Leggi in ItalianoWhen it comes to artificial intelligence and journalism, the question often asked is whether an automated system can write an article in place of a journalist. In academia, the question is similar: will it be able to produce essays, prepare lectures or contribute to research? These are important questions, but there is another one that may matter more: what happens to our capacity to know and freely discuss reality when the tools used to search, select and interpret information increasingly depend on a small number of companies?
Journalism and scientific research have different tasks and methods. A journalist reconstructs facts of public interest, gathers testimony and verifies documents; a scholar formulates hypotheses, compares interpretations and submits conclusions to criticism. Both, however, contribute to an essential condition of democracy: the existence of people and institutions capable of producing knowledge independently of the accounts offered by those who hold power. Freedom of the press and academic freedom, therefore, do not protect only two professional categories. They also protect the possibility, for everyone, of forming an autonomous judgment.
I do not believe this autonomy is defended by rejecting artificial intelligence. A newsroom can use it to examine large numbers of documents, compare sources and work with materials in different languages. In research, it can help identify connections and address problems that would otherwise require far more time. Yet the speed with which an answer is obtained does not guarantee that the answer is well-founded. A tidy, confident and seemingly exhaustive text may conceal errors, questionable assumptions or important elements that have not been considered.
This is a point worth pressing, including for those who regularly work with legal documents and sources. A summary may be useful for orientation, but it does not replace reading the provision, the judgment or the original document. Sometimes a single word, a qualification or an overlooked passage is enough to change the meaning of a conclusion. The problem, then, is not only whether an answer is right or wrong: it is whether one can reconstruct how it was reached and on what evidence it rests. Without that possibility, it becomes harder to challenge it as well.
We are accustomed to thinking about threats to freedom of information in terms of censorship: a story that cannot be published, a suppressed opinion, a critical voice silenced. These threats remain real. But the ability to influence what we know begins before publication: in the choice of sources, in the selection of materials, in the order in which they are presented and in what is left out. If newsrooms, universities and research centres entrusted a growing share of these operations to the same technological intermediaries, they might end up sharing not only the tools but also their limitations and selection criteria.
There is no need to imagine deliberate censorship. Commercial choices, the characteristics of the data used to develop the systems and their technical limitations can all affect the results. The risk increases when we tend to regard a selection as neutral simply because it was performed automatically. The absence of an explicit prohibition does not mean that all information has the same chance of being found and taken into account.
There is also another, less visible form of dependence. A newsroom may produce more articles thanks to efficient tools and, at the same time, have fewer journalists capable of conducting investigations or verifying testimony directly. A university may increase the number of publications without improving the quality of research. The ability to produce texts and the ability to acquire knowledge are not the same thing. Independence requires resources, expertise, time to go deeper and the freedom to reach uncomfortable conclusions. It is not enough to declare that humans remain in control of artificial intelligence: the question is whether those who verify the output are genuinely able to understand it, correct it and, if necessary, reject it.
On this point, a comparison with intelligence work is instructive, despite the different purposes. Journalism, scientific research and intelligence share certain requirements: evaluating sources, seeking independent corroboration, distinguishing facts from hypotheses and taking alternative explanations seriously. In all these fields there is a risk of privileging information that confirms what one already believes. Artificial intelligence can amplify this problem when it rapidly offers a plausible reconstruction that aligns with existing expectations. Verification serves precisely this purpose: to keep open the possibility of being contradicted by the facts.
The question ultimately concerns institutional responses. Countering information manipulation and the deceptive use of new technologies is necessary. Yet it would be contradictory to defend pluralism by granting any single authority the exclusive power to determine which interpretations are admissible. What is needed instead is accessible sources, the possibility of independent verification, protection of journalism, freedom of research and the means to critically examine even the technologies on which we depend. Regulating these systems is important; no less important is ensuring that regulation does not narrow the space for public debate.
The question that interests me, then, is not so much whether artificial intelligence will replace journalists or scholars. It is whether we will know how to use it without losing the competences and the autonomy needed to judge what it proposes. New technologies can enormously expand our possibilities for knowledge, but no answer, however fast and convincing, can entirely replace the verification of sources, the confrontation of views and the exercise of doubt.
Freedom of the press and academic freedom serve to guarantee precisely this: that it is always possible to discover something that contradicts what we thought we knew.