Leggi in ItalianoSomething curious seems to be happening in our relationship with artificial intelligence. The more effective these tools become, the more we tend to trust their answers. So far, nothing unusual. The problem begins when trust gradually turns into something else: a kind of faith in AI, in its capacity to know, explain and judge virtually everything.
It is not difficult to see why. We ask ChatGPT, Gemini or other systems about even highly complex issues and receive immediate, well-constructed and often correct answers. After a while, we stop asking ourselves where those answers come from. The response itself becomes the source. And when something does not convince us, we may simply ask the same AI to verify what it has just told us — as though consulting the oracle a second time were enough to produce an independent confirmation.
I notice it in a much more everyday situation as well. From time to time, talking with my daughter, I realise that the level of trust she places in ChatGPT’s answers is even higher than the trust she places in what I tell her. I admit it makes me smile, but it also gives me pause. It is not so much a matter of parental authority — which has always been debatable — as of the fact that an answer produced by a machine sometimes seems to enjoy a presumption of reliability that we humans no longer have. As though the apparent absence of personal opinions, interests or biases automatically made that answer more objective.
Examples abound. There are now several cases in which lawyers have filed non-existent court decisions, generated by artificial intelligence, before tribunals without bothering to verify their authenticity. These are not first-year students but professionals who should know perfectly well the importance of sources and their reliability. The problem, then, is not only that AI makes mistakes, but above all our willingness to accept its answers without subjecting them to critical scrutiny.
Naturally, I do not think we should systematically distrust artificial intelligence. That would be an unreasonable and pointless position. I use these tools every day and continue to regard them as one of the most remarkable innovations of recent decades. But precisely by using them, I realise how easy it is to move from assistance to delegation. First we delegate certain operations, then the search for information, and finally, almost without noticing, part of our own judgment.
The real novelty may not be that machines can err, but that they can lead us to give up verification precisely because they work so well. This seems particularly significant for schools, universities and research, where the result matters, of course, but the path by which we reach it should matter just as much.
To speak of a new religion is obviously a provocation. But not an entirely unfounded one. We have created something that appears to answer every question, that speaks with confidence, that possesses seemingly unlimited knowledge and whose workings remain essentially incomprehensible to most people. It would be surprising if a form of almost unconditional reliance did not develop.
And yet, between trusting a tool and believing in a tool there is an enormous difference. Trust should be open to being tested. Faith, by its nature, can survive even in the absence of evidence.
The risk, in short, is not so much that artificial intelligence might become a new deity. It is that we might turn it into an oracle, forgetting that critical thinking does not consist merely in finding answers, but above all in knowing how to question them.
And perhaps the real paradox is precisely this: we have built an extraordinary tool to expand our intellectual capabilities, yet we risk using it to exercise them less and less.