Power in the Age of Artificial Intelligence

By Matthew Parish

Tuesday 8 September 2026

Power is one of those concepts that everybody understands until they are asked to define it. A policeman has power. A billionaire has power. A government has power. An army plainly has power. Yet so does a beautiful woman walking into a crowded room, a newspaper capable of destroying a reputation, a priest whose congregation believes his words and a mathematician who understands something upon which everybody else depends. These things have almost nothing in common.

The history of political philosophy might therefore be understood, in substantial part, as an extended disagreement about what power actually is. Is it the capacity to compel? The ownership of property? The ability to persuade? The capacity to determine what other people believe? Or is power something stranger still — the ability to construct the categories within which other people think?

Artificial intelligence makes these old questions suddenly urgent because it introduces something unprecedented into human history. We are creating machines that possess no armies, vote in no elections, own little property in their own names and occupy no offices of state — yet that may increasingly influence what billions of people know, believe and do.

We may therefore be entering an age in which the most important forms of power become almost invisible.

Hobbes and the power to obtain

Thomas Hobbes offered one of the simplest conceptions of power. A person’s power consists, roughly speaking, in the means presently available to obtain some future good. In this tradition, power is fundamentally capacity — the ability to make something happen.

Artificial intelligence produces an extraordinary multiplication of precisely this kind of power.

Imagine two people attempting to establish businesses. One has access to sophisticated artificial intelligence and the other does not. The first can obtain instantaneous translations, draft contracts, analyse accounts, create advertising materials, write computer programs, investigate markets and conduct preliminary research across almost every academic discipline.

The second must either perform these tasks personally or employ other human beings.

Their formal legal rights may be identical. Their bank accounts might initially contain identical sums. Yet their effective power is radically different. AI therefore represents what might be called cognitive capital.

Industrial machinery multiplied muscular power. A mechanical excavator permits one person to move quantities of earth that once required hundreds of labourers. Artificial intelligence performs something analogous for certain forms of intellectual activity. One person accompanied by machines may increasingly possess the effective intellectual resources of an organisation. That is an enormous transfer of power. But to whom?

Marx and ownership

Karl Marx would immediately ask the question that technological enthusiasts sometimes neglect: who owns the machines? Marx understood that technology cannot readily be separated from the economic relationships surrounding it. The organisation of production helps determine the organisation of society and therefore the distribution of political power. Philosophers of technology still return to Marx precisely because technological systems can embody and reinforce structures of economic authority.

From this perspective, artificial intelligence is not principally interesting because machines may become clever. The important question is who owns the computational infrastructure, data, models, electricity, semiconductor fabrication facilities and intellectual property upon which machine intelligence depends. This produces a paradox.

AI may be one of the greatest democratising technologies ever invented. An obscure student in a provincial town can suddenly obtain intellectual assistance that once required access to an excellent university, an expensive law firm, a team of programmers or an army of research assistants. Yet the infrastructure making this possible is extraordinarily concentrated. The means of intellectual production may consequently become simultaneously democratised and centralised. Marx would recognise the contradiction immediately. The individual becomes more powerful while becoming increasingly dependent upon enormous systems that the individual neither owns nor understands.

Nietzsche and the will to power

Friedrich Nietzsche approached power differently. Power was not merely political authority or material possession. His elusive idea of the will to power concerned striving, overcoming resistance, creation and self-transformation. One influential interpretation understands Nietzschean power precisely through the overcoming of resistance and the capacity to reshape one’s environment in pursuit of one’s purposes.

Artificial intelligence presents a fascinating Nietzschean problem because it dramatically reduces resistance. Writing a book is difficult. Learning a language is difficult. Programming a computer is difficult. Researching an unfamiliar subject is difficult. Difficulty imposes resistance upon the will.

AI removes some of that resistance. At first sight this represents an extraordinary expansion of Nietzschean power. Human beings can suddenly accomplish things that previously exceeded their abilities. Yet there is another possibility. Perhaps resistance itself is essential to human development. The pianist becomes a pianist through thousands of hours struggling with the piano. The philosopher becomes a philosopher through struggling with arguments. The writer becomes a writer through confronting the blank page. Human capacities emerge partly through overcoming obstacles.

What happens when the obstacle disappears? A civilisation possessing enormous technological power might paradoxically produce human beings with diminished individual power because they have outsourced the process through which capacities are acquired.

The machine becomes stronger precisely as its user becomes weaker. That possibility would have fascinated Nietzsche.

Foucault and invisible power

Michel Foucault may nevertheless be the philosopher whose conception of power is most unsettlingly relevant to artificial intelligence. Foucault rejected the idea that power exists principally in obvious commands issued by identifiable rulers. Modern societies exercise power through institutions, classifications, observation, examinations and systems of knowledge. His analysis of disciplinary society emphasised hierarchical observation, normalising judgement and examination — mechanisms through which people become both objects of knowledge and subjects of control.

The prison watches the prisoner. The hospital classifies the patient. The school grades the pupil. The bureaucracy records the citizen. Knowledge and power become intertwined. Now imagine Foucault encountering the twenty-first century. Our telephones know where we travel. Search engines know what interests us. retailers can infer what we desire. Social networks know whom we know. Financial institutions model our creditworthiness.

Governments and corporations accumulate gigantic quantities of behavioural information. Artificial intelligence transforms these archives from passive records into systems of prediction. The distinction is profound. Traditional surveillance asks: What did this person do?

Algorithmic surveillance increasingly asks: What will this person do next? The first records behaviour. The second attempts to anticipate it. And anticipation creates opportunities for intervention.

The algorithmic Panopticon

Jeremy Bentham’s famous Panopticon was a proposed prison constructed so that prisoners could never know whether they were being watched. Foucault transformed this architectural curiosity into a metaphor for disciplinary society. The genius of the Panopticon was that continuous surveillance was unnecessary. The possibility of surveillance was sufficient. People who believe they might be watched begin supervising themselves.

Artificial intelligence carries this principle considerably further. The digital Panopticon need not merely observe people. It can categorise them, compare them, predict them and subtly modify the environments through which they make decisions. And the person exercising power may disappear altogether. Nobody needs to order you to watch a particular video. A recommendation algorithm places it before you.

Nobody needs to prohibit you from reading an article. Another article simply appears first.

Nobody necessarily decides that you are unreliable. A statistical model produces a score.

Power becomes architectural. This may be one of the defining political transformations of our century.

Arendt and collective action

Yet Hannah Arendt offers a very different conception of power — and perhaps a more hopeful one. For Arendt, power should be distinguished from violence and coercion. Genuine political power arises when human beings act together. It belongs not simply to an individual but to a plurality of people cooperating towards a common political purpose.

This distinction becomes exceptionally important in thinking about AI. A machine may become extraordinarily capable without possessing Arendtian power at all. An artificial intelligence cannot create political power merely by calculating faster than us. Political power emerges from human association, recognition and collective action.

Indeed AI might strengthen such power. Small groups can organise more efficiently. Citizens can understand legislation more readily. Language barriers can collapse. Complex government documents can be analysed instantaneously. Expertise previously monopolised by bureaucracies can become widely available. The citizen may acquire intellectual weapons formerly reserved for the state. AI therefore need not necessarily produce technological oligarchy. It could produce technological republicanism. Everything depends upon how it is organised.

Power over truth

There nevertheless remains a deeper problem. Human societies have always depended upon intermediaries that determine what counts as knowledge. Priests once performed this function. Universities do so. Newspapers do so. Governments do so. Encyclopaedias do so. Search engines increasingly do so. Artificial intelligence may become the most powerful epistemic intermediary ever created. Millions of people are beginning to ask machines questions they previously asked teachers, doctors, lawyers, librarians, colleagues and friends.

This creates a peculiar kind of authority. The machine does not need to command anybody. It merely answers. Yet whoever determines the answer may exercise extraordinary influence.

The crucial political question of the AI age may therefore become not: Who controls the government? but: Who controls the systems through which society understands reality? This is power at its most Foucauldian. It determines not necessarily what we must do but what appears reasonable for us to think.

The disappearance of the powerful

There is another peculiarity about algorithmic power. Traditional power normally has a face.

The king sits upon his throne. The dictator appears upon the balcony. The general wears his uniform. The policeman carries his badge. The judge sits elevated in court. One knows who possesses authority. Algorithmic power can be strangely anonymous. Suppose an automated system refuses somebody credit. Who exercised power? The programmer? The bank? The training data? The executives who purchased the software? The mathematical model? The regulator who permitted its use? The customer whose previous behaviour became part of the dataset?

Perhaps all of them did. Perhaps none of them did individually. This creates a serious philosophical problem because our concepts of responsibility developed around human agency. We assume that when power is exercised there exists somebody capable of explaining why. Artificial intelligence can create consequential decisions without there necessarily being any individual who understands completely how the decision arose.

Power may thereby become detached from intention. That is historically unusual and politically dangerous.

The power to define the normal

Foucault’s idea of normalisation becomes especially important here. Artificial intelligence learns patterns. But patterns inevitably establish norms. A model trained upon millions of successful employees develops some representation of what successful employees resemble. A medical system develops representations of normal and abnormal patients. A security system distinguishes ordinary from suspicious behaviour.

Statistical description can quietly become social prescription. What usually happens becomes what ought to happen. The unusual person then becomes an anomaly. This has always been one of the dangers of bureaucracy. AI can magnify it because classification can occur on a previously unimaginable scale. The eccentric, the dissident, the innovator and the genius share an inconvenient characteristic. They are statistical outliers. A society governed excessively by predictive systems might become extraordinarily efficient while becoming intellectually sterile.

Power without consciousness

We then encounter perhaps the strangest question of all. Must something understand that it possesses power in order to possess power? A hurricane possesses enormous causal power but no political power because it has no intentions. An emperor possesses political power because his intentions matter. Where should artificial intelligence sit between these categories?

For the moment the sensible answer is that AI systems remain embedded within human institutions. Their apparent agency is ultimately deployed through human organisations, infrastructures and purposes. Contemporary philosophical work on AI and power accordingly emphasises that artificial power still operates through human technological practices.

But increasingly autonomous systems complicate the picture. Suppose an AI system negotiates contracts, manages investments, employs other AI systems, operates machinery and communicates persuasively with human beings. At what point does describing it merely as a tool become philosophically inadequate? A corporation provides an interesting analogy.

A corporation has no biological brain. Nevertheless corporations own property, employ people, sue governments and shape societies. Human civilisation already knows how to construct artificial entities that exercise enormous power. Artificial intelligence may simply create much more intelligent ones.

Democracy after intelligence

This brings us to the political problem. Democracy developed upon an assumption that human beings are the principal reasoning agents in society. One citizen, one vote.

But intelligence was never evenly distributed and neither was knowledge. Representative institutions partly emerged because no individual could understand everything necessary to govern a complex society.

Artificial intelligence alters this informational landscape. A minister accompanied by sufficiently sophisticated AI might command analytical capabilities once requiring an entire civil service. A citizen might possess them too. The consequences could run in opposite directions. AI could produce the most centralised societies humanity has ever known because governments and corporations obtain unprecedented capacities to observe, predict and influence populations.

Or it could produce radically decentralised societies because individuals acquire capabilities once monopolised by large institutions. The technology itself does not resolve this tension. Politics will.

The final paradox

The great philosophers of power each illuminate a different part of the emerging world.

Hobbes reminds us that power is capacity — and AI enormously expands human capacity. Marx asks who owns the productive machinery upon which that capacity depends. Nietzsche asks what happens to human beings when resistance, struggle and self-overcoming are delegated to machines. Foucault asks how knowledge, classification and surveillance quietly manufacture obedience. Arendt reminds us that genuine political power may ultimately remain something distinctively human: people choosing to act together.

Artificial intelligence brings all these conceptions into collision.

Yet perhaps the greatest transformation is simpler. For most of history, power belonged principally to those who controlled things — land, gold, factories, weapons, ships and armies.

During the twentieth century it increasingly belonged to those who controlled institutions — governments, corporations, banks, universities and media organisations.

In the twenty-first century, power may increasingly belong to those who control interpretation. Who decides what information matters? Who determines what is credible? Who predicts what we will want? Who constructs the intellectual environment within which choices appear before us? These questions are becoming inseparable from artificial intelligence.

The most powerful machine may not be the machine that can force us to do something.

It may be the machine that persuades us that we wanted to do it anyway. And here lies the philosophical danger peculiar to our age.

Human beings have spent several thousand years devising ways to restrain visible power. We have constitutions to restrain governments, courts to restrain officials, elections to remove politicians and laws to constrain corporations. We are considerably less experienced at restraining power that presents itself merely as useful information. The tyrant tells you what to think. The algorithm tells you what you might find interesting. The distinction seems enormous. One day we may discover that it was smaller than we imagined.