Embracing innovation but preventing it from eroding the foundations of coexistence
When it comes to artificial intelligence, attention has, for more than two years now, been captured by generative applications: tools that write texts, produce images, compose music or simulate conversations. It has been a surprising and partly destabilising phenomenon, because it has made it clear that activities that we thought were reserved for human ingenuity could be emulated by sophisticated statistical computing. The new frontier is represented by systems that no longer just wait for a command, but are able to move on their own, learn from interactions, take initiatives. We call them agents, but in reality they are much more than a technical function: they are tools that take on proactive behaviour, capable of organising work, making economic choices, directing our daily digital experience.
To understand the implications of this shift, it is not enough to look at efficiency or market opportunities. One must ask oneself what happens to the principles that underpin civil coexistence in Europe, and which derive from a long constitutional tradition. The first of these principles is the centrality of the human being. This is not an abstract slogan: it means that the person can never be reduced to a mere tool of technical efficiency. Yet, with intelligent agents, the temptation is precisely this: to let software take care of choices that used to belong to the individual sphere, turning man into an observer of the decisions taken in his place.
The second point concerns the ability to choose. European constitutions have always linked freedom to the ability to make informed decisions, with adequate information and without undue pressure. In generative systems, the problem was the quality of the content: how reliable it was, how much it could manipulate public opinion. With agents, however, a more insidious terrain opens up. It is not just a question of receiving correct information: it is the very logic of the decision that is in danger of being shifted elsewhere. The agent anticipates desires, constructs paths, realises actions before the individual has even decided. Thus freedom becomes apparent: the external forms of choice remain, but the authentic moment of personal deliberation is missing.
The third theme is the protection of the mind, understood as an intimate space for reflection and autonomy. We increasingly speak of cognitive freedom, a notion that did not appear in law textbooks until a few years ago. With generative models, there was a risk of stimulus overload, which could be disorientating but still left the responsibility of filtering to the individual. With agents, the dynamic is different: they become permanent companions who filter, select and organise experience for us. If this habit becomes entrenched, critical capacity can weaken. One no longer stops to question, one no longer seeks doubt: one accepts what the system proposes, as if it were the only possible way.
Faced with this transformation, European law cannot simply update the list of technical obligations. The Ai Act took an important step with the risk-based regulation model, but it is not enough. The dangerousness of an agent depends not only on the sector in which it operates, but on its very nature as an autonomous delegate. It is therefore necessary to think of new guarantees: for example, the possibility of interrupting the agent's action at any time, clarity as to when a decision has been made by software and not by the individual, clear legal liability in the event of errors or damage.
