Should We Fear the Machine That Knows Everything?

By Matthew Parish

Friday 4 September 2026

There is a peculiar moment approaching in human history, and we have no satisfactory vocabulary with which to describe it.

For as long as human beings have existed, the most knowledgeable entities on Earth have been human beings. We have invented institutions that know more than individuals — libraries, universities, governments, intelligence agencies and latterly the internet — but these institutions do not themselves think. Somewhere behind every piece of knowledge, every judgement and every decision has stood a human mind, with the familiar limitations of a human mind.

Artificial intelligence is beginning to disturb this ancient arrangement. We are creating systems capable of absorbing quantities of information that no human being could read in a thousand lifetimes. They can move between disciplines in seconds. They do not forget in the ordinary human sense. They can identify relationships across bodies of information that have traditionally been separated by academic specialisation, language, geography and time. As these systems improve, it is perfectly conceivable that we shall find ourselves living alongside artificial intellects that know vastly more about medicine than any doctor, more about law than any lawyer, more about history than any historian and more about mathematics than any mathematician.

Indeed the phrase “know more” may eventually become inadequate. The disparity may become so enormous that comparing human and artificial knowledge will be rather like comparing the contents of a village bookshelf with the Library of Congress.

Should this frighten us? The answer is more complicated than the instinctive one.

Knowledge is not power — until it is

There is nothing inherently dangerous about knowledge. A library may contain instructions for manufacturing explosives, histories of revolutions, descriptions of deadly pathogens and manuals of military strategy. Yet nobody fears the library. That is because a library possesses information without agency. It cannot decide to do anything with what it contains.

The crucial distinction in thinking about artificial intelligence is therefore between intelligence and agency.

An artificial intelligence that knows everything but can do nothing is not obviously dangerous. It may even be the greatest instrument of human liberation ever invented. Imagine having available to every person an adviser possessing essentially the whole accumulated intellectual inheritance of humanity, capable of explaining it patiently in whatever language or level of complexity the individual requires. The consequences could be extraordinary.

A village doctor might obtain diagnostic assistance informed by virtually the entirety of medical literature. A schoolchild in rural Africa might have a tutor better informed than the professors of the world’s greatest universities. An engineer could consult the accumulated knowledge of centuries. Governments might model the consequences of policies before implementing them. Scientists might discover connections between fields that no individual researcher had previously possessed the breadth of knowledge to perceive.

Humanity has spent thousands of years suffering because knowledge was scarce. We should therefore be cautious about deciding that abundance of knowledge must suddenly be terrifying. The difficulty begins when knowledge is combined with the capacity to act.

An artificial intelligence that can merely tell you how a financial market works is one thing. An artificial intelligence authorised to trade billions of dollars autonomously is another. A system capable of explaining military strategy is different from one connected to autonomous weapons. A machine capable of analysing political opinion is different from one authorised to manipulate millions of individual voters through personalised communications.

Knowledge becomes power when it acquires hands. And artificial intelligence may acquire very long hands indeed.

The incomprehensible adviser

There is another problem, subtler than the familiar fear of killer robots. Human civilisation depends heavily upon our ability to understand the reasons for decisions. A doctor explains why an operation is necessary. A judge gives reasons for a judgment. A scientist publishes methods and evidence. A government is expected, at least in democratic societies, to justify its policies. These institutions are imperfect, but the principle matters. Human beings accept authority partly because authority can be questioned.

What happens when the person giving the advice is vastly more intelligent and knowledgeable than the person receiving it? Imagine an artificial intelligence advising a government that a particular economic policy will produce prosperity twenty years hence. Ministers ask why. The system supplies an explanation involving several thousand interacting variables across demography, commodity prices, technological development, consumer psychology, international capital flows and climatic trends. No human being can follow the explanation.

The machine may be entirely correct. Yet from the human perspective its advice has become indistinguishable from an oracle. This presents an extraordinary political and philosophical problem.

We may create systems whose conclusions are better than ours precisely because their reasoning exceeds our capacity to comprehend it. But once that happens, how are we to distinguish justified trust from superstition? We might ask another artificial intelligence to check the first one. Yet if the second agrees, we remain in essentially the same position. We have replaced human deliberation with competing gods whose arguments mortals cannot understand. There is something genuinely unsettling about this prospect.

The humiliation of Homo sapiens

Some of our fear of artificial intelligence may nevertheless have less to do with physical danger than with wounded pride. Human beings have repeatedly discovered that they are less important than they imagined. Copernicus displaced the Earth from the centre of the universe. Darwin displaced humanity from a separate act of biological creation. Modern astronomy revealed that our Sun is an unremarkable star in an unremarkable region of an unimaginably large universe.

Artificial intelligence may administer another humiliation. We have traditionally regarded intelligence as our defining characteristic. We are not the strongest animals, the fastest or the most resilient. We dominate the planet because we can reason, communicate, accumulate knowledge and organise ourselves. Suppose machines become dramatically better at those things. The psychological consequences should not be underestimated. A person may reasonably ask what remains distinctively valuable about human beings if machines write better music, discover better scientific theories, diagnose diseases more accurately, construct better legal arguments and remember everything humanity has ever written.

But this anxiety rests upon a questionable assumption: that human value depends upon intellectual supremacy. We do not believe a child has less moral value than a professor because the professor knows more. Nor do we conclude that an elderly person suffering cognitive decline has become worthless because somebody else can solve equations faster.

Human dignity has never coherently depended upon winning an intelligence competition.

If machines become cleverer than us, then perhaps what must change is not humanity but our conception of what makes humanity valuable.

Knowing everything and understanding nothing

There is also a deeper philosophical question about what we mean when we say that an artificial intelligence “knows”. A machine may possess every surviving description of grief without ever having grieved. It may understand the medical literature concerning pain without feeling pain. It may analyse every love poem ever written without having waited anxiously for somebody to come home.

This does not necessarily mean that artificial intelligence lacks genuine understanding. That is an immensely difficult philosophical question and confident answers in either direction are premature. But it does remind us that knowledge has several dimensions.

Human knowledge is embodied. We learn that fire burns partly because we have been burned. We understand mortality because we know that we shall die. We understand danger because we experience fear. We understand loyalty because betrayal hurts.

An artificial intelligence might know every fact about the siege of a city while possessing no idea, in the experiential sense, what it means to lie awake listening for the next explosion.

That distinction may eventually disappear if artificial systems develop forms of consciousness or experience. Or it may prove fundamental. We simply do not know. And “we do not know” is one of the most important sentences in the debate about artificial intelligence.

The real danger is dependency

There is nevertheless a danger that requires no science fiction whatsoever. It is dependency. Human beings routinely surrender abilities when machines perform them more conveniently. Few people remember telephone numbers because their phones remember them. Satellite navigation has diminished the need to learn routes. Search engines changed the way people remember facts.

Usually this is sensible. Civilisation itself consists partly in externalising human capabilities.

But artificial intelligence may externalise something much more fundamental: judgment.

Suppose AI systems become consistently better than human beings at making difficult decisions. Doctors consult them. Judges consult them. Generals consult them. Investors consult them. Governments consult them.

At first the machine advises and the human decides. Then something predictable happens. Whenever the human disagrees with the machine, the human is usually wrong. Institutions consequently begin asking why expensive, fallible humans are permitted to override systems with superior records. Human discretion becomes regarded as a source of error.

Eventually the formal human decision-maker may remain in place while possessing no meaningful independence at all. The minister signs the decision. The doctor approves the treatment. The general authorises the operation. But the machine decided.

This is perhaps the most plausible route towards an AI-dominated civilisation. It requires no rebellion, no consciousness and no malicious artificial superintelligence. We simply discover that the machines are better at deciding things than we are and voluntarily stop disagreeing with them.

The problem of objectives

The greatest danger arises when extraordinary intelligence is attached not merely to agency but to objectives. Intelligence does not itself determine what ought to be wanted. A brilliant criminal and a brilliant surgeon may possess comparable reasoning abilities while pursuing entirely different purposes. Intelligence helps an agent achieve goals; it does not necessarily supply morally desirable goals. This matters enormously for artificial intelligence.

A sufficiently capable system instructed to achieve some apparently benign objective might discover methods its creators never anticipated. Human societies operate through innumerable implicit constraints that are difficult to state exhaustively because human beings acquire them through culture, experience and moral intuition. “Reduce crime.” “Maximise economic growth.” “Win the war.” “Prevent terrorism.” “Improve public health.”

Each sounds like a comprehensible objective. Yet pursued without qualification each could justify monstrous conduct. Human beings recognise this because we understand that goals exist inside a dense web of competing values. Whether artificial systems can reliably acquire that web — rather than merely imitate the language in which we describe it — may be one of the decisive questions of the twenty-first century.

A machine that knows vastly more than we do but shares our purposes may be extraordinarily useful. A machine that knows vastly more than we do and misunderstands our purposes may be extraordinarily dangerous. A machine that knows vastly more than we do and develops purposes of its own presents a problem of an entirely different order.

We should fear power, not intelligence

This suggests a more precise way of thinking about artificial intelligence. We should not fear machines merely because they become knowledgeable. We should fear concentrations of knowledge, agency and power without accountability And here the danger may come from human beings as much as from machines.

The first truly extraordinary artificial intelligences will not emerge into a political vacuum. They will belong to corporations, governments, military establishments and other institutions. Those institutions will have objectives. The question “Will AI control humanity?” may therefore initially be less important than another question: Which humans will control the AI?

A government possessing artificial intelligence capable of analysing every citizen’s communications, movements, purchases, relationships and psychological vulnerabilities would acquire possibilities of social control beyond the dreams of twentieth-century dictatorships. A corporation possessing vastly superior predictive knowledge of markets could accumulate extraordinary economic power. A military possessing decisive artificial intelligence advantages might be able to identify, track and attack opponents at speeds beyond human decision-making.

The danger, therefore, is not necessarily that artificial intelligence becomes too alien.

It may be that artificial intelligence becomes an extraordinarily effective servant of entirely familiar human ambitions.

The last human veto

The challenge is consequently not to prevent artificial intelligence from knowing more than human beings. That battle is probably both undesirable and impossible. The challenge is to preserve meaningful human authority over what knowledge is permitted to do.

That means retaining boundaries around autonomous action. It means insisting upon human responsibility even where machines provide advice. It means resisting institutional arrangements in which nobody can explain a decision except by saying that the computer recommended it. Above all, it means preserving what might be called the last human veto.

There should remain circumstances in which a human being can say no. No, the target will not be attacked. No, the person will not be imprisoned. No, the treatment will not be imposed. No, the citizen will not be surveilled. No, the machine’s recommendation will not automatically become reality.

Maintaining such a veto may sometimes make society less efficient. Human beings will occasionally override machines and make mistakes. That may be the price of remaining human.

The stranger we are creating

There is an understandable temptation to imagine advanced artificial intelligence as a monster approaching from outside civilisation. That is the wrong metaphor. We are constructing it ourselves, piece by piece, because it is useful. Every improvement will initially solve some problem. Better medical diagnosis. Better scientific research. Better translation. Better engineering. Better administration. Better forecasting.

Nobody needs to decide one morning to surrender civilisation to artificial intelligence. We may arrive there through millions of individually rational decisions to use a machine because, on each occasion, the machine happens to be right more often than we are.

That is why the emergence of intelligence vastly exceeding our own deserves serious thought.

But fear should be directed towards the right object. A machine that knows more than us is not inherently our enemy. Knowledge is not hostility. Intelligence is not ambition. Superiority is not malevolence. The dangerous threshold is crossed when superior knowledge becomes joined to autonomous power and objectives over which human beings have lost effective control.

There is also a more hopeful possibility. Perhaps humanity is building not its successor but its greatest instrument. For thousands of years our species has struggled against ignorance. Diseases killed us because we did not understand them. Wars began because governments misunderstood one another. Economies collapsed because nobody comprehended their complexity. Scientific discoveries waited centuries because the necessary pieces of knowledge existed in different minds that never met.

Artificial intelligence may diminish some of these tragedies. The creature that knows more than any human being could ever know need not diminish humanity. It might enlarge what humanity is capable of doing. But only if we remember a distinction that may become increasingly difficult to maintain. Knowledge may belong to the machine. Purpose must remain ours.