Anonymous Power: Government That Nobody Governs

We used to think that behind a decision there was at least a recognisable face. A manager hired or fired, a bank officer assessed a loan application, an editor decided which article would be published, a civil servant examined the file in front of them. These decisions might be unfair; the person deciding could be prejudiced, self-interested or arbitrary. But at least there was somebody in front of us to speak to, to question, to object to. Today, faced with many of the decisions that affect our lives, we can find no human being. Our loan application is refused by “the system,” our job application is eliminated in an automated assessment, our social media account is closed for an unclear rule violation, our insurance premium rises according to our risk score. A decision has been made, but when we ask who made it, nobody is there.

This new form of power does not shout, does not climb onto a platform, does not construct sentences of command. It has no uniform, no office, no leader everyone knows. Most of the time it merely sends us a notification: “Your request has not been deemed appropriate.” “Your content violates community standards.” “Your application did not meet the assessment criteria.” “Your account has been restricted for security reasons.” What these sentences share is that the decision is definite while the agent is indefinite. There is an outcome, but no subject to take responsibility for it. This is exactly where anonymous power appears: a force able to govern people’s lives without showing itself as a governor.

There Is a Decision, but No Decider

Anonymous power does not consist only of algorithms. We already knew its older forms in bureaucracy, in large companies and in complex hierarchies. When an employee asks why they were dismissed, their manager may say “company policy.” An official may carry out a transaction they can see is plainly unjust by saying “those are the regulations.” A bank employee may say they understand the customer’s situation but the system does not permit it. Not all the people uttering these sentences are necessarily lying. Their authority to decide really is limited. But the problem begins precisely here: in an order where everyone performs only a small operation and nobody carries responsibility for the whole, who will account for the outcome that emerges?

A company dismisses hundreds of its employees. The human resources unit says the decision came from management. Management points to economic conditions. Economic managers explain the requirements of the market. Shareholders say they want to protect the value of their investment. The consultancy firm points to an efficiency report. Those who prepared the report state that they merely analysed the data. In the end hundreds of people are unemployed, but nobody says “I made this decision and I am responsible for its consequences.” Everyone is the small implementer of a necessity larger than themselves. Power does not vanish as it is broken up among individuals; on the contrary, it becomes far harder to object to.

For you can oppose a recognisable manager. You can criticise their decision, demand their removal or fight to have their authority limited. But how does one argue with “the market,” “the procedure,” “efficiency,” “the risk model” or “a system requirement”? These are not speaking subjects, yet they can issue judgements more definite than people’s. Moreover, they present themselves as necessity rather than choice. When a manager says “this is what I want,” the arbitrariness of the decision is visible. When they say “the data show this,” the same decision looks as if it were purified of personal will — scientific and inevitable.

In such a system nobody even needs to have bad intentions. While everyone thinks they are doing their job properly, an extremely unjust outcome can emerge. A system in which people are divided into small pieces of responsibility leaves the responsibility for large outcomes unowned. At every link of the chain producing the decision there is somebody; yet looking at the chain as a whole, there is nobody. This is where the power of anonymous power comes from: it works everywhere and nowhere entirely reveals itself.

An Algorithm Is Not a Neutral Referee

Algorithmic decisions have strengthened this anonymity further. For the algorithm is presented as a neutral referee purified of human prejudice. A person can show favour, get tired, be swayed by emotion; a machine looks only at the data. At first sight this seems quite persuasive. We might all want an order in which acquaintance, personal relations and arbitrary decisions are not decisive. But the moment we think the algorithm is a neutral intelligence deciding on its own, independent of people, we miss an important truth: the algorithm does not itself decide which data it will look at, or what it will count as success, risk, reliability or threat.

If a hiring system examines the characteristics of employees the company counted successful in the past, it can turn the company’s past preferences into the criterion of the future. If in the past managerial posts were given to people from particular social groups, the system can read this not as an inequality but as a data pattern correlated with success. Even though no written rule says “eliminate people from this group,” past discrimination goes on living as a mathematical prediction. Instead of removing prejudice, the algorithm can give it an impersonal, technical appearance.

A similar thing can happen in credit assessment. The system says it does not look directly at a person’s identity; but it decides through hundreds of indirect data such as the district they live in, their shopping habits, their educational background, their income pattern, their phone use and their social circle. Class inequalities are thereby converted into a numerical score called “risk.” Someone living in more fragile conditions because they are poor is counted more risky; because they are counted more risky they reach only more expensive credit; expensive credit increases their fragility further. The system then records this outcome as new data and concludes that its first decision was correct. Inequality has not merely gone unmeasured; it has been turned into a self-confirming cycle.

The issue here is not that algorithms always decide badly. In some situations they can produce more consistent, faster and more accurate results than people. The problem is that how those results are produced becomes invisible, and that passing through a mathematical operation makes them appear beyond debate. Numbers hide the value judgements behind the decision. The sentence “this person is 74 per cent risky” looks as if it carried scientific certainty; yet how risk is defined, which data are used and who pays the price of a wrong decision rest wholly on social choices.

An algorithm does not merely measure the world; it also carries a view of the world about what is worth measuring. When we measure an employee’s success by the number of emails they send, their keyboard movements or the time they stay online, we have made a particular decision about what work is. Time spent thinking, support given to a colleague, intuition earned through experience or invisible care labour can fall outside the measurement. Then we begin to take what we cannot measure as worthless and what we can measure as real. And people turn towards producing the signals the system recognises as success rather than doing their work well.

Finding No Door to Object At

The justice of a decision does not depend only on its producing the correct result. The person affected must also be able to explain themselves, to learn the reasoning and, if there is an error, to ask for it to be corrected. Yet the most distinctive feature of anonymous power is that it removes the door at which objection could be directed. When your social media account is closed, what usually appears is a standard form. You fill in the form and a few seconds later receive another automatic message saying your request has been re-examined. You cannot know whether a human being really looked at your file in so short a time. The system that made the first decision and the system assessing your objection may be the same.

On digital platforms we meet countless everyday examples of this. A seller’s account can be frozen for “suspicious activity”; someone who earns their living from that platform may be unable to reach a real official for weeks. A content creator’s audience, built up over years, can be taken away without any clear statement of which expression constituted a violation. A courier, driver or temporary worker can become unable to get work because of a low score, without being able to learn what assessments produced it. A customer may have made an unjust complaint, the location system may have malfunctioned, traffic conditions beyond the worker’s control may have lengthened the delivery time. But the system looks not at the account of the person living through this but at the few numbers it recorded.

When a person becomes unable to defend themselves against a decision about their own life, they lose not only their right; they also lose the authority to narrate their own reality. “You were not there,” says the system, though they were. “You delayed the delivery,” it says, though the road was closed. “You showed suspicious behaviour,” it says, though they merely carried out an unusual but wholly legitimate transaction. When the system’s record conflicts with a person’s lived experience, the record usually prevails. For the record looks measurable, storable and reportable; a person’s word is taken as subjective.

When this situation is carried into public services it can produce far graver consequences. A wrong decision by automated systems used in areas such as social assistance, healthcare, educational support or security assessment does not merely mean losing access to an application. A person can lose their housing support, their access to treatment or their liberty. Even so, when the system is not explained on grounds of commercial secrecy, security or technical complexity, the person cannot even learn why they were punished. They are forced to defend themselves against an invisible accusation.

Reducing the right of objection to the existence of a form is therefore not enough. For objection to be real, one must reach somebody with the authority to change the decision. The reasoning for the decision must be explained intelligibly, the data used must be visible and wrong data must be correctable. Otherwise the objection mechanism becomes a decoration that makes the decision look fair. The system pretends to listen to us while leaving no space in which what we say could change the outcome.

Human Oversight Is Sometimes Only a Decoration

One of the answers often given when algorithmic decisions are criticised is that the final decision is made by a human being anyway. We are told there is an officer behind the screen and that the algorithm merely makes a recommendation. On paper this looks reassuring. But if that officer has no real possibility of questioning the algorithmic recommendation, a human’s presence in the process changes little by itself. With hundreds of files in front of them, a narrow window of time and a definite-looking score produced by the system, most officers tend to approve the recommendation. For opposing the system requires extra explanation, responsibility and risk; complying with it is safe.

Consider a bank employee. Suppose the system has flagged a loan application as high risk. If the officer approves that decision, any later problem can be seen as “the model’s prediction.” But if they override the system, approve the loan and the customer runs into difficulty, personal responsibility for the decision can be loaded onto them. In such an order a human having the last word is a merely formal authority. In reality a structure has been built in which not changing the system’s decision is rewarded and changing it is punished.

In time employees learn not to trust their own judgement. A teacher knows the student but looks at the performance table; a doctor listens to the patient but follows the recommendation on the risk screen; a manager sees an employee’s effort but takes the scoring system’s result as the basis. Nobody openly says the machine is always right. Yet all the institution’s incentives, records and structures of responsibility push people to comply with the machine. The algorithm thus becomes the actual decider although it is not formally the decision-maker.

Moreover, as automated systems spread, managers also find the opportunity to load their own decisions onto the algorithm. Managerial choices such as increasing workloads, reducing staff numbers, changing wages or excluding certain customers can be presented under the name of “data-driven optimisation.” Yet which target will be optimised was again determined by people. The company chose more profit, less cost or faster delivery; it pushed employees’ health, customers’ particular circumstances or social effects into second place. Here the algorithm is not the owner of the decision but the curtain behind which it hides.

The question “did a human or a machine decide?” is therefore not enough on its own. More important is who set the target, who built the system, who selected the data, who pays the price of a wrong decision and who holds the authority to change the outcome. Sometimes a human presses the button on the screen, but the decision was made much earlier, while the system was being designed. The officer merely applies a preference established in advance to individual files.

Making Invisible Power Visible Again

Fighting anonymous power does not require rejecting technology entirely. The matter is not simple enough to be solved by shutting down a few computer programs. When algorithms are removed, fairer people do not always take their place. Arbitrary managers, personal prejudices and networks of acquaintance also produce serious injustices. The real problem is that power is presented as a technical necessity and thereby taken out of political debate. If a system affects people’s lives, how that system will work must also be a shared matter.

First we must bring back simple but uncomfortable questions: who benefits from this decision? Who pays the price of a wrong one? Which data were collected and which ignored? Through which values were success or risk defined? Whom can a person who thinks the system is mistaken actually reach? Is there anyone with the authority to change the decision? These questions may look like technical details, but in fact they touch power’s oldest matter: who decides, on whose behalf and by what right?

Accountability does not consist of showing everyone the source code of every algorithm. Most people cannot examine thousands of lines of code; even if the code is open, the data used, the corporate targets and the manner of application can remain invisible. What people need is to learn the intelligible reasoning for a decision, to reach the information used about them, to correct errors and to find a real responsible party in front of them. The sentence “the system decided so” must never be the final answer for any institution. For the system did not build itself, did not choose its targets and did not decide which errors were acceptable.

Perhaps anonymous power’s greatest success is convincing us that there is nobody governing any more. The market wants it, the data show it, the algorithm determines it, the procedure requires it, the system does not allow it… As these sentences are repeated, orders built by human hands begin to resemble laws of nature. Yet every procedure was written by somebody, every criterion was selected by somebody, every piece of software was designed towards particular goals. All of this can be changed. That they look unchangeable stems not from their really being unchangeable but from the hiding of who holds the authority to change them.

In periods when power was visible there was at least an address to which anger could be directed. In today’s nameless government a person finds themselves arguing with a void. The call-centre worker has no authority, the manager is bound to the system, the company that built the system is protected by contract, and the company points to the requirements of the market. Responsibility is thus passed continually to the next link and finally returns to us: if we could not get the loan we are risky, if we were not hired we are not good enough, if our content is not seen it is not interesting, if our account was closed we must have done something wrong. Instead of letting itself be questioned, the system makes us doubt ourselves.

Perhaps the first objection should begin exactly here: in refusing to accept that every automated outcome is our personal fate. In remembering that the score we are given does not define us entirely, that data are not our life itself, and that the sides of us which cannot be calculated cannot be counted worthless. Then we must insist on looking for the agent of the decision. Without making visible again the institutions, interests and choices behind what we call “the system,” it is not possible for us to hold anyone to account.

For in a place where nobody is said to govern, usually it is not true that nobody governs. Somebody sets the targets, somebody profits, somebody is protected from the consequences; it is only that these relations become invisible behind the decision screen. Anonymous power is not a power that has lost its face but one that has learned not to show it.

Freedom, meanwhile, is not merely demanding that a decision be made by a human being. It is that the people affected by a decision can take part in how it is produced, question its reasoning and, when necessary, change it. Otherwise code takes the place of the manager, a score takes the place of the order and a blocked account takes the place of punishment; but the relation of being governed remains exactly as it was.

Perhaps one of the most important questions of our age is no longer “who governs us?” but “why does whoever governs us present themselves as nobody?” For where there is no responsibility, power does not disappear. It merely withdraws to a safer, quieter place that is harder to object to.

In the current form of anonymous power, generative artificial intelligence and automated decision assistants take up an ever larger place. The summary of a CV, a patient’s priority ranking or the first response given to a citizen’s application can be prepared by a machine. Even if the human officer thinks they are making the final decision, the frame that reaches them has already determined which information will look important. When a system reduces a person to a short description, the officer usually has neither the time nor the incentive to step outside that description.

A human being present at the final decision is therefore not a sufficient guarantee. That human must have the authority, the time, the knowledge and the institutional protection to question the recommendation. Otherwise “human oversight” remains an ethical decoration that circulates responsibility between the technology and the officer. Real accountability requires making the whole decision chain visible, from the manager who bought the system to the company that built the model, from whoever selected the data to the person who applies it.