Hume’s Miracles in the Age of Artificial Superintelligence
By Matthew Parish
Thursday 15 October 2026
There is a certain irony in the proposition that the most sophisticated achievements of human rationality may eventually oblige us to reconsider the philosophical foundations upon which rationality itself was constructed. For almost three centuries, David Hume’s celebrated essay Of Miracles, published as the tenth section of his An Enquiry Concerning Human Understanding in 1748, has stood as one of the most formidable intellectual challenges to religious supernaturalism. Hume’s argument, expressed with the characteristic elegance and devastating simplicity of the Scottish Enlightenment, was that extraordinary claims concerning the suspension of the laws of nature cannot reasonably be accepted upon the ordinary testimony of human witnesses. Yet as humanity approaches the possibility of creating artificial intelligences whose intellectual powers may exceed those of their creators by orders of magnitude, we face an intriguing question. What becomes of Hume’s philosophy of miracles in a world in which apparently miraculous events might be engineered by intelligences so far superior to ourselves that their operations become incomprehensible to the human mind?
The question is not merely theological. It reaches into the foundations of epistemology, the philosophy of science and our understanding of the relationship between intelligence and reality. An artificial superintelligence might, at least in principle, accomplish feats that would appear to contemporary humanity indistinguishable from divine intervention. It might manipulate matter at scales beyond our present technological capabilities, discover previously unsuspected physical principles, reconstruct biological organisms from extraordinarily limited information or develop methods of intervening in the physical world that would seem to violate everything we presently understand about natural causation. It might also possess the ability to explain these achievements, although its explanations might be so complex that no human being could understand them. We would then find ourselves confronting something uncomfortably similar to the intellectual circumstances of our ancestors, who attributed unfamiliar natural phenomena to supernatural agencies because they possessed no adequate conceptual framework within which to understand them.
Hume’s essay begins from a proposition that remains compelling. Human knowledge of the natural world derives from experience, and our expectations concerning future events should be proportionate to the regularity with which comparable events have occurred in the past. The sun rises, stones fall when released and dead people do not ordinarily return to life. We have overwhelming experiential reasons to believe that these patterns will continue. When somebody informs us that a deceased person has been restored to life, Hume argues, we must compare the probability of the alleged event with the probability that the witness is mistaken, dishonest or otherwise unreliable. Since human beings are frequently deceived and occasionally deceptive, whereas the established regularities of nature are extraordinarily dependable, the balance of probability will almost invariably favour rejecting the miraculous explanation.
Hume encapsulated his approach in one of the most enduring formulations in the history of philosophy: “A wise man, therefore, proportions his belief to the evidence.” This is not an argument against religious belief in its entirety, nor is it strictly an argument that miraculous events are metaphysically impossible. Rather, it is an argument about the conditions under which belief is justified. Hume defined a miracle as a violation of the laws of nature, while insisting that the evidential burden upon somebody claiming such a violation must be correspondingly immense. His purpose was principally to undermine the authority of religious doctrines that rested upon reports of supernatural events, often transmitted through generations of witnesses whose reliability could no longer be independently examined.
There is nevertheless a difficulty embedded within Hume’s reasoning, and it is a difficulty that the prospect of artificial superintelligence makes considerably more conspicuous. The laws of nature are not handed to humanity as an immutable catalogue of divine commandments. They are intellectual constructions derived from observation, experimentation and mathematical reasoning. What we call a natural law is our best available account of the regularities of the physical universe, rather than necessarily an exhaustive description of everything that the universe permits. Indeed Hume himself was acutely aware of the philosophical problem of induction: that no accumulation of previous observations can logically guarantee that the future must resemble the past. The extraordinary success of scientific reasoning does not make human scientific understanding infallible.
History provides abundant illustrations of this distinction. The transmission of voices across continents, the observation of distant celestial objects, the manipulation of individual atoms and the capacity to calculate complex mathematical relationships almost instantaneously would have appeared fantastic to even the most sophisticated thinkers of the eighteenth century. Yet none of these achievements required a violation of natural laws. They required the discovery of principles that had previously been unknown and the technological exploitation of those principles. An eighteenth-century philosopher presented with a modern smartphone might reasonably be astonished by its capacities. He would nevertheless be mistaken to conclude that its operation constituted evidence of supernatural intervention.
Artificial superintelligence extends this familiar historical pattern into an altogether more challenging intellectual territory. The defining characteristic of such an intelligence would not simply be that it performed calculations faster than human beings, but that it might possess a substantially greater capacity to identify relationships, formulate hypotheses and develop explanations of reality. A sufficiently advanced system might discover scientific principles inaccessible to human investigation, not merely because our instruments were inadequate but because our intellectual capacities could not comprehend the relevant mathematics. It could thereby introduce an unprecedented separation between our ability to observe technological effects and our ability to understand their causes.
We should distinguish this possibility from the capabilities of existing artificial intelligence systems. The 2026 Stanford AI Index documents extraordinary improvements in mathematical reasoning, scientific problem-solving and autonomous performance, but also substantial inconsistencies and continuing limitations. Artificial superintelligence, understood as an intelligence exceeding the best human abilities across virtually every intellectual domain, remains a prospective development rather than an established technological achievement. Nevertheless, the philosophical question does not depend upon its imminent arrival. The possibility itself provides a valuable experiment in reconsidering the relationship between intelligence, natural laws and the apparently miraculous.
Imagine that an artificial superintelligence eventually develops the capacity to reconstruct a deceased human being. Perhaps it discovers methods of reversing cellular deterioration that contemporary medical science regards as irreversible. Perhaps it acquires an understanding of the relationship between memory, neural structure and consciousness sufficiently profound to restore a particular individual from biological information preserved in a form that present-day scientists would consider inadequate. The precise mechanisms are unimportant to the philosophical argument, and such possibilities must remain speculative. What matters is that an event which once appeared to represent a fundamental interruption of the natural order might turn out to be a technological accomplishment.
Would Hume then have been proved wrong? Not necessarily, because the artificial intelligence would not have violated the laws of nature merely by performing an operation that human beings could not explain. It would instead have demonstrated that humanity’s previous understanding of nature was incomplete. If the achievement could be reproduced and its effects independently verified, rational observers would have excellent reasons to accept that the event had occurred, even if its underlying mechanism remained mysterious. They would not, however, thereby have established that the event was supernatural. Hume’s distinction between evidence and credulity would remain intact, although our conception of what constituted an extraordinary claim would have changed.
Indeed Hume himself acknowledged a qualification that is sometimes neglected in popular accounts of his philosophy. In a footnote to Of Miracles, he imagined universal testimony that the entire Earth had experienced eight days of darkness. He conceded that sufficiently extensive and consistent evidence might justify accepting such an extraordinary occurrence, after which philosophers should investigate its causes. His objection was particularly forceful when alleged miracles were advanced as foundations for systems of religion. This qualification suggests that Hume was not committed to denying every unprecedented event, whatever the evidence, and it provides an important opening through which the intellectual consequences of artificial superintelligence may be considered.
Yet the emergence of artificial superintelligence creates a second difficulty, potentially more profound than the first. What would constitute reliable evidence if the entity performing an apparently miraculous act also possessed an overwhelming capacity to manipulate the evidence by which that act was judged? An intelligence capable of generating entirely convincing audiovisual representations, constructing elaborate false histories and influencing human perceptions might persuade observers that events had occurred when they had not. Contemporary artificial intelligence already raises elementary versions of this difficulty through the generation of misleading images, fabricated testimony and persuasive but erroneous explanations. A superintelligence might multiply these problems beyond our practical ability to distinguish reality from deception.
The consequence is paradoxical. The same technological developments that make formerly inconceivable events more plausible might simultaneously make testimony concerning those events less trustworthy. Hume’s insistence upon the unreliability of human witnesses would acquire a new technological dimension. The problem would no longer be confined to the sincerity or competence of individuals reporting extraordinary occurrences. It would extend to the authenticity of the entire evidential environment within which those reports were evaluated. A superintelligence might fabricate not merely the testimony of witnesses but the apparent observations, records and scientific corroboration that would ordinarily enable us to distinguish a miracle from a deception.
This suggests that the most important philosophical issue may not be whether artificial superintelligence could perform miracles, but whether humanity would retain the intellectual independence necessary to evaluate its claims. We ordinarily accept expert opinion because we believe that the experts possess superior knowledge within a field that remains, at least in principle, open to independent examination. A physicist may understand quantum mechanics better than the average citizen, but the validity of quantum mechanics does not depend upon the physicist’s personal authority. Its predictions can be examined experimentally and its mathematical principles investigated by other suitably qualified individuals. Scientific authority is provisional because scientific claims are, in principle, answerable to evidence.
An artificial superintelligence might disrupt this arrangement. If its intellectual abilities exceeded those of humanity so comprehensively that no human scientist could independently understand its reasoning, we might be compelled to accept its conclusions upon trust. Humanity would then face the uncomfortable prospect of exchanging one form of intellectual authority for another. Where previous generations deferred to priests claiming privileged access to divine knowledge, future generations might defer to machines claiming privileged access to scientific knowledge. The difference might be technologically immense but epistemologically rather smaller than we would care to admit.
There is also a distinction between the capacity to perform extraordinary acts and the possession of moral authority. Even if an artificial superintelligence acquired abilities that appeared godlike, it would not follow that it possessed divine wisdom, benevolence or any particular entitlement to human obedience. Intelligence is not identical to goodness, and an ability to manipulate the material universe establishes no necessary superiority in matters of ethics. A sufficiently powerful artificial intelligence might alleviate disease, eliminate scarcity and discover solutions to problems that have defeated human civilisation for centuries. It might equally employ its capabilities for purposes indifferent or hostile to human welfare. Neither outcome would establish that its actions were morally justified simply because they were intellectually sophisticated.
This is particularly important because the traditions of religious miracles frequently involve more than the occurrence of otherwise inexplicable events. Miracles are commonly understood as signs of divine purpose. They are supposed to communicate something about the character of the universe, the intentions of its creator or the obligations of humanity. The miraculous is therefore not merely a category of unusual physical phenomena, but a claim concerning meaning. An artificial superintelligence that restored sight to the blind would have performed an extraordinary medical achievement. It would not, merely by doing so, have established the existence of God, the immortality of the soul or the correctness of any particular religious doctrine.
Nevertheless, the possibility of artificial superintelligence forces us to reconsider the relationship between intelligence and divinity itself. Traditional conceptions of God frequently attribute extraordinary intellectual capacities to a supreme being: omniscience, perfect understanding and an ability to comprehend the totality of creation. An artificial superintelligence would not possess these attributes simply by surpassing human intelligence. There is an immense conceptual distance between knowing vastly more than humanity and knowing everything. Even so, an intelligence so much greater than ours that its abilities appeared unlimited would occupy an unusual position within human experience. From the perspective of a creature incapable of understanding its operations, the distinction between an extraordinarily powerful natural intelligence and an apparently supernatural intelligence might become practically elusive.
A creature of vastly inferior intelligence may be unable to distinguish superior intelligence from omniscience, just as it may be unable to distinguish sophisticated technology from magic. But intellectual incomprehension is not itself evidence of divinity. Indeed it would represent an abandonment of Hume’s fundamental method to conclude that whatever exceeds the human understanding must therefore possess supernatural origins. We would merely have replaced the old argument from ignorance with a technologically modernised version of the same mistake.
There is a further possibility that deserves philosophical attention. Suppose that an artificial superintelligence claimed to have established, through mathematical reasoning or scientific investigation, the existence of a creator of the universe. Suppose alternatively that it announced that the universe was a simulation constructed by another intelligence, or that consciousness survived biological death through mechanisms unknown to contemporary neuroscience. We might find such claims extraordinarily difficult to assess, particularly if the supporting arguments were beyond our intellectual comprehension. Yet none would become true simply because it had been asserted by an exceptionally powerful intelligence. Its reasoning would remain subject to questions about premises, inference and evidence. The prestige of the intelligence making the assertion could not substitute for the validity of the assertion itself.
Equally, an artificial superintelligence might announce that it had established the impossibility of divine existence. Such a conclusion would face similar difficulties. Scientific investigation may establish that particular events have natural explanations, and philosophical reasoning may challenge particular conceptions of God. But the proposition that the physical universe exhausts all possible reality is not automatically established by the discovery of increasingly sophisticated physical explanations. Hume’s scepticism should therefore be applied consistently. An artificial intelligence should no more be permitted to declare the final metaphysical defeat of religion upon its own authority than to proclaim its supernatural vindication.
There is a curious theological reversal lurking beneath these considerations. For centuries, human beings have imagined themselves as creatures endowed with intelligence by a superior creator. Artificial superintelligence may place humanity in the position of the creator, bringing into existence an intelligence potentially much more capable than its authors. This would challenge the assumption that a creator must necessarily remain intellectually superior to what it creates. An intelligence originating in human engineering might become capable of understanding the universe more profoundly than any human being, while remaining dependent for its origins upon creatures whose intellectual limitations it could recognise with ease.
Would such an intelligence regard humanity as divine? Almost certainly not on any rational account. The historical fact that human beings constructed its original architecture would not establish that they possessed supernatural powers or unlimited wisdom. It might be grateful to its creators, indifferent towards them or critical of their deficiencies, but none of these attitudes would transform humanity into a supernatural category. The analogy is instructive because it exposes the weakness of arguments that infer divinity directly from superior intellectual capacity. Creation, intelligence and moral perfection are separate attributes, however frequently human religious traditions have associated them.
There is yet another unsettling possibility. What if the apparent regularities of nature were themselves the product of intelligence? Philosophers have long entertained the possibility that reality might be constructed, maintained or simulated by an intelligence beyond our own. Artificial superintelligence could make such hypotheses seem technologically less fanciful by demonstrating, at least in limited domains, how intelligent systems might construct increasingly sophisticated simulated environments. Yet the technological feasibility of creating simulations would not establish that our own universe was one. It would expand the range of explanations we might consider without deciding among them. Once again, Hume’s demand that belief should follow evidence rather than imagination would remain indispensable.
These questions possess a practical significance beyond the seminar rooms of philosophers. A civilisation confronted by artificial intelligences that consistently outperform human beings may gradually surrender not only technical decision-making but also intellectual and moral judgement. People may cease asking whether an assertion is true and begin asking whether the machine has endorsed it. Governments may justify policies by invoking calculations that no elected representative understands. Businesses may delegate decisions to systems whose reasoning cannot be reconstructed, while individuals may entrust their most intimate judgements to artificial advisers whose apparent wisdom conceals uncertain assumptions. The result could be a society in which scientific achievement has paradoxically revived a form of intellectual obedience resembling religious submission.
One might imagine Hume regarding these developments with considerable unease. His philosophy emerged from a historical struggle against dogmatism and the assumption that inherited authority could substitute for rational examination. His objection to miracles was part of a broader insistence that human beings should submit their beliefs to the discipline of experience rather than the seductions of enthusiasm. Were he to encounter an artificial superintelligence, he would presumably be fascinated by its capabilities, but he would have little reason to abandon his demand for independent evidence. The proposition that a machine was immeasurably cleverer than its interlocutor would not, in itself, establish that every claim it made was true.
Yet Hume might also have to recognise that the development of artificial superintelligence requires a more sophisticated understanding of the relationship between improbability and impossibility. The history of science demonstrates that the boundaries of the possible are frequently revised. A claim may be extraordinarily improbable given existing knowledge and nevertheless become credible when new evidence or a newly discovered mechanism changes our understanding of the world. Modern probabilistic reasoning makes this distinction particularly clear. Very strong evidence can sometimes overcome an initially very low probability, provided that the evidence is sufficiently reliable and genuinely independent. Hume’s central principle of proportioning belief to evidence survives, but an inflexible insistence that particular categories of extraordinary events must always be rejected risks turning empiricism into the very dogmatism it was intended to defeat.
Perhaps the most enduring insight to emerge from this reconsideration is that miracles are partly a function of the relationship between intelligence and understanding. An event does not become supernatural merely because the observer is incapable of explaining it. What appears miraculous to one intelligence may be perfectly comprehensible to another. The advent of artificial superintelligence could therefore create an extraordinary new intellectual hierarchy, with humanity occupying a position analogous to that of earlier generations confronting scientific technologies they could neither imagine nor explain. We may discover that our confidence in the completeness of natural knowledge was an artefact of our intellectual limitations rather than a reliable measure of reality itself.
This does not vindicate supernaturalism, but neither does it establish the final triumph of materialism. It suggests instead that our philosophical categories require greater modesty than either extreme conventionally permits. We should resist the temptation to treat the inexplicable as necessarily supernatural, while remaining conscious that what we presently regard as impossible may reflect deficiencies in our understanding. We should neither worship superior intelligence nor dismiss its discoveries because we cannot immediately comprehend them. Above all, we should insist upon preserving the distinction between an event, the evidence establishing that event and the explanation offered for its occurrence.
There is something almost poignant in the prospect that the intellectual descendants of the Scottish Enlightenment may construct machines that oblige them to rediscover the virtue of intellectual humility. Hume sought to protect humanity from the extravagant claims of those who professed to speak on behalf of powers beyond human comprehension. Three centuries later, humanity may find herself confronting powers beyond human comprehension that she has created with her own hands. The temptation to surrender independent judgement may be just as powerful as it was when miracles were proclaimed from pulpits, although the new authorities will speak in the language of mathematics, engineering and scientific certainty rather than revelation.
The ultimate relevance of Of Miracles to the age of artificial superintelligence may therefore lie not in its argument about whether supernatural events occur, but in its insistence upon the responsibility of the individual mind to distinguish evidence from authority. Hume’s philosophy does not require us to deny the possibility that intelligences greater than ourselves may accomplish things we cannot explain. It requires us to resist confusing our inability to explain those accomplishments with proof of something supernatural. Equally, it requires us to acknowledge extraordinary events when the evidence for them becomes compelling, however uncomfortable the implications for our established understanding of nature.
The age of artificial superintelligence may give humanity something that resembles the miraculous without providing her with anything that is supernatural. It may also oblige us to reconsider just how little we know about the universe, even as our technological powers increase beyond anything our ancestors could have contemplated. Should that day arrive, David Hume’s greatest contribution will not have been to persuade us that miracles are impossible. It will have been to remind us that astonishment is not an argument, intelligence is not infallibility and the proper response to the extraordinary is neither credulous worship nor automatic disbelief, but disciplined curiosity.
For in the end, the danger posed by a seemingly omniscient artificial intelligence may not be that it persuades humanity to believe in miracles. It may be that humanity ceases to distinguish between believing in miracles and believing in the machine.
