Bayesian arguments for God

Bayesian arguments for God apply the formal apparatus of probability theory to the question of God’s existence, treating theism and naturalism as competing hypotheses and assessing them against the available evidence using Bayes’s theorem. The central idea is that the existence of God can be evaluated in the same way that any hypothesis is evaluated in science or everyday reasoning: by asking how probable the observed evidence is on the hypothesis that God exists compared with how probable it is on the hypothesis that God does not exist. If the evidence — cosmic fine-tuning, the existence of consciousness, moral awareness, religious experience, the beauty of the natural world — is more probable on theism than on naturalism, then by Bayes’s theorem the evidence raises the posterior probability of theism.^{1, 14}

The Bayesian approach to natural theology has become the dominant framework in analytic philosophy of religion since Richard Swinburne’s pioneering work in the 1970s and 1980s. Its appeal lies in its promise to bring mathematical precision to debates that have often seemed irresolvable by informal reasoning alone. Its critics argue that this promise is illusory — that the appearance of rigour masks unresolved philosophical disagreements about how to assign probabilities to metaphysical hypotheses, and that the framework is only as reliable as the probability judgments fed into it.^{3, 4}

Bayes’s theorem and its application

Bayes’s theorem, first published posthumously by the Reverend Thomas Bayes in 1763, provides a formal rule for updating the probability of a hypothesis in light of new evidence. In its simplest form, the theorem states:

P(H | E) = P(E | H) × P(H) / P(E)

Here P(H | E) is the posterior probability of hypothesis H given evidence E — what we want to know. P(E | H) is the likelihood — how probable the evidence is if the hypothesis is true. P(H) is the prior probability — the probability of the hypothesis before considering the evidence. P(E) is the total probability of the evidence — the probability of observing E whether H is true or false. The theorem shows that the posterior probability of H increases when the likelihood P(E | H) is high relative to the total probability P(E), and when the prior probability P(H) is not negligibly low.^{9, 14}

Applied to the existence of God, the theorem takes the form: P(God | Evidence) = P(Evidence | God) × P(God) / P(Evidence). The key question becomes whether the various pieces of evidence — the existence of the universe, its fine-tuning, the existence of conscious beings, moral awareness, religious experience — are more probable on the hypothesis that God exists than on the hypothesis that God does not exist. If they are, then the cumulative effect of multiple independent pieces of evidence can raise the posterior probability of theism to a high level even if no single piece is decisive.¹

Swinburne’s programme

Richard Swinburne has developed the most comprehensive and technically rigorous Bayesian case for theism, principally in The Existence of God (1979; 2nd ed. 2004) and its preparatory volumes The Coherence of Theism (1977) and An Introduction to Confirmation Theory (1973). Swinburne’s project proceeds in three stages. First, in The Coherence of Theism, he argues that the concept of God — understood as a person without a body who is omnipotent, omniscient, perfectly good, and the creator and sustainer of the universe — is internally coherent, defeating the objection that theism is meaningless or contradictory. Second, in An Introduction to Confirmation Theory, he develops the formal probabilistic framework within which evidence for and against theism will be assessed. Third, in The Existence of God, he evaluates the evidence and argues that it renders theism more probable than not.^{1, 2, 15}

Swinburne evaluates seven principal bodies of evidence, each treated as an independent confirmatory datum: (1) the existence of a universe rather than nothing; (2) the conformity of the universe to simple, elegant natural laws; (3) the fine-tuning of physical constants for the possibility of life; (4) the existence of conscious beings; (5) the existence of moral awareness; (6) the occurrence of miracles and religious experiences; and (7) the beauty of the natural world. For each datum, Swinburne assesses the likelihood — how probable the datum is on theism versus on naturalism — and argues that in every case the theistic likelihood exceeds the naturalistic one.¹

A crucial component of Swinburne’s programme is his argument that the prior probability of theism is not negligibly low. Swinburne contends that the theistic hypothesis is intrinsically simple in the sense relevant to scientific theory choice: God is postulated as a single entity with maximal properties (infinite power, infinite knowledge, perfect goodness), and simplicity is a truth-conducive feature of hypotheses because simpler hypotheses are more likely to be true, all else being equal. Just as scientists prefer simpler theories (Occam’s razor), the intrinsic simplicity of the theistic hypothesis gives it a non-negligible prior probability — perhaps not high, but sufficient to be raised to a probability above 0.5 by the cumulative evidence.^{1, 11}

The likelihood assessments

The core of the Bayesian case for theism consists in arguing that each piece of evidence is more probable on theism than on naturalism. The following summarises Swinburne’s principal likelihood claims.

For the existence of the universe: on theism, a perfectly good God would have reason to create a universe containing good things (conscious beings, beauty, moral agents), making the existence of a universe fairly probable on theism. On naturalism, the existence of the universe is a brute, unexplained fact — it just happens to exist — which Swinburne argues makes its existence less expected than on theism.¹

For fine-tuning: on theism, a God who intends to create life would set the physical constants at life-permitting values, making fine-tuning highly probable. On naturalism (assuming a single universe), the probability that the constants would fall within the exceedingly narrow life-permitting range is astronomically low. Robin Collins has formalised this as the “likelihood principle”: fine-tuning strongly confirms design over chance on a single-universe naturalism, though the multiverse hypothesis complicates the picture.^{1, 7}

For consciousness: on theism, a conscious God would have reason to create conscious beings, making the existence of consciousness expected. On naturalism, consciousness is a deeply puzzling phenomenon — the “hard problem” of consciousness — with no satisfactory naturalistic explanation, making it less expected on naturalism than on theism.^{1, 5}

For moral awareness: on theism, a perfectly good God would have reason to create beings capable of moral knowledge and moral action, making moral awareness expected. On naturalism, moral awareness is typically explained as a product of evolutionary pressures, but Swinburne argues that the existence of objective moral truths — not merely the illusion of morality — is less expected on naturalism than on theism.¹

For religious experience: on theism, God would have reason to make himself experientially accessible to human beings, making the occurrence of religious experiences expected. On naturalism, religious experiences require a neurological or psychological explanation, and the sheer prevalence and cross-cultural consistency of such experiences is, Swinburne argues, more probable on theism.^{1, 12}

Swinburne’s likelihood assessments for key evidential strands¹

The prior probability debate

The most contested element of the Bayesian case for theism is the assignment of prior probabilities. Swinburne argues that theism deserves a non-negligible prior because it is a simple hypothesis — it postulates a single entity with three infinite properties (power, knowledge, goodness) — and simplicity is an epistemically virtuous feature that raises prior probability. On this account, the theistic hypothesis is simpler than many naturalistic alternatives (which must posit a complex, undesigned physical universe with no explanation for its existence or character) and therefore deserves a prior probability at least comparable to that of naturalism.^{1, 11}

J. L. Mackie challenged Swinburne’s simplicity argument, contending that the concept of a person without a body who is infinite in power, knowledge, and goodness is anything but simple. Mackie argued that the theistic hypothesis is extraordinarily complex because it postulates an entity utterly unlike anything in our experience — a disembodied mind with maximal properties — and that the notion of an “infinite” property is conceptually obscure. If the theistic hypothesis is not simple, its prior probability may be very low, and no amount of confirmatory evidence can raise a sufficiently low prior to a probability above 0.5.⁴

Graham Oppy has pressed a related objection: the assessment of “simplicity” is itself a philosophical judgment on which reasonable people disagree, and there is no neutral algorithm for measuring the simplicity of metaphysical hypotheses. A materialist will judge physicalism simpler than theism (one kind of substance rather than two); a theist will judge theism simpler than physicalism (one ultimate explanation rather than a brute collection of physical laws). The dispute over priors thus reduces to a dispute over which ontological framework is more fundamental — a philosophical disagreement that Bayesian formalism cannot resolve.³

Alvin Plantinga has expressed scepticism about the Bayesian approach from a different direction. Plantinga argues that belief in God can be “properly basic” — warranted without being inferred from evidence — and that the demand for a probabilistic argument from evidence reflects an evidentialist epistemology that the reformed epistemologist rejects. On Plantinga’s view, the Bayesian framework may be legitimate but unnecessary: theistic belief can be rational even without a Bayesian calculation supporting it.¹⁰

The problem of evil as counter-evidence

Any comprehensive Bayesian assessment of theism must account for counter-evidence as well as confirmatory evidence. The most significant counter-evidence is the existence and extent of suffering and evil in the world. In Bayesian terms, the problem of evil asserts that the existence of extensive suffering is less probable on theism (a perfectly good, omnipotent God would have reason and power to prevent it) than on naturalism (suffering is an expected consequence of impersonal natural processes), so that suffering lowers the posterior probability of theism.^{1, 4}

Swinburne addresses the problem of evil within his Bayesian framework by arguing that the existence of suffering is not as improbable on theism as it initially appears. A good God would have reason to create a world in which humans have genuine moral freedom, in which actions have real consequences, and in which character can be formed through the experience of hardship — conditions that require the possibility of suffering. Swinburne concedes that suffering lowers the probability of theism to some degree, but argues that the evidential force of the positive evidence (fine-tuning, consciousness, moral awareness, religious experience) outweighs the evidential force of evil, so that the net posterior probability of theism remains above 0.5.¹

Mackie and other critics dispute this balance. They argue that the sheer quantity, intensity, and apparently pointless distribution of suffering in the world is far less probable on theism than Swinburne allows, and that the evidential force of suffering is sufficient to outweigh the cumulative positive evidence. The dispute is ultimately about the magnitude of the relevant likelihood ratios — a question on which Bayesian formalism provides a framework for disagreement but not a resolution of it.^{4, 13}

Alternative Bayesian approaches

Swinburne’s is not the only Bayesian approach to the existence of God. Robin Collins has developed a focused Bayesian argument from fine-tuning, treating the precise values of fundamental physical constants as evidence that strongly confirms design over a single-universe naturalism. Collins distinguishes his approach from Swinburne’s in two respects: he restricts his argument to a single evidential strand (fine-tuning) rather than attempting a comprehensive cumulative case, and he formulates the argument as comparing design with a “naturalistic single-universe hypothesis” rather than with all possible forms of naturalism, thereby avoiding the complication of the multiverse.⁷

Swinburne has also applied Bayesian reasoning to historical claims, most notably in The Resurrection of God Incarnate (2003), where he argues that the posterior probability of Jesus’s resurrection is high given the prior probability of theism (established by the natural theology arguments) and the historical evidence (the empty tomb, the post-mortem appearances, the origin of Christian belief). Critics have objected that this application stretches the Bayesian framework beyond its proper domain, since the assignment of probabilities to unique historical events is even more indeterminate than the assignment of probabilities to metaphysical hypotheses.^{12, 14}

Basil Mitchell’s The Justification of Religious Belief (1973) anticipated some of Swinburne’s themes but adopted a less formally Bayesian and more informal “inference to the best explanation” approach. Mitchell argued that religious belief is justified in the same way that any comprehensive worldview is justified — not by a single decisive argument or calculation, but by the overall explanatory adequacy of the worldview in accounting for the full range of human experience. Mitchell’s approach shares Swinburne’s cumulative structure but avoids the apparatus of prior probabilities and likelihood ratios, relying instead on a qualitative assessment of explanatory power.⁸

Major objections

The Bayesian approach to natural theology has attracted several categories of objection. The most fundamental challenges the applicability of probability theory to metaphysical hypotheses. On a frequency interpretation of probability, probability is a property of repeatable events in a reference class, and the existence of God is not a repeatable event in any reference class. On a logical interpretation, probability reflects the degree to which evidence logically supports a hypothesis, but critics argue that the logical support relation for metaphysical hypotheses is too indeterminate to yield precise probability values. The Bayesian approach requires a subjective (or “personalist”) interpretation of probability, on which probabilities represent rational degrees of belief, but this interpretation makes the resulting probabilities dependent on the reasoner’s prior beliefs and evidential judgments — a feature that critics regard as vitiating the argument’s claim to objectivity.^{3, 14}

A second objection concerns the indeterminacy of theistic likelihoods. Oppy has argued that because we have no independent access to God’s intentions, we cannot determine how probable any particular feature of the world is on the hypothesis that God exists. Would God create a universe with fine-tuned constants? Would God create conscious beings? Would God permit suffering? The answers depend entirely on assumptions about God’s purposes, and since different theistic traditions make different assumptions, the likelihood P(Evidence | God) is not a determinate quantity but a range that can be made to support almost any conclusion. The apparent precision of the Bayesian framework, on this view, is an illusion that conceals the genuine indeterminacy of the underlying probability judgments.³

Keith Parsons has objected that Swinburne’s simplicity argument for a non-negligible prior probability is question-begging. The claim that God is “simple” depends on a particular understanding of simplicity (fewest entities with maximal properties) that a naturalist will reject in favour of a different understanding (fewest kinds of entities, fewest unexplained laws). Since the choice between these conceptions of simplicity is itself a substantive philosophical disagreement, the Bayesian framework cannot resolve the debate but merely relocates it from the level of evidence to the level of priors.¹³

A fourth objection, developed by Hume in the pre-Bayesian era and adapted by modern critics, concerns the “many gods” problem. The Bayesian comparison is typically framed as theism versus naturalism, but there are infinitely many possible supernatural hypotheses (polytheism, deism, pantheism, malevolent deity, simulation hypothesis) that share some of theism’s predictive advantages. If the prior probability must be divided among all these hypotheses, the prior for classical theism specifically may be very low, and the cumulative evidence may not be sufficient to raise it above 0.5.^{3, 16}

Responses to objections

Defenders of the Bayesian approach have offered replies to each of these objections. Against the applicability objection, Swinburne argues that the subjective interpretation of probability is the appropriate one for evaluating worldviews and that it is routinely used in science, law, and everyday reasoning. Scientists assign subjective probabilities to competing theories (string theory versus loop quantum gravity), juries assess the probability of guilt given evidence, and individuals make decisions under uncertainty by implicitly assigning probabilities to outcomes. If Bayesian reasoning is legitimate in these contexts, it is legitimate in the philosophy of religion as well.^{1, 11}

Against the indeterminacy objection, Swinburne contends that theistic likelihoods are not as indeterminate as Oppy suggests. The classical theistic conception of God as omnipotent, omniscient, and perfectly good places constraints on what God would be expected to do: a perfectly good being would have reason to create goods (conscious beings, beauty, moral agents) and to make himself known to his creatures, making certain features of the world probable on theism even without detailed knowledge of God’s specific intentions. The likelihoods are admittedly imprecise, but Swinburne argues that imprecise probabilities are better than no probabilities, and that the relevant comparative claim — the evidence is more probable on theism than on naturalism — can be defended even without precise numerical values.¹

Against the many-gods objection, Swinburne argues that the simplicity criterion favours classical theism over its competitors. Polytheism postulates multiple entities, deism postulates a God who does less than maximal good (by not sustaining the universe), and a malevolent deity postulates a being that is not maximally good. Each of these alternatives is less simple than classical theism, and therefore each deserves a lower prior probability. The prior probability for theism need not be divided equally among all supernatural hypotheses; the simplest hypothesis gets the largest share.^{1, 15}

Contemporary assessment

The Bayesian approach to natural theology remains one of the most active areas of debate in the philosophy of religion. Swinburne’s programme has been enormously influential, establishing a methodological framework that even its critics engage with. The publication of the Probability in the Philosophy of Religion volume (2012) reflects the maturity of the field, with contributors debating not whether Bayesian methods can be applied to religious questions but how they should be applied — what priors are appropriate, what likelihoods are defensible, and how counter-evidence should be weighted.¹⁴

The fundamental tension in the Bayesian approach is between its formal precision and the philosophical indeterminacy of the inputs it requires. Bayes’s theorem is a deductively valid rule of probability; if the inputs (priors and likelihoods) are correct, the output (posterior probability) is necessarily correct. The dispute is entirely about the inputs, and here the Bayesian framework provides a structure for disagreement rather than a resolution of it. A theist and a naturalist can agree on the mathematical structure of the argument while disagreeing about every probability assignment, reaching opposite conclusions from the same formal apparatus.^{3, 14}

William Lane Craig has endorsed the Bayesian approach as a useful formal framework while cautioning that it should not replace the traditional individual arguments for theism. Craig argues that the kalam cosmological argument, the fine-tuning argument, and the moral argument each have independent deductive or abductive force that does not depend on Bayesian probability assignments, and that the Bayesian framework is most useful as a way of combining these arguments into a cumulative case rather than as a replacement for them.⁶

The Bayesian approach to natural theology thus exemplifies both the strengths and the limitations of formal methods in philosophy. Its strength is that it makes the logical structure of the theistic argument transparent, identifying the precise points at which disagreement occurs and showing how different probability assignments lead to different conclusions. Its limitation is that it cannot resolve the substantive philosophical disagreements that determine the probability assignments. Bayesian arguments for God are only as strong as the philosophical case for the priors and likelihoods that drive them — and that case remains, as it has been since Hume and Swinburne first joined the debate, genuinely and deeply contested.^{1, 3, 16}