The Argument from Miracles: Testimony, Probability, and Hume

Miracle claims occupy a peculiar place in religious epistemology: they are offered as evidence for the supernatural, yet they arrive almost exclusively through human testimony, the most error-prone channel we have. Understanding what it would actually take for such testimony to be convincing is not a matter of faith or cynicism — it is a matter of probability.

Hume's original argument and what it actually says

David Hume's famous essay "Of Miracles" (1748) is often misread as a flat dismissal of the supernatural. His actual argument is more careful. A miracle, for Hume, is a violation of a law of nature. Laws of nature are established by uniform, repeated human experience — the strongest kind of inductive evidence we have. When testimony for a miracle conflicts with that uniform experience, we face a contest of probabilities: the probability that the law holds versus the probability that the testimony is accurate.

Hume's maxim is that we should proportion our belief to the evidence. Because no amount of testimony from fallible, motivated, culturally situated witnesses easily outweighs a law established by millions of observations, the rational default is skepticism. Note that Hume does not say miracle testimony is always false — he says it nearly always fails to clear the evidential bar.

Where the argument is strongest

Hume's probabilistic core holds up well when tested against what we now know about testimony and cognition. Several factors compound the difficulty for miracle claims:

The base rate problem. If genuine miracles are rare — even theists typically grant this — then the prior probability of any given miracle report being accurate is very low. Bayesian reasoning requires that low prior to be overcome by strong, independent evidence before belief is warranted. A single sincere eyewitness rarely supplies that.

Source independence. Many historical miracle reports trace back to a single community, manuscript tradition, or interested party. Corroboration only multiplies the evidential force if the sources are genuinely independent. Witnesses who share a culture, a prior expectation, or a single original account are not independent data points.

The asymmetry of errors. People misperceive, misremember, exaggerate, and sometimes fabricate. These error modes are common and well-documented. The competing hypothesis — that the laws of nature were suspended — requires an extraordinary mechanism with no established precedent. When two explanations compete, the one invoking known failure modes beats the one invoking unknown mechanisms.

Where the argument faces real pushback

Hume's critics have made serious objections that deserve engagement rather than dismissal.

The philosopher John Earman, in Hume's Abject Failure (2000), argues that Hume misapplied probability. Specifically, Hume seems to treat the prior against miracles as so close to 1 that no testimony could ever move the posterior — which would make miracle belief not merely hard but logically impossible. That is a stronger claim than his evidence justifies. Bayesian frameworks do allow, in principle, for multiple independent high-quality witnesses to raise the posterior probability of an unusual event substantially. Earman's point is that Hume's argument is rhetorically decisive but mathematically sloppy.

A second objection targets Hume's definition of miracle itself. If a miracle is simply an event with a very low natural probability rather than a strict "violation" of a law, the argument loses some force. Quantum mechanics allows for events of vanishingly small but nonzero probability; calling them violations of nature becomes harder. Critics like Richard Swinburne argue that a God who acts within a probabilistic universe need not "break" any law.

These are not fatal objections to skepticism about miracles, but they do sharpen what skeptics need to show: not that miracles are conceptually incoherent, but that the evidence for any particular miracle claim is insufficient by identifiable standards.

What a rigorous standard actually demands

A defensible contemporary position runs as follows. Miracle claims should be evaluated like any extraordinary empirical claim: by asking how well the evidence survives alternative explanations. The relevant questions are specific. How many independent witnesses were there, and in what sense were they independent? What were their incentives? How close in time was the account recorded to the alleged event? What naturalistic explanations — hallucination, legend formation, deliberate fabrication, honest misidentification — have been examined and ruled out?

Applying this standard to canonical miracle stories in the major religions reveals a consistent pattern: the closer historians look, the less the evidence meets it. That is not proof of absence. It is a straightforward finding that the evidentiary record, as it stands, does not do the work that miracle apologetics needs it to do.

The honest conclusion is that Hume identified a genuine and important asymmetry, even if his formulation was imprecise. Testimony is a weak instrument for establishing that the ordinary order of nature was interrupted. That does not settle every theological question, but it does mean that anyone who builds a case for religious belief on miracle reports has more explaining to do than is usually acknowledged.