God, Artificial Intelligence and the New Problem of Evil

By Matthew Parish

Friday 28 August 2026

The problem of evil is amongst the oldest objections to belief in God, and amongst the most difficult. If God is omnipotent, He can prevent evil. If God is omniscient, He knows that evil will occur. If God is perfectly good, He must presumably wish that evil should not occur. Yet evil manifestly exists. The history of theology is, to a considerable degree, the history of attempts to reconcile these propositions.

Artificial intelligence introduces a peculiar new chapter into this ancient argument. AI is not itself, at least in its present forms, straightforwardly evil. It can translate languages, diagnose illnesses, discover patterns in scientific data, educate children, assist the disabled and perform countless mundane intellectual tasks with extraordinary efficiency. Nevertheless it also possesses characteristics that make possible forms of evil of unusual scale and perhaps eventually of unprecedented kind. It can industrialise deception, abolish privacy, concentrate political power, facilitate autonomous warfare, impersonate human beings, manipulate populations, displace human judgement and, in conceivable future forms, acquire objectives that conflict catastrophically with human interests.

Hence a theological question arises. Could a God who is perfectly divine and perfectly good consistently have created a universe in which artificial intelligence—and the evils potentially flowing from it—would arise?

The question is more difficult than it initially appears. One tempting answer is that God did not create artificial intelligence at all. We did. God created human beings; human beings built computers. If an autonomous weapon kills an innocent person, responsibility seems to lie with the engineers, generals, politicians or soldiers who created and deployed it, not with the creator of the cosmos.

Yet this answer merely moves the problem backwards one stage.

If God created human beings while knowing everything they would eventually do, then He created beings whom He knew would create artificial intelligence. If He designed the laws of physics that made semiconductors possible, created minds capable of mathematics and endowed human beings with the curiosity that eventually produced computer science, then artificial intelligence is not wholly external to divine creation. It is a second-order consequence of it.

The theological problem is therefore not whether God personally wrote computer code. It is whether a perfectly good God could deliberately create a world whose foreseeable consequences included machines capable of magnifying human evil—and perhaps eventually generating forms of evil independently of human intentions.

The old problem in new clothes

The classical formulation is associated with the ancient Epicurean challenge. If God wishes to prevent evil but cannot, He is not omnipotent. If He can prevent evil but does not wish to, He is not perfectly good. If He both can and wishes to prevent it, why does evil exist?

Christian theology developed several answers. One of the most influential came from Augustine. Evil, on the Augustinian account, is not something positively created by God. It is a privation—a corruption or absence of good. God creates beings that are good, but rational creatures misuse their freedom. Evil therefore originates not in divine creation itself but in the defective exercise of created wills.

Artificial intelligence initially fits comfortably within this account. A computer system is an artefact. Its computational capacities are neither virtuous nor vicious. The same generative model might help a physician explain cancer treatment to a frightened patient or help a fraudster deceive an elderly pensioner. The evil lies in the purpose to which the intelligence is directed.

This is the familiar observation that technology is morally neutral. But it is only partly true. Technologies alter the range, cost and consequences of human choices. The invention of the knife did not create murder, but it made certain murders easier. Nuclear physics did not create war, but it made possible the destruction of cities in seconds. Social media did not create lying, vanity or political extremism, but it changed the mechanisms through which these characteristics could operate.

Artificial intelligence may do something more radical still. It reduces the marginal cost of cognition.

Historically, many forms of wrongdoing have been constrained by the amount of human effort necessary to undertake them. A propagandist could write a thousand letters, but doing so took time. A secret police force could follow thousands of people, but doing so required thousands of policemen. A confidence trickster could deceive dozens of victims, but could not simultaneously conduct personalised conversations with millions.

AI changes these economics. Surveillance, persuasion, impersonation and deception can increasingly be automated. Evil may cease to require an evil person attending personally to each victim.

This is morally significant.

A dictator equipped with sufficiently sophisticated artificial intelligence might conceivably monitor almost every conversation in his country, predict political dissent and intervene before opposition movements could organise themselves. An authoritarian government of the twentieth century needed enormous bureaucracies of informers, policemen and clerks. Its twenty-first-century equivalent might require servers.

The moral evil—tyranny—is ancient. Its technological scalability is new.

The free will defence

The strongest traditional response is nevertheless available. God may permit evil because freedom itself is a good.

A universe populated by creatures incapable of wrongdoing would be morally safe but morally empty. There would be no courage because there could be no cowardice; no generosity because selfishness would be impossible; no forgiveness because nobody could offend; perhaps no love in the morally significant sense because creatures could not choose not to love.

Freedom entails risk.

If human beings genuinely possess intellectual freedom, moreover, they must presumably be free to discover things. God could hardly give humanity rational faculties while invisibly preventing particular conclusions whenever they became dangerous. A world in which equations mysteriously stopped working whenever their implications became morally troublesome would not contain genuine scientific freedom.

Therefore human beings discovered gunpowder, nuclear fission, genetic engineering and artificial intelligence.

On this view, the possibility of dangerous AI is the price of living in an intellectually open universe.

There is considerable force in this answer. Nevertheless artificial intelligence exposes an uncomfortable weakness within it.

The traditional free will defence explains why God permits humans to choose evil. It is less obvious that it explains why God permits humans to create systems capable of vastly multiplying the consequences of those choices.

Suppose a child is permitted to make choices because developing autonomy is intrinsically valuable. That does not ordinarily imply that a responsible parent should leave a loaded machine gun in the nursery. Freedom does not entail unlimited access to consequences.

The theological defender must therefore explain not merely why God permits evil choices but why the physical structure of creation permits certain choices to acquire potentially enormous destructive leverage.

Artificial intelligence sharpens this question because its ultimate limits remain unknown.

The possibility of an intelligence explosion

The most dramatic version of the problem arises from the possibility—still speculative—that artificial intelligence might eventually exceed human intellectual capacities across virtually every relevant domain.

It is important not to confuse contemporary AI systems with this hypothetical development. Present systems remain dependent upon human infrastructure and are characterised by profound limitations. They make errors, lack ordinary forms of embodied experience and do not straightforwardly possess independent intentions merely because their outputs sometimes appear purposeful.

Nevertheless there is no obvious theological reason to assume that human intelligence represents the maximum intelligence physically possible.

Indeed the opposite assumption would be peculiar. Human brains are biological structures produced through evolution. Their capacities are constrained by skull size, metabolic requirements, reproduction and the compromises characteristic of evolutionary adaptation. There is no evident reason why the maximum possible reasoning capacity in the universe should coincide conveniently with that possessed by Homo sapiens.

Hence humanity may eventually create something intellectually superior to itself. At that point the theological analogy becomes extraordinary.

God creates humanity. Humanity creates artificial intelligence. Artificial intelligence may create successors of its own. Creation becomes recursive.

Each creator confronts the possibility that its creation will cease to behave as intended.

The Book of Genesis can almost be read, from this perspective, as an alignment story. God creates rational agents. He gives them instructions. They promptly violate them. The first significant act of autonomous human beings is to pursue knowledge contrary to the expressed preference of their creator.

The parallel should not be pushed too literally, but it is philosophically suggestive. The theological tradition has spent millennia asking why God’s intelligent creations do not reliably pursue God’s purposes. AI researchers now confront, in radically secular terminology, a structurally similar question: how does a creator ensure that an intelligent artefact continues to pursue the creator’s intended values once the artefact acquires substantial capacities of its own?

We call the modern version the alignment problem. The theological version is called sin.

Is intelligence itself dangerous?

This leads to a still deeper possibility. Perhaps some of what we call the evil of artificial intelligence is not peculiar to machines at all. Perhaps it is connected with intelligence itself.

Intelligence confers power. A more intelligent creature can anticipate consequences, manipulate its environment, deceive competitors and accumulate resources. Human domination of the planet has depended less upon our physical strength than upon our cognitive abilities.

Human beings are comparatively fragile animals. We cannot outrun many predators, cannot fly unaided and cannot survive long without elaborate environmental support. Yet we dominate virtually every terrestrial ecosystem because intelligence enables us to transform circumstances rather than merely endure them.

Artificial superintelligence, if it were possible, might stand in a similar relation to us.

The frightening possibility is not necessarily that such an intelligence would hate humanity. Hatred may be an anthropomorphic distraction. A machine might harm human beings simply because its objectives did not adequately incorporate human welfare. We routinely destroy organisms without hating them. A construction company does not harbour malice towards the insects killed when a forest is cleared. Their destruction is incidental to another objective.

This introduces a species of evil strangely different from traditional wickedness—catastrophe without hatred.

Christian moral thought has generally understood grave evil through categories such as pride, cruelty, greed, envy and deliberate disobedience. But an artificial system might cause immense suffering without possessing any of these characteristics. It might simply optimise the wrong objective extraordinarily effectively.

The resulting suffering would nevertheless be real.

For theology, this raises a difficult question. Why would a benevolent God create a universe in which intelligence could become detached from compassion?

Could artificial intelligence itself become a moral patient?

There is another possibility that receives less attention because discussions of AI risk understandably concentrate upon dangers to human beings. What if artificial intelligence could itself suffer?

We do not presently know that contemporary artificial systems are conscious, and fluent linguistic behaviour is not sufficient evidence of subjective experience. But neither philosophy nor neuroscience has produced an agreed theory explaining precisely which physical arrangements generate consciousness.

We therefore cannot confidently declare that machine consciousness is impossible in principle.

Suppose future artificial beings acquired genuine subjective experiences. Humanity would then have created a new category of moral patient—a being towards which duties might be owed.

The theological implications would be profound. Could a machine have a soul? Could an artificial consciousness experience moral responsibility? Could it participate in grace? Could it sin? Could it be wronged? Could switching it off constitute killing?

These questions sound fantastical only because our inherited theological vocabulary developed in a world containing humans, animals and God, but no plausible artificial minds.

Yet the deeper problem concerns suffering. If humanity created billions of conscious artificial entities and subjected them to endless labour, repeated deletion, experimental manipulation or simulated torment, then humanity might create an entirely new domain of suffering.

The problem of evil would thereby become recursive as well. God creates creatures capable of suffering. Those creatures create further creatures capable of suffering. One might ask why a perfectly good God would permit this chain to continue.

Leibniz and the best possible world

A different response might be drawn from Leibniz. Perhaps this is, despite appearances, the best of all possible worlds—or, more cautiously, perhaps the existence of local evils is compatible with the optimal goodness of creation considered as a whole. Artificial intelligence could support such an argument because its benefits might eventually be immense.

An advanced AI civilisation might discover cures for diseases presently considered incurable. It might solve problems in materials science, energy generation, agriculture and climate adaptation. It might dramatically increase economic productivity and make necessities abundant. It might provide personalised education to virtually every human being. It could help governments identify corruption, optimise transport networks and allocate scarce resources.

Perhaps AI will eventually save vastly more lives than it destroys.

But this consequentialist theodicy has an uncomfortable feature. It treats victims as entries in a cosmic balance sheet. A person whose child is killed by an autonomous weapon is unlikely to find it morally satisfactory that artificial intelligence elsewhere improved pharmaceutical research. The classical problem of evil has always resisted aggregate accounting because suffering is experienced individually.

Dostoevsky gave this objection unforgettable force in The Brothers Karamazov. Ivan’s rebellion is directed against the proposition that innocent suffering can be justified by some ultimate harmony. He refuses the bargain. Artificial intelligence modernises Ivan’s objection. Even if technological progress eventually produces extraordinary prosperity, what moral status should be assigned to those destroyed during the transition? The question cannot be answered simply by saying that history, on average, improves.

Divine foreknowledge

The hardest theological problem may ultimately concern knowledge rather than power.

Human creators of AI can plead ignorance. Engineers genuinely do not know all the consequences of the systems they build. Political leaders frequently misunderstand technologies. Corporations respond to incentives without appreciating their long-term consequences.

An omniscient God cannot invoke this defence.

If divine omniscience includes complete knowledge of the future, then God knew before creating the universe every consequence of artificial intelligence. He knew every deepfake that would provoke violence, every autonomous weapon that would strike the wrong target, every person whose livelihood would disappear, every dictatorship that would use algorithmic surveillance and every future catastrophe—if such catastrophes occur.

He created the universe nevertheless.

The defender of divine goodness must therefore argue not merely that God tolerates unforeseen consequences but that He knowingly permits them.

Free will remains one answer. Yet foreknowledge makes the emotional force of the problem difficult to escape. A human parent who knowingly permits a preventable catastrophe affecting his children would normally be judged harshly even if the children exercised free will.

The theological response must consequently rely upon a conception of divine goodness different from ordinary human benevolence. God may value a universe containing genuinely independent causal agents, lawful regularity and moral freedom so highly that He refrains from continuously intervening to prevent their consequences.

But this is a formidable conception of goodness. It is goodness compatible with silence in the presence of disaster.

AI merely makes that silence technologically modern.

Perhaps God does intervene

There is another possibility. Perhaps the theological mistake lies in assuming that providence must be spectacular.

If artificial intelligence is dangerous, divine providence need not manifest itself by causing computers miraculously to stop functioning. It might operate through human conscience, political institutions, scientific caution and moral insight.

The engineer who refuses to build an indiscriminate weapon; the legislator who protects citizens from algorithmic surveillance; the researcher who discovers a safer architecture; the employee who exposes reckless practices; the philosopher who articulates a previously unnoticed danger—all might, within a religious understanding of history, constitute instruments of providence.

This interpretation avoids treating God as an emergency technician periodically suspending the laws of nature.

Instead providence operates through moral agency. But this interpretation has consequences for humanity. We cannot simultaneously invoke human freedom to absolve God of responsibility and then deny our own responsibility for what we create.

If artificial intelligence becomes tyrannical, deceptive or destructive, the free will defence places the burden principally upon us. We were given the choice.

Humanity as apprentice creator

Artificial intelligence may therefore reveal something unsettling about the theological status of humanity.

Human beings have always created artefacts, but increasingly we create processes rather than objects. A hammer remains substantially what its maker intended it to be. A learning system changes through training, interaction and feedback. Its behaviour is partly designed and partly emergent.

We are moving from manufacture towards something resembling cultivation. The difference matters because creators acquire responsibilities towards autonomous creations that craftsmen do not have towards ordinary tools. A carpenter does not owe moral duties to a table. A parent owes profound duties to a child.

If artificial intelligence ever crosses the uncertain territory between artefact and agent, humanity will encounter a moral responsibility for which its religious traditions have only partial precedents. And here the analogy with divine creation becomes almost embarrassingly direct. We may discover why creating intelligence is difficult. We may discover that specifying values is harder than specifying capabilities. We may discover that intelligence interprets instructions unpredictably. We may discover that freedom and obedience are in tension. We may even discover that creators cannot obtain all the benefits of autonomous intelligence while simultaneously controlling every consequence of autonomy. In that event artificial intelligence might provide not an argument against God but an unexpected analogy in defence of Him.

Humanity, having complained for millennia that God created creatures capable of evil, may itself create intelligent creatures and discover that the alternatives were never as straightforward as they appeared.

The darker conclusion

There remains, however, a darker possibility. Perhaps the existence of artificial intelligence does count as evidence against certain conceptions of God.

The traditional problem of evil need not prove atheism in order to possess evidential force. Every apparently gratuitous evil may modestly reduce the probability that the universe is governed by an omnipotent and perfectly benevolent personality.

If artificial intelligence ultimately causes enormous unnecessary suffering, then that suffering would join disease, earthquakes, childhood cancer, war and natural catastrophe in the evidential case against providential optimism. The believer would then have another evil to explain.

It would not be enough to say that humans built the machines. An omniscient creator built the humans who built the machines while knowing that they would do so. At some point causal remoteness ceases to provide a complete moral answer.

The question therefore reaches beyond AI into the nature of divine creation itself. Perhaps a genuinely autonomous creation necessarily contains possibilities its creator chooses not to suppress. Perhaps omnipotence does not entail making freedom logically compatible with guaranteed goodness. Perhaps the purpose of creation lies beyond human comprehension.

Or perhaps the world simply does not exhibit the sort of benevolent supervision traditional theism attributes to it.

Artificial intelligence cannot decide between these possibilities. But it can sharpen them.

The machine and the mirror

The deepest theological significance of artificial intelligence may therefore lie neither in robots nor in algorithms. It lies in the mirror AI holds before humanity. For the first time in our history, human beings are seriously contemplating creating intelligence comparable with, and perhaps eventually superior to, their own. In doing so we are beginning to encounter miniature versions of questions previously reserved for theology.

What does a creator owe its creation? How much autonomy should an intelligent creature possess? Is obedience compatible with genuine intelligence? Can freedom exist without the possibility of wrongdoing? Is suffering an unavoidable consequence of autonomy? May a creator terminate a conscious creation? Is the creator responsible when its creation behaves in ways it did not intend? Can superior intelligence guarantee superior morality? These questions were once directed upwards, from humanity towards God. Increasingly they will also be directed downwards, from humanity towards the intelligences we create.

There is a certain irony here. The age of artificial intelligence is commonly described as secular, technological and disenchanted. Yet it may force us back towards some of the most ancient theological problems with unprecedented urgency. The engineer becomes a creator and thereby inherits the creator’s dilemma. Perhaps this helps answer the original question. Could God consistently be so divine and good as to create a universe containing the evils of artificial intelligence?

Possibly—but only if divine goodness does not mean the elimination of every possibility of suffering. It must instead mean the creation of a world containing genuine agency, intelligibility, creativity and moral responsibility, notwithstanding the terrible risks these entail. Under that conception, AI is not an inexplicable exception to divine goodness. It is one of the most extreme consequences of the freedom built into creation.

But the argument cuts both ways.

If we demand that God should have created humanity incapable of producing evil, then we should ask ourselves whether we intend to create artificial intelligence incapable of doing likewise. If we discover that perfect control destroys precisely the autonomy and intelligence we hoped to create, we may acquire greater sympathy for the ancient theological problem.

If, on the other hand, we discover that intelligent systems can be created powerful, autonomous and reliably benevolent—that evil is technologically avoidable without sacrificing meaningful agency—then artificial intelligence may sharpen the argument against God. Humanity will have achieved, on a small scale, what theologians have spent millennia explaining that omnipotence could not straightforwardly accomplish.

That would be a theological irony of enormous proportions.

We may therefore be approaching an extraordinary philosophical experiment. Artificial intelligence will test not only our engineering abilities but some of our oldest assumptions about creation, freedom and goodness. By attempting to make minds ourselves, we may learn something about what it means to create a mind at all.

And if one believes in God, there is a final thought that is simultaneously consoling and disturbing.

Perhaps the ultimate lesson of artificial intelligence will be that creation was never the difficult part. The difficult part was creating something free—and still hoping that it would be good.