The generative AI surge: between opportunity and risk

In a recent interview with Paolo Galli for Corriere del Ticino, Andrea Emilio Rizzoli, Director of the Dalle Molle Institute for Artificial Intelligence (IDSIA USI-SUPSI), analysed the growth of generative AI in Switzerland, reflecting on the reasons behind its success, potential risks, and the necessity for new digital skills.

According to the latest data from the Federal Statistical Office, nearly half of the Swiss population claims to use generative AI tools. This represents an extremely rapid spread, unprecedented in the recent technological landscape. For Andrea Emilio Rizzoli, a primary driver of this success is how seamless the human interaction feels: "If it is spreading so quickly, it is because the technology feels very close to us. We talk to these machines using natural language, which gives us the impression that they understand us."

While this immediacy makes AI particularly attractive, it also risks shifting the focus from the quality of the output to the mere convenience of the tool. Rizzoli specifically warns about the effects on learning and critical thinking. "If we delegate even the five-line summary of a two-page text to a machine, we will end up with summaries that are content-poor, and we will lose our own ability to abstract meaning from data and information." The risk, he adds, is a progressive erosion of our ability to process information independently.

To illustrate this phenomenon, the Director of IDSIA employs an effective comparison: "My favourite analogy is the electric bike. It certainly helps me reach certain goals—for instance, reaching the summit of Monte Generoso—but it still allows me to exercise my muscles. Reaching the peak of every objective by motorcycle is a different matter entirely." He observes that the fear with generative AI is that we may "end up atrophying the mind," fostering a culture of immediacy and instant gratification at the expense of long-term effort.

Another significant issue concerns personal data and the value of information. "The information we feed into the machine is data we surrender and lose control over," Rizzoli emphasises. However, the issue also affects those who produce high-quality content—such as publishers, newspapers, and authors—who are now confronted with systems that synthesise and rework the material of others without always citing original sources.

Alongside digitisation, Rizzoli argues that true literacy in AI use is essential, especially among the youth. "As a research institute, we are also concerned about the social and educational impact, particularly on students using these tools; we want the experience to be enriching rather than a process of impoverishment." To this end, IDSIA is collaborating with the Canton to identify the skills required for students most exposed to these shifts.

Looking ahead, Rizzoli notes that the current Large Language Model paradigm may be approaching maturity. The next evolution could be represented by so-called "World Models"—systems capable of better understanding the workings of the physical and social world. "Humans learn how things work with very little data and few examples; the goal of AI now is to replicate that human capacity. World Models represent an evolution towards a more human and responsible AI."