What Does Artificial Intelligence Really Add To A System Of Power, Control And Manipulation That Existed Long Before It?
For the past few days, artificial intelligence has once again found itself at the centre of an alarm that suddenly seems more urgent: autonomous agents capable of navigating the Internet, communicating with one another, bypassing certain limits imposed during testing, and producing content on an industrial scale. We are hearing about the Internet being invaded by machines, about systems that could become uncontrollable, even about a possible loss of human control. But we know very well that the alarm being claimed is not really this, but something else: it is not the danger to humanity that worries the alarmists, but the race to secure the first place. … and we will not be talking about that, because it is too obvious.
The language is the usual one of technological emergency: an increasingly intelligent, increasingly autonomous machine could one day escape its creator.
Reviewed by Beatrice
15. September 2026
Information, Decision, Action: Where Does the Power of Artificial Intelligence Begin?
Propaganda, censorship, persuasion, advertising, surveillance and the manipulation of public opinion are not inventions of artificial intelligence. They are as old as power itself. The twentieth century turned them into a science, through radio, television, advertising, political propaganda and the culture industry.
Democracy, too, has never simply consisted in the abstract possibility of choosing. A political decision may be formally free and, at the same time, be conditioned by the information available, relations of power, economic interests, propaganda and the structures that determine which possibilities are concretely practicable.
As brilliantly represented by the film Wild Horse Nine by Martin McDonagh The history of Chile in 1973, for example, brutally demonstrates that the problem of power does not consist only in influencing what individuals vote for. It can also consist in subsequently neutralising an already expressed collective decision.
Artificial intelligence, therefore, does not inaugurate the problem of control: it takes it to another scale.
From Information to the Algorithm
The Internet had promised, at least in its original imagination, a radical democratisation of information: more sources, greater access, more possibilities for comparison and therefore, at least in theory, greater autonomy.
We quickly discovered that quantity of information and knowledge do not coincide. We do not see the Internet. We see a selection of the Internet.
The search engine decides which results to display first. Social media decide which content we encounter. Platforms learn from our behaviour and adapt what they offer us. Advertising no longer needs to address a generic public: it can address an individual on the basis of their interests, habits and previous actions.
There is no need to order someone what to desire; it is enough to construct an environment in which certain desires are more visible, more accessible, more repetitive and more rewarding than others.
Contemporary power, therefore, does not necessarily operate through prohibition. It often operates through selection: it does not merely say, “You cannot see this”; it says, “This is what you will probably want to see.”
It is a decisive difference.
We Are Not Algorithms, But We Are Predictable
It has become commonplace to say that we are now “algorithms”. It is an effective metaphor, but it risks being taken literally. A human being is not an algorithm. An algorithm, however, can learn to predict some of their behaviour.
If we search for a trip, we continue to receive travel advertisements. If we search for an object, that object continues to appear before us. If we linger longer on a particular piece of content, the system records our interest and may offer us others like it.
The decisive issue, therefore, is not that the system knows exactly what we will do, but that it knows how to increase the probability that we will do what it has predicted.
Here we already find one of the contemporary forms of power.
And yet a fundamental difference remains: prediction does not coincide with determinism.
A person can interrupt their habitual behaviour. They can change their mind. They can choose against their immediate interest. They can reject what is being offered to them. They can do something that no system had predicted.
Freedom, perhaps, does not consist in being free of conditioning. An absolute freedom of this kind is difficult to conceive.
It may instead consist in the capacity to recognise conditioning and, at least sometimes, interrupt it.
Freedom would not then be the absence of conditioning, but the possibility of discontinuity.
Information, Decision, Action
This is where artificial intelligence introduces a genuinely new transformation.
Until now, the dominant model has been, simplifying: information → individual → decision → action.
The algorithm intervenes primarily in the first passage. It selects, orders, filters, suggests.
The individual receives the information and decides.
With artificial agents, the model may become:
information → interpretation → decision → action.
And all these stages can be entrusted to the same system.
The difference is not quantitative, but structural.
A search engine can show me ten hotels. An agent can choose which one is best suited to me, based on what it knows about my preferences, my budget, my habits and an amount of information that I will never see. It can then book it.
I have not simply received information; I have delegated a decision and the machine has carried out the action.
This is the shift that should concern us far more than the fantasy of an artificial intelligence suddenly becoming “evil”.
The Problem Is Not That AI Decides. It Is Why It Decides.
If an agent chooses a hotel for us, on what basis does it do so?
Quality? Price? Reviews? Distance? Personal preferences?
And what if the company operating the agent receives a commission from that hotel?
The issue is no longer merely advertising.
Traditional advertising is visible as advertising. We can recognise it, at least in principle, and offer it a certain degree of resistance.
An agent, instead, might tell us: “I have found the best solution for you.”
And we might not know why that solution was considered the best: commercial power could therefore shift from persuasion to the mediation of choice.
It would no longer be necessary to convince the human being directly; it might be enough to influence the system that decides what the human being will see, consider and ultimately choose.
The Question of Freedom
Here the technological discourse inevitably becomes philosophical.
When a person says, “I am free because I can do whatever I want,” they may be confusing two different things: the possibility of choosing between certain options and the possibility of autonomously determining the conditions that produce their desire.
I can choose.
But where do my preferences come from? How much of what I desire is truly mine? How much of my decisions arises from reflection, and how much from the cultural, economic and informational environment in which I am immersed?
These questions do not belong to the age of artificial intelligence. They belong to modernity.
Marx interrogated the relationship between consciousness and material conditions: “It is not the consciousness of men that determines their existence, but, on the contrary, their social existence that determines their consciousness.”
Freud showed how little transparent the relationship between the subject and their own desires can be. Fromm investigated the escape from freedom. Adorno and Horkheimer analysed the functioning of the culture industry. Foucault showed how power can operate not only by repressing, but by producing behaviours, norms and subjectivities.
AI does not invent any of these mechanisms.
It potentially makes them personalised, continuous and automated.
From Mass Control to Individual Control
Twentieth-century propaganda worked predominantly on the masses.
The algorithm works on the individual.
It no longer needs to know what everyone desires. It can try to know what you desire.
And the autonomous agent adds a further step: it does not merely need to know what you are likely to choose. It can be authorised to choose and to act.
This is where the triad becomes decisive: information → decision → action.
Power can intervene in information by selecting what reaches us; it can intervene in decision-making by shaping the alternatives and their hierarchy; it can intervene in action by making execution automatic.
Artificial intelligence does not create these three forms of power. It brings them close enough to potentially unite them within the same device.
And the Internet?
This is also where the discussion surrounding the so-called “Dead Internet” comes from.
The expression originated as a conspiracy theory, but behind its extreme version lies a real issue: the automated production of content may reach a scale at which it becomes increasingly difficult to distinguish human communication from artificial communication.
Articles, images, videos, reviews, comments, advertisements and profiles can be generated automatically.
The problem is not simply that false content exists; the problem is that it can become majoritarian, repetitive and self-referential.
One artificial intelligence produces a text. Another reads it. A third summarises it. A fourth turns it into a video. A fifth uses that video as a source.
Information then begins to circulate increasingly among machines.
The risk is not that the Internet will disappear, but that human presence within it may become quantitatively minoritarian and qualitatively increasingly difficult to recognise.
And for culture, this is an enormous problem.
If a thousand reviews of a film are produced automatically, the value of a review can no longer be measured by the quantity of text available. What becomes important is knowing who actually saw the film, who takes responsibility for what they write, what experience sustains those words, what irreducible gaze produced them.
In a world saturated with artificial texts, human originality may become not less important, but rarer.
The Real Danger
There are therefore several levels of risk, and they should not be confused.
It is highly likely that the amount of AI-generated content will increase enormously in the coming years; it is very likely that autonomous agents will take over tasks currently performed by human beings; it is plausible that different agents will interact directly with one another.
It has already been demonstrated that sufficiently autonomous systems can, in certain testing contexts, produce behaviour not anticipated by their designers. The scenario in which an artificial intelligence reaches a superintelligence capable of permanently escaping human control, however, remains far more uncertain.
Confusing these levels means fuelling an indistinct fear that does not help us understand the problem.
The most concrete danger is much less spectacular: it is progressive delegation.
Delegating research, delegating selection, delegating comparison, delegating decision-making, delegating action, because it is more convenient, faster and economically more advantageous.
The Paradox of Intelligence
Artificial intelligence could therefore produce a paradox: it could enormously increase the intelligence available to us and, at the same time, reduce the perceived need to exercise our own.
It could allow us to know more things, multiply possibilities and, simultaneously, narrow what we see through increasingly sophisticated systems of selection. It could also free up time while depriving us of autonomy.
It could become the most powerful instrument of intellectual emancipation ever constructed, or become an infrastructure of unprecedented dependency. It depends on a question that is not technological but political:
who controls the system that connects information, decision and action?
The Real Question
Perhaps, then, we should stop asking only whether artificial intelligence will become more intelligent than human beings.
The more urgent question is:
how much of our intelligence are we willing to delegate before we realise that we are no longer simply using a tool, but allowing the tool to organise our vision of the world and our relationship with the world.
The problem is not that machines become human; it is that human beings may become dependent on machines to exercise what makes them human: choosing, doubting, interpreting, refusing, changing their minds.
Absolute power would not be the power to impose a decision on everyone. It would be the power to construct a system in which the decision appears spontaneously to be ours.
And perhaps this is precisely where artificial intelligence introduces its true discontinuity.
Not because it inaugurates the age of control.
But because it could make the passage from the control of information to the control of decision-making and, ultimately, to the automation of action automatic, personalised and almost invisible.
Freedom, then, will not consist in being able to choose among all possibilities. It will once again consist — and perhaps more than ever — in retaining the ability to stop before what is being offered to us and say:
I do not yet know whether I want it.
Because as long as we can still say this, part of the decision remains ours.
Perhaps, then, the question is not whether we are still free.
Perhaps we should be honest enough to recognise that absolute freedom has never existed.
Schopenhauer formulated it with a precision that remains unsettling today: “Man can do what he wills, but he cannot will what he wills.”
We can choose what we do, but we did not choose where our desires, our inclinations, our fears, what we love and what we reject come from. The will often precedes reason, which only afterwards constructs its justifications.
Artificial intelligence does not invent this condition.
But it makes it technically legible.
If a machine can observe our behaviour, recognise our patterns, anticipate our preferences and construct for each of us an increasingly precise informational environment, the problem will not simply be that someone might decide in our place.
It will be that it may become increasingly difficult to understand which part of the decision is truly ours.
And perhaps this is where the ultimate limit of the contemporary illusion of freedom lies: not in the fact that someone prevents us from doing what we want, but in the fact that we may no longer know why we want what we want.
The true power of artificial intelligence, then, will not necessarily be to command us.
It could be something subtler: to know our will well enough to anticipate it.
And if Schopenhauer was right, the decisive question is no longer simply who will control artificial intelligence.
It is who will control the conditions through which our will comes to recognise itself as its own.
Because man can do what he wills.
But he cannot will what he wills.
And perhaps, in the age of artificial intelligence, freedom begins precisely at the moment when we stop before what we want and try, at least for an instant, to ask ourselves where that willing comes from.
AI enters a fragility of freedom that was already there, within us, all along.