We often judge the power of a technology by what appears on its screen. It calculates faster, writes more smoothly, and compares more data. Yet the kind of power that changes a society extends beyond a product’s features. When a system begins to influence who becomes visible, which information rises to the top, and what decisions seem reasonable, it enters the way our lives are governed.

A September 23 report on AI and human development from the United Nations Office for Partnerships brings together views from different disciplines. Drawing on conversations in Cairo, Mexico City, Xi’an, and Rome, it asks questions beyond technical performance. It asks what happens to human agency and to the relationships that hold a society together.

Computer science helps us understand how a system works. Public administration adds another question: what authority have we given it? International business asks who owns the infrastructure, who bears the cost, and who can negotiate the terms. If we separate these questions too sharply, we may miss the different consequences of the same development.

A model’s strong performance does not make every decision by the institution using it legitimate. Whose data does it use? For what purpose, and within what limits? A company’s recommendation engine cannot be held to the same standard as a system that helps a public agency decide access to a right or service. As the consequences of a decision grow, so should the clarity of its explanation.

Power lies beyond the model

Power also resides in the conditions of access: computing infrastructure, data, expertise, distribution channels, and the habits of users. Control over one of these can influence the others. That is why I do not think the number of competing products tells us enough about competition.

As an institution becomes more dependent on one provider, it needs to consider the future cost of today’s convenience. Can it move its data elsewhere? How difficult would changing its workflow become? If contract terms change, does it have a real alternative? These may sound like technical details. In practice, they shape an institution’s freedom to decide.

For the same reason, I do not understand digital sovereignty as simply putting a domestic name on a product. A country also needs the capacity to understand its decisions, oversee its infrastructure, change providers when necessary, and answer to the public. A closed system with a local label does not automatically solve problems of accountability or competition.

Nor can every country reasonably build every layer of the infrastructure on its own. International cooperation, shared research, and the exchange of expertise matter. The distinction is between buying a service without understanding the dependency it creates and entering a partnership on terms that can be examined. In a global technology, independence often means managing connections well.

From the United States and Europe to countries with very different levels of capacity, no single measure of success is sufficient. The speed of innovation may matter most in one place, access to services in another, and computing costs in a third. If global rules treat the needs of the strongest companies and states as universal, they may narrow everyone else’s room to act.

Oversight belongs inside progress

I resist the idea that oversight and innovation must be opposing goals. Good oversight can make innovation more trustworthy. Unnecessary and vague obligations can burden smaller ventures, of course. Rules should therefore focus on uses with substantial consequences, be understandable, and remain proportionate to real risk.

When a system’s recommendation shapes an important decision, an institution cannot transfer responsibility to a screen. We should know who made the final decision, what information was used, and how an appeal will be heard. “The algorithm said so” is no substitute for a reason. Human oversight, likewise, means more than someone clicking an approval button.

That person needs the knowledge and time to question the recommendation, the authority to stop a flawed outcome, and the freedom to do so without punishment. Otherwise, a human presence offers only the appearance of a safeguard. An old question of public administration returns in a new setting: do authority and responsibility still sit together?

There is an international question of fairness, too. A country that only consumes a technology occupies a different position from one that helps research, set standards, and oversee it. A shared AI order requires institutions that allow countries with limited capacity to participate meaningfully. Otherwise, “partnership” may mean learning decisions after others have made them.

I do not claim that machines will suddenly hold all power. A nearer possibility is that power exercised by people and institutions will become more concentrated among the actors behind the technology. To understand that shift, we need to look at contracts, infrastructure, institutions, and bargaining relationships as closely as we look at software.

We will continue to ask which model is smartest. We should ask, with equal seriousness, which arrangements are most accountable. A society’s technological maturity lies not only in possessing powerful tools, but in explaining for whose benefit they are used and within what limits. That is how people retain their place on the new map of power.