relationships while utilitarian "manual mode" reasoning is necessary for intergroup and policy-level dilemmas—has gained significant empirical support but remains philosophically contested. Critics like Jesse Prinz argue that Greene's framework smuggles in a utilitarian normative commitment under the guise of descriptive cognitive science. The dual-process architecture of moral cognition does not, by itself, tell us which process produces correct moral judgments. It merely tells us which process is operative in which conditions.

For practical domains—AI alignment, institutional design, public policy—the most productive path forward may be a context-sensitive pluralism that neither valorizes partiality as natural nor demands impartiality as unconditional. The empirical evidence supports designing systems that preserve space for legitimate partiality in personal contexts while building institutional guardrails that prevent parochial bias from distorting decisions with broad moral consequences. This is less a philosophical resolution than an engineering challenge: calibrating the boundary between domains where group loyalty serves moral purposes and domains where it undermines them.

Takeaway

The deepest challenge is not choosing between partiality and impartiality but recognizing that the same psychological mechanisms produce both defensible loyalty and indefensible prejudice—and that only external structures, not internal moral intuition, can reliably distinguish between them.

The research surveyed here converges on an uncomfortable thesis: moral perception is constitutively shaped by group identity in ways that precede and constrain rational deliberation. This is not a peripheral finding that can be patched over with better arguments or appeals to fairness. It strikes at foundational assumptions in normative ethics about the accessibility of impartial moral judgment.

Yet the same research reveals that moral circles have expanded—through individuation, narrative, and above all through institutions that externalize the cognitive work of impartiality. The challenge is not to eliminate parochialism, which may be psychologically impossible and in personal contexts undesirable, but to build structures that compensate for it where the moral stakes are highest.

For those working in AI ethics, institutional design, or policy, the implication is direct: systems that assume impartial moral agents will fail. Systems that account for the architecture of moral cognition—its speed, its partisanship, its resource dependence—stand a chance of producing outcomes that our reflective moral commitments can endorse, even when our automatic moral perceptions cannot.