Training cognitive effort to guard against the comfort of Ai

Whoseresponsibility is it if artificial intelligence produces damage. And where is the truth to be found in the current age of cognitive flattening and the quick and easy dissemination of false news and unverified facts. These two questions were at the centre of the debate that took place yesterday afternoon during the third day of 'Talk to the future', the event organised by the Milan Bar Association and dedicated to the confrontation between Ai, law and society.

It starts with an event that was unthinkable, until recently, for liberal democracies: the cancellation of elections - it happened in Romania on 6 December 2024 - due to the large amount of foreign interference on social media by a foreign power. "The novelty does not lie in the attempt to influence the outcome of an election, but in being able to do so remotely, easily and at little economic expense," explains Matteo Flora, entrepreneur and adjunct professor at the European School of Economics and the University of Pavia. "It has been shown both that the Ai convinces people more of a political idea than a human being and that the models that work best are those that mention the most because they are more responsive to our prejudices".

The scientific studies recalled by Elisabetta Fersini, professor at the Bicocca University and president of the Italian Association of Computational Linguistics, show the cognitive settlement one experiences when using Ai tools, the sense of understanding and acceptance similar to that experienced in the therapist-patient relationship, and the importance of defining, at a regulatory level, what content is safe for us as users.

"Regulatory guardrails certainly need to be put in place, so much so that today the legal frontier is questioning the liability of the algorithm," comments Alessandro Mezzanotte, councillor of the Milan Bar Association. "In my opinion, some form of liability on the part of platforms must emerge at the regulatory level, especially in relation to their design. Indeed, the class action injunction underway at the Court of Enterprises of Milan against Meta and TikTok, brought by Moige, is very interesting'.

Artificial intelligence is carrying out a process that began decades ago with the birth of Google, namely usingcognitive shortcuts, losing the ability to ask questions, because the machine is too comfortable. Ai, moreover, is affected by what in the Anglo-Saxon world is called 'sycophancy', i.e. an assertive and condescending mode of response that is particularly worrying when one considers that eight out of 10 minors in the United States prefer to talk about serious matters with Ai instead of with a human being, convinced of the neutrality of the technology. "Other data show that those who use Ai are more convinced they are right and less willing to apologise, while seven out of 10 people don't trust anyone whose opinion differs from their own," Flora concludes. "The biggest investment we can make, therefore, is training ourselves to make cognitive efforts because that is where the truth lies."