Artificial intelligence is often presented as a neutral tool: a capacity available to anyone who wants to use it. But no technology reaches an empty space. Someone defines their objectives, possesses infrastructure, sets conditions for use and decides which people will have to adapt. That’s why talking about artificial intelligence is also talking about power.
The word may be uncomfortable, though it should not. Power does not necessarily mean bad intention. It means the ability to decide and to affect. Recognizing it allows asking questions that otherwise stay out: who participates, who gets the benefit, who takes the risk and what choice does anyone disagree with.
The designer also sets priorities
Each system needs to decide what counts as success. It can be speed, cost reduction, number of responses, or time of use. These metrics seem technical, but they express a priority. If success is to close queries quickly, you may miss the conversation that a person needs. If success is to be classified efficiently, perhaps there are simplified paths that do not fit into a category.
Design decisions are never innocent when they affect other people. That is why organizations must be able to explain what objective they pursue and what limits they have set. It is not enough to say that the tool works; it is necessary to say for who it works and what happens when it does not work.
It also matters who can question that design. An expert person can detect a technical problem; the recipient of the service often detects a problem of experience, language or treatment. Both knowledge are necessary. Leaving one of them out does not simplify the decision: it makes it more fragile.
The choice is not the same for everyone
There are voluntary uses, such as testing a creative tool. And there are uses that come incorporated into a process, a job or a necessary service. In the latter, asking for consent is not enough if refusing means staying without alternative. Freedom needs a real option, not a box that no one can mark without paying a cost.
Accessibility is a concrete way of sharing power. Keeping human attention, explaining decisions in clear language, and providing channels of correction allows not only those who know how to move through an interface to participate. It also protects those who have less time, less resources, or a situation that the system does not understand.
This is especially relevant for minors, older people and groups exposed to discrimination. A system that seems the same for everyone can widen inequalities if it requires devices, knowledge or data that not everyone can provide. Equity is not about applying the same screen to everyone; it is about not turning a difference of access into a permanent disadvantage.
Sharing control improves decisions
Sharing control does not mean that each person has to understand the model inside. It means creating spaces where decisions can be explained, discussed and corrected. A test with affected people, an impact assessment, an incident protocol or a review date are simple mechanisms to prevent power from being locked in who buys or configures the tool.
Transparency serves when it gives the ability to act. A document that is incomprehensible can fulfill a formal obligation and not help anyone. A useful explanation allows us to know what happens, who to turn to and how to ask for a change. That difference is the one that turns information into autonomy.
Organizations can also decide what they will not do. Do not use sensitive data when they are not essential, do not create imitations of real people, do not leave decisions with serious consequences without human review. These express limits have more value than a generic promise of responsible use because they allow to check if they are fulfilled.
Technology is more just when those who live its consequences have a real way of influencing it.
Talking about power does not force you to reject artificial intelligence. It forces you to use it with your eyes open. It can expand capabilities and facilitate tasks, but it can also concentrate decisions on places hard to see. The difference depends on how it is designed, explained and monitored.
The next time a tool seems inevitable, it is worth asking who decided it was. Often discovering that answer opens up a possibility that seemed closed: changing an adjustment, demanding an alternative, hearing an experience, or simply not adopting a solution that does not respond to a real need.
That is a good starting point for more democratic innovation: not asking people to adapt quietly, but building technologies that can respond to them.
The concentration of power may appear when a platform becomes the gateway to knowledge or when an automatic criterion ceases to be questionable. Avoiding this dependence requires preserving alternatives, forming the team and not confusing comfort with renunciation of control.
Sharing procedures and explaining boundaries reduces the distance between the person who sets up the tool and the person who has to live with it. It is not about making everyone a specialist, but about preventing technical knowledge from being a barrier to participation.
Educational, cultural and public institutions have an additional duty: they must be able to justify their decisions to different people. When they incorporate artificial intelligence, they must do so with clear information, human review and an alternative for those who need different treatment.
The conversation about power is, in the end, a conversation about dignity. A tool is worth if it leaves people with more ability to understand, decide and claim; it loses value if it makes them invisible within a process that no one can explain.
This commitment begins in a daily decision: to ask, explain and always leave an open door to correct.
