Cognitive science of religion

The cognitive science of religion (CSR) is an interdisciplinary research program that applies theories and methods from cognitive psychology, evolutionary biology, and anthropology to explain why human beings form religious beliefs. Rather than asking whether any religion is true, CSR asks a prior question: why are religious beliefs so widespread across cultures, so persistent across history, and so readily acquired by children? The field answers by identifying specific cognitive mechanisms — agency detection, theory of mind, teleological reasoning, memory biases favoring certain kinds of concepts — that, operating together, make religious belief a natural and predictable output of the ordinary human mind. The field emerged in the early 1990s through the converging work of Pascal Boyer, Scott Atran, Stewart Guthrie, and others, and has since grown into a substantial body of empirical and theoretical research spanning developmental psychology, neuroscience, anthropology, and philosophy.^{1, 2, 3, 7}

CSR draws a deliberate distinction between explaining the origins and persistence of religious belief and evaluating whether religious beliefs are true. The mechanisms it identifies could, in principle, produce true beliefs, false beliefs, or some mixture of both. The field’s contribution is to show that the existence of widespread religious belief does not, by itself, require a supernatural explanation: the cognitive architecture of the human mind is sufficient to generate religious concepts even in the absence of any divine reality. What follows from this observation — whether it undermines the epistemic standing of religious belief, is neutral toward it, or is even compatible with theistic design — is a philosophical question that CSR itself does not settle.^{7, 15, 16}

Hyperactive agency detection

The most foundational mechanism identified by CSR is the hyperactive agency detection device (HADD), a term introduced by Justin Barrett building on Stewart Guthrie’s earlier work on anthropomorphism. The core insight rests on asymmetric error costs in the ancestral environment. For an organism navigating a world of predators, competitors, and potential mates, the cost of failing to detect a real agent — a predator concealed in tall grass, a rival lurking behind a rock — was dramatically higher than the cost of falsely detecting an agent that was not there. A hominid who mistook wind for a predator wasted a moment of vigilance; a hominid who mistook a predator for wind became a meal. Natural selection therefore favored a cognitive system biased toward overdetection of agency, one that systematically produces false positives (perceiving agents where there are none) rather than false negatives (failing to perceive agents where they are).^{3, 4}

Guthrie argued in Faces in the Clouds (1993) that this bias is the cognitive root of religious belief. Humans perceive intentional agency in natural phenomena — storms, earthquakes, disease, patterns in the night sky — because their cognitive systems are calibrated to detect agents on minimal or ambiguous evidence. The resulting “agents” are invisible, powerful, and purposive: they cause events that have no visible agent, they operate on scales that exceed human capability, and they appear to direct natural processes toward outcomes. These are precisely the attributes assigned to gods, spirits, and ancestors across human cultures. Religion, on Guthrie’s account, is systematic anthropomorphism — the projection of human-like agency onto a non-human world.^{3, 7}

Barrett extended this analysis by showing that HADD does not operate in isolation. Once the device posits an unseen agent behind an unexplained event, other cognitive systems engage automatically. Theory of mind attributes beliefs, desires, and intentions to the detected agent. Causal reasoning constructs narratives about why the agent acted. Memory systems encode the resulting concept and its associated narrative. The output is not a vague sense of “something out there” but a fully elaborated representation of a purposive, psychologically complex agent — a mind without a body, acting on the world for reasons. This, Barrett argued, is the cognitive foundation of theistic belief: the intuitive perception of an intentional mind behind the events of the natural world.^{4, 14}

Theory of mind and supernatural agents

Theory of mind (ToM) is the cognitive capacity to attribute mental states — beliefs, desires, intentions, knowledge, emotions — to other agents. It develops early in human ontogeny, typically emerging between ages three and five, and is fundamental to social cognition. ToM enables cooperation, deception, empathy, moral judgment, and the prediction of others’ behavior. CSR researchers have argued that this same faculty is the mechanism that allows humans to sustain rich, psychologically detailed beliefs about invisible supernatural agents over time.^{1, 7}

Boyer (2001) argued that the human mind is equipped with an inference system for agents that automatically generates expectations about any entity categorized as possessing a mind. Once an entity is tagged as an agent — whether by HADD, cultural transmission, or direct perceptual experience — the inference system produces default assumptions: the agent has a perspective, possesses information, can act on the world, has goals, and can be pleased or displeased. These inferences are automatic and largely unconscious. A child told that God watches over her does not need instruction in the implications of this claim; the inference system generates them immediately: God has beliefs about her behavior, God has preferences about what she does, God may respond emotionally to her actions.^{1, 5}

Experimental studies by Barrett and others have revealed a striking divergence between theological concepts of God and the cognitive representations people actually deploy when reasoning about divine action. Theologically sophisticated adults who affirm that God is omniscient, timeless, and incorporeal nevertheless process narratives about God using the same cognitive architecture they use for narratives about human persons — attributing to God a temporal perspective, limited attention, and sequential processing of information. Barrett described this as the difference between “theological correctness” (the explicit doctrines a person endorses) and the intuitive “online” God concept that operates in real-time cognition. The theological God of systematic theology and the cognitive God of intuitive religious thought are not identical: the cognitive God is a person-like agent generated by ToM, while the theological God is an abstract construction maintained by effortful, reflective reasoning that works against the grain of natural cognition.^{4, 14}

Jesse Bering (2006) extended this research by investigating whether humans have a default cognitive bias toward attributing mental states to the dead. In experiments with children and adults, Bering found that even participants who explicitly denied an afterlife continued to attribute psychological states — emotions, desires, epistemic states — to deceased persons when reasoning about specific scenarios. He interpreted this as evidence that the human ToM system does not have a natural “off switch” for agents who have died: the mind continues to generate inferences about the mental states of persons it has modeled, even after those persons no longer exist. This cognitive persistence, Bering argued, provides a natural foundation for afterlife beliefs and ancestor veneration across cultures.²³

Minimally counterintuitive concepts

Pascal Boyer’s theory of minimally counterintuitive (MCI) concepts addresses a different question: not why humans detect agents, but why certain religious concepts spread and persist while others do not. Boyer argued that human cognition operates with a set of intuitive ontological categories — person, animal, plant, natural object, artifact — each of which comes with a package of default expectations. Persons are expected to have bodies, be born and die, have limited knowledge, and occupy a single location. Plants are expected to grow, lack locomotion, and lack mental states. A minimally counterintuitive concept is one that violates a small number of these default expectations while otherwise conforming to the category template. A ghost is a person without a body. A talking tree is a plant with the capacity for speech. An omniscient god is a person without the limitation of finite knowledge.^{1, 6}

Boyer predicted, and experimental research has confirmed, that MCI concepts enjoy a mnemonic advantage over both fully intuitive concepts (which are unmemorable because they are unremarkable) and maximally counterintuitive concepts (which violate so many expectations that they become cognitively unprocessable and are quickly forgotten). The sweet spot for cultural transmission is minimal counterintuitiveness: one or two violations of default assumptions embedded in an otherwise normal conceptual structure. Religious concepts — gods, spirits, enchanted objects, talking animals in sacred narratives — overwhelmingly occupy this sweet spot. They are memorable precisely because they are slightly surprising, and they are processable precisely because they are mostly normal. Narratives containing a small proportion of MCI elements are better recalled and more readily transmitted than narratives containing none or many such elements.^{1, 6, 7}

This framework explains a recurrent pattern in comparative religion: the gods and spirits of actual religious traditions are far more anthropomorphic and cognitively tractable than the God of philosophical theology. The God of Anselm or Aquinas — a being with no body, no temporal location, no composition, and no limitation of any kind — is maximally counterintuitive and cognitively difficult to sustain. The God of popular devotion, who listens to prayers, responds to petitions, gets angry, feels love, and acts in history, is minimally counterintuitive: a person with a few extraordinary properties. Boyer’s theory predicts that the latter concept will always be more cognitively natural, more culturally widespread, and more psychologically compelling than the former — a prediction that matches the observed distribution of religious concepts across cultures.^{1, 15}

Teleological reasoning and promiscuous teleology

Deborah Kelemen’s research on teleological reasoning has identified another cognitive bias that contributes to religious thought: the tendency to explain natural objects and events in terms of purposes. Teleological reasoning is the attribution of function or goal-directedness to things — explaining that rain exists “so that plants can grow” or that rocks are pointy “so that animals can scratch themselves.” While teleological explanations are appropriate for artifacts (a chair exists for sitting) and biological traits shaped by natural selection (wings exist for flying, in a restricted sense), extending them to nonbiological natural phenomena constitutes what Kelemen calls “promiscuous teleology.”^{9, 10}

In a series of developmental studies, Kelemen (2004) found that children between the ages of four and ten overwhelmingly prefer teleological explanations for natural objects and phenomena, even when physical-causal explanations are available. Asked why mountains exist, young children consistently choose answers like “so that animals have a place to climb” over answers like “because of the way the earth’s plates push up.” Kelemen described children as “intuitive theists” — not because they have explicit theological commitments, but because their default explanatory framework treats the natural world as purposefully designed. The inference from “everything exists for a purpose” to “someone must have purposed it” is cognitively natural, and the agent most readily recruited to fill the designer role is the kind of powerful, intentional being that religious traditions describe.⁹

Critically, promiscuous teleology is not confined to children. Kelemen and Rosset (2009) demonstrated that adults under cognitive load — when forced to answer quickly, when distracted, or when mentally fatigued — revert to teleological explanations for natural phenomena that they would reject under reflective conditions. Even professional scientists accepted scientifically unwarranted teleological statements (such as “the sun makes light so that plants can photosynthesize”) at significantly elevated rates when placed under time pressure. This suggests that promiscuous teleology is not simply a developmental stage that education eliminates but a deep cognitive default that persists into adulthood and reasserts itself whenever reflective override is weakened.^{10, 14}

Dual-process theory and intuitive versus reflective belief

The distinction between intuitive and reflective cognition is central to CSR’s account of religious belief and draws on the broader dual-process framework in cognitive psychology. Dual-process theories, as synthesized by Evans and Stanovich (2013), distinguish between Type 1 (fast, automatic, effortless, associative) and Type 2 (slow, deliberate, effortful, rule-based) cognitive processes. Type 1 processes operate by default and generate intuitive judgments; Type 2 processes can override these judgments but require cognitive resources and motivation to do so.^{21, 22}

Robert McCauley (2011) argued in Why Religion Is Natural and Science Is Not that religious cognition is primarily a product of Type 1 processing while scientific reasoning depends heavily on Type 2 processing. The cognitive mechanisms CSR identifies — HADD, ToM applied to invisible agents, promiscuous teleology, MCI concept formation — are all Type 1 processes: fast, automatic, developmentally early, and cross-culturally universal. They produce religious intuitions without effort or instruction. Scientific thinking, by contrast, requires sustained Type 2 effort: suppressing intuitive teleological explanations, resisting the temptation to see agency in natural events, overriding the default assumption that minds require bodies. This asymmetry explains why religion is culturally universal while science is culturally rare, why children acquire religious concepts easily but struggle with counterintuitive scientific ideas like natural selection, and why religious belief persists even among scientifically educated populations.¹⁴

The dual-process framework also illuminates the phenomenon Barrett identified as the gap between theological correctness and online religious cognition. Explicit theological doctrines — God is omniscient, incorporeal, timeless, and omnipresent — are products of reflective Type 2 reasoning. But when believers reason about God in real time, they default to Type 1 processing, which generates a person-like God with limited attention and a temporal perspective. Reflective theology can override intuitive theology, but the override is fragile: it requires cognitive effort and collapses under load, stress, or distraction. The result is that most religious believers operate with two parallel God concepts, one reflective and doctrinally correct, the other intuitive and anthropomorphic, and the intuitive concept dominates in everyday religious life.^{4, 14, 21}

Ritual and costly signaling theory

A complementary strand of CSR focuses not on the cognitive mechanisms that generate religious belief but on the social mechanisms that sustain religious commitment and practice. The costly signaling theory of religion, developed by Richard Sosis, Joseph Bulbulia, and others, argues that religious rituals and behavioral requirements function as honest signals of group loyalty. Behaviors such as fasting, tithing, pilgrimage, dietary restrictions, painful initiation rites, and time-consuming prayer schedules are costly: they demand the expenditure of time, energy, resources, and sometimes physical well-being that could be devoted to other purposes. Ritual theorists have argued that such practices create and maintain a shared subjunctive world — a domain of “as if” that binds participants into common frameworks of meaning regardless of individual sincerity.¹⁹ Precisely because they are costly, they are difficult to fake. A free rider who does not genuinely share the group’s commitments will be reluctant to bear these costs, making costly behavior a reliable indicator of sincere in-group membership.^{18, 11}

Joseph Henrich (2009) introduced the concept of credibility enhancing displays (CREDs) to capture a related transmission mechanism. CREDs are behaviors that demonstrate a person’s commitment to the beliefs they verbally profess. People are more likely to adopt the beliefs of models whose actions are consistent with their stated beliefs, especially when those actions are costly. A parent who merely tells a child that God exists is less persuasive than a parent who demonstrates that belief through regular worship attendance, dietary restrictions, financial sacrifice, and emotional investment in religious practices. CREDs provide a mechanism for the intergenerational transmission of religious belief that does not depend on the truth of the beliefs being transmitted; it depends only on the observable consistency between a model’s stated beliefs and costly behaviors.¹¹

David Sloan Wilson (2002) argued in Darwin’s Cathedral that religion functions as a group-level adaptation that enhances cooperation and coordination within religious communities. Religious beliefs provide shared narratives, moral codes, and institutional structures that solve collective action problems, enabling groups to cooperate on scales that would otherwise be impossible. Ara Norenzayan (2013) extended this analysis in Big Gods, arguing that belief in moralizing high gods — gods who monitor human behavior and punish defectors — was a critical cultural innovation that enabled the scaling of human cooperation from small kin-based bands to large, anonymous societies. On this view, religions that featured omniscient, morally concerned gods outcompeted religions that did not, not because those gods actually existed but because belief in such gods promoted prosocial behavior and group cohesion.^{13, 12}

The byproduct versus adaptation debate

The central theoretical debate within CSR concerns whether religious belief is an evolutionary byproduct or a direct adaptation. The byproduct hypothesis, associated primarily with Boyer, Atran, Barrett, and Guthrie, holds that religious cognition is a side effect of cognitive systems that evolved for non-religious purposes. Agency detection evolved because detecting predators and prey was adaptive. Theory of mind evolved because navigating social relationships required modeling the mental states of conspecifics. Teleological reasoning evolved because identifying the functions of tools and environmental features was useful. These systems, operating together, incidentally produce religious concepts as a byproduct — much as moths are drawn to artificial lights not because light-seeking is adaptive in itself but because the navigational system that uses moonlight for orientation misfires in the presence of novel light sources.^{1, 2, 7}

The adaptationist hypothesis, associated with Wilson, Sosis, Bulbulia, and to some extent Norenzayan, holds that religion itself was selected for — either at the individual level (because religious individuals enjoyed fitness advantages through group membership, social support, and reduced anxiety) or at the group level (because religious groups outcompeted non-religious groups through superior cooperation). On this view, the cognitive mechanisms CSR identifies are not merely incidental producers of religious concepts; they were shaped or co-opted by natural selection specifically because of the fitness advantages that religious belief and practice conferred.^{13, 18, 20}

Key cognitive mechanisms and their proposed roles^{1, 7, 8}

The byproduct and adaptationist positions are not mutually exclusive, and many researchers hold intermediate views. It is possible that religious cognition originated as a byproduct and was subsequently co-opted by cultural or genetic selection for its group-beneficial effects. Boyer (2001) acknowledged that religion may have social consequences that are selected for even if the cognitive mechanisms that produce it were not originally selected for religious functions. The empirical question — whether any genetic or cultural selection pressure has specifically shaped the cognitive architecture of religion, as distinct from the more general cognitive systems from which religion emerges — remains open.^{1, 16, 20}

Cross-cultural findings

A central claim of CSR is that the cognitive foundations of religious belief are universal features of human cognition rather than products of specific cultural traditions. This claim has been tested through cross-cultural research spanning societies with very different religious traditions, levels of technological development, and degrees of exposure to organized religion. The results broadly support the universality thesis, though with important qualifications.^{8, 20}

Agency detection biases have been documented across a wide range of cultures. People in societies as different as urban North America, rural India, indigenous Amazonia, and East Africa show similar tendencies to attribute ambiguous events to intentional agents. Kelemen’s findings on promiscuous teleology have been partially replicated in non-Western samples, with children in diverse cultural settings preferring purpose-based explanations for natural phenomena, though the specific forms of teleological reasoning vary with cultural context and available causal frameworks. Bering’s work on intuitive afterlife beliefs has found cross-cultural support: even in societies without strong doctrinal traditions about the afterlife, individuals tend to attribute ongoing mental states to deceased persons.^{9, 23, 8}

At the same time, cross-cultural research has revealed significant variation in how these cognitive foundations are elaborated into specific religious systems. The universal tendency to detect agency does not produce a universal religion; it produces a universal susceptibility to religious concepts, which is then shaped by local cultural inputs into the specific gods, spirits, rituals, and narratives of particular traditions. Norenzayan’s work on “big gods” has shown that belief in omniscient, moralizing deities is far from universal among small-scale societies — it is concentrated in large, complex societies where monitoring and punishing defectors through face-to-face interaction is impossible. This suggests that while the cognitive raw materials for religion are universal, the specific form that religious belief takes is powerfully shaped by ecological, demographic, and historical factors.^{12, 2}

The cross-cultural evidence also poses challenges for certain versions of CSR. If specific religious concepts (moralizing high gods, afterlife beliefs tied to moral behavior, creation narratives) are culturally variable rather than cognitively determined, then the cognitive mechanisms CSR identifies constrain but do not determine the content of religious belief. The mechanisms explain why humans are disposed to religious thinking in general but do not explain why specific religions take the particular forms they do. Cultural transmission, institutional structures, and historical contingency play irreducible roles that purely cognitive accounts cannot capture.^{2, 8, 20}

What CSR does and does not imply about religious truth

CSR researchers have been generally careful to distinguish between explaining the cognitive origins of religious belief and evaluating whether religious beliefs are true. Explaining why people believe something is logically independent of whether what they believe is correct. A complete causal account of why humans believe the earth is round does not show that the earth is not round; similarly, a complete causal account of why humans believe in God does not show that God does not exist. The genetic fallacy — rejecting a belief solely on the basis of its causal origin — is a genuine logical error, and the leading CSR researchers have typically acknowledged this.^{7, 16}

That said, the relationship between causal explanation and epistemic evaluation is more complex than a simple invocation of the genetic fallacy suggests. If a belief is produced by a mechanism known to be unreliable — one that produces false positives at a high rate — then learning that the mechanism produced a given belief is legitimate evidence for reducing confidence in that belief. HADD is known to produce massive numbers of false positives in non-religious contexts: people perceive faces in clouds, attribute intentions to computer animations, and detect agency in random noise patterns. A mechanism that systematically overdetects agents is precisely the sort of mechanism whose deliverances warrant caution. The question is whether the religious deliverances of HADD are more like its accurate detections of real agents or more like its false detections of phantom ones — and CSR, by itself, cannot answer that question.^{7, 15}

Alvin Plantinga has offered the most prominent theistic engagement with CSR, arguing in Warranted Christian Belief (2000) and Where the Conflict Really Lies (2011) that the cognitive mechanisms CSR identifies are entirely compatible with theism. If God exists and desires human beings to know him, God may have designed the very cognitive faculties that CSR describes — making HADD, ToM, and promiscuous teleology truth-tracking rather than truth-distorting. On this view, the sensus divinitatis that John Calvin posited is simply the cognitive architecture that CSR has now described in naturalistic terms. The fact that we can describe these faculties mechanistically does not show that they were not designed by God, any more than a mechanistic description of the eye shows that the eye was not designed for seeing.^{17, 24}

Critics respond that Plantinga’s argument is available only to those who already accept theism. The sensus divinitatis hypothesis cannot serve as evidence for God’s existence without circularity: it assumes God exists in order to explain why the cognitive mechanisms that produce belief in God are reliable. Furthermore, the specific mechanisms CSR identifies are poor candidates for a divinely designed truth-tracking faculty. A mechanism selected for producing adaptive false positives is precisely not a mechanism designed for accuracy. De Cruz and De Smedt (2015) have argued for a more nuanced view: some cognitive inputs to religious belief (such as the perception of cosmic order or the intuition of moral realism) may be more epistemically respectable than others (such as HADD-generated agent detection), and the philosophical evaluation of religious belief should proceed mechanism by mechanism rather than treating CSR findings as a monolithic block.^{15, 16}

A further worry, pressed by Plantinga and others, is that naturalistic debunking arguments are self-undermining. If evolved cognitive mechanisms are unreliable in the domain of religion, the same evolutionary processes produced all human cognitive mechanisms, including those used in science, philosophy, and the formulation of CSR itself. This is the core of Plantinga’s evolutionary argument against naturalism. CSR proponents respond that the objection proves too much: it would invalidate all of empirical science if taken seriously, since all scientific conclusions are products of evolved cognition. The relevant distinction, they argue, is between cognitive mechanisms that were selected for tracking environmental truths (perception, spatial reasoning, causal inference) and mechanisms that were selected for producing adaptive behavior regardless of truth (HADD, anxiety responses, coalition detection). CSR claims that the cognitive foundations of religion fall predominantly into the second category.^{24, 20, 15}

What CSR establishes with reasonable confidence is that the existence of widespread religious belief does not, by itself, constitute strong evidence for the truth of religious claims. The cognitive architecture of the human mind is sufficient to generate religious concepts whether or not any gods exist. This does not prove that no gods exist; it removes one category of evidence — the sheer prevalence of religious belief — from the theistic side of the ledger. The truth of religious claims must be evaluated on independent grounds: philosophical arguments, empirical evidence, and the internal coherence of theological systems. CSR contributes to this evaluation by clarifying which aspects of religious belief are products of general cognitive mechanisms and which, if any, require explanations that go beyond what cognitive science can provide.^{7, 8, 15, 16}