AI does not need to become conscious to reduce freedom. It only needs to help powerful systems make decisions faster, cheaper and harder to challenge.
The public argument about AI often gets pulled into the wrong story. We are asked to imagine machines waking up, becoming conscious, and deciding to rule us. That makes good cinema, but it is also a useful distraction.
The more ordinary danger is easier to see. AI can become the hidden machinery between people and power. It can help decide who gets a benefit, who gets watched, who gets flagged as risky, who gets a job interview, who gets insurance, who gets housing, who gets access to a service, and who is pushed to the back of the queue.
That does not require a robot ruler. It only requires an institution with a problem to solve, a budget to cut, and a system that promises speed.
The real danger is automated power
The danger is not that AI has opinions. The danger is that AI can be used by systems that already have power over people.
A council may use a system to sort applications. A department may use one to detect fraud. An employer may use one to screen job applicants. A platform may use one to decide what is seen, hidden, boosted, or buried. A police force may use one to identify patterns. An insurer may use one to price risk.
Each step can be presented as technical. Each step can sound harmless: it is only a tool, it only assists, a human is still involved, the system only makes recommendations, the final decision is not fully automated.
That language matters.
Power often protects itself by making decisions sound procedural. The more technical the process becomes, the harder it is for the affected person to see where the decision was really made.
A person may be told no. They may not be told why. They may not know what data was used. They may not know whether a score, model, flag, or risk category shaped the outcome. They may not know who can change it.
At that point, freedom has already been reduced. Not because a machine has become conscious, but because the route to challenge has been made unclear.
Speed is not neutral
AI is often sold as efficiency.
That is not automatically wrong. Some public systems are slow, overloaded, and painful to use. Staff are under pressure. People wait too long. Mistakes happen. Repetition wastes time. Used carefully, technology can help.
But speed is not neutral when the system has power over people.
A faster wrong decision is still wrong. A cheaper unfair decision is still unfair. A smoother exclusion is still exclusion.
If AI helps an organisation process more people with fewer staff, the important question is not only whether the system is efficient. The question is simpler: can the person understand the decision, challenge it, and reach a human being with authority to change it?
If the answer is no, the system has not simply become modern. It has become harder to resist.
Human in the loop is not enough
One of the most comforting phrases in AI policy is human in the loop. It sounds reassuring. It suggests that a person remains responsible.
But the phrase can hide more than it explains.
A human may be in the loop too late. They may only see the output after the system has already shaped the options. They may be under pressure to agree with the machine. They may not understand how the score was produced. They may not have time to investigate. They may be trained to treat the system as objective.
That is not real oversight.
Real oversight means a person can question the system, reject the system, explain the decision, and be held responsible for the outcome. A human rubber-stamping an automated recommendation is not accountability. It is theatre.
The computer says no
For ordinary people, the danger will not always look dramatic.
It may look like a form that never leads to a person, an appeal process that asks the same questions again, a risk score nobody can explain, or being told that the decision was made according to policy.
It may look like a public body saying the system only supports staff, while staff behave as if the system’s answer is final.
This is how freedom can shrink without a single new prison or police power.
The citizen is still technically free. They can still apply. They can still complain. They can still appeal. But the process becomes harder to understand, harder to challenge, and harder to survive.
That matters most for people who already have the least spare capacity: disabled people, poor people, migrants, people with unstable housing, people with complex health needs, people who struggle with forms, and people who cannot afford legal help.
Automation does not harm everyone equally. A middle-class person with time, confidence, literacy, and professional support may eventually get through the system. Someone exhausted, poor, distressed, or already treated with suspicion may simply disappear from the process.
The political question
The real political question is not whether AI is clever. The question is who gets power from it.
If AI gives staff better tools while preserving human judgement, public explanation, and appeal rights, it may help. If it gives institutions cheaper ways to sort, rank, deny, monitor, and manage people, it becomes a control system.
That distinction is the point. The problem is not automation itself. The problem is automated power without visible responsibility.
A democratic society should not allow major decisions about people’s lives to disappear into systems they cannot inspect, understand, or challenge.
The public test
Every AI system used in public life should face a simple test.
Can the person affected by the decision know that AI was used?
Can they understand the main reason for the decision?
Can they challenge the decision before a real person?
Can that person change the outcome?
Can the public see which systems are being used, where, and why?
If not, the system should not be treated as safe just because it is efficient.
Freedom does not only depend on the right to vote or speak. It also depends on the ability to deal with power when power makes a decision about your life.
AI makes that decision-making faster. It can also make it more hidden.
That is why the danger is not the machine waking up. The danger is the machine being useful to institutions that already prefer the public to be quiet, processed, scored, and managed.
AI will not need to rule us. It may only need to make control cheaper.