Italy is making giant strides in developing models of Artificial intelligence (IA), with the birth of tools such as Minerva And Llamantino 3. During the first general conference of Fair – Future AI Research in Naples, these new Italian linguistic models, fundamental for the progress of research in our country, were presented. Roberto Naviglihead of the Minerva project, explained why it is crucial to develop Llm (Large Language Models) in Italian: “Dialogue with an AI only in English leads to cultural and linguistic biases, which is why it is essential to create native language models in our language.”
The importance of Italian AI models
The project Minerva it is one of the first open linguistic models developed entirely in Italian, with the aim of guaranteeing greater control over the data used. Navigli illustrated the advantages of creating AI in Italy: “Models adapted to our language have excellent performance, but developing native models allows you to verify the contents used, guaranteeing greater security”. In parallel, Llamanteno 3 is an example of an existing model that has been “re-trained” to operate in Italian.
The new challenges of human-AI co-evolution
The conference also addressed issues related to coexistence between man and AIwith Dino Pedreschi of the University of Pisa which raised the alarm on the risks for democracies and the possible collapse of online content. “We increasingly rely on AI for daily decisions,” explained Pedreschi, “but we must keep in mind that our choices influence the data that feeds the AI, creating a cycle of co-evolution that can lead to unpredictable deviations.” A concrete example is the risk of creating “digital garbage” when AI generates content that is reused by other models, leading to a loss of data diversity.
Physics seeks to reveal the hidden laws of AI
Another theme explored at the conference was the link between physics and IA. Federica Gerace from the University of Bologna underlined how AI are “complex systems” that we do not yet fully understand, similar to the first machines of the industrial revolution before the discovery of the laws of thermodynamics. “We want to find the hidden laws that govern AI,” said Gerace, who together with other researchers is exploring how AI works. transformerthe “building blocks” behind models like ChatGpt.
Diversity as a resource in the digital world
Valentina Pansanella of the National Research Council highlighted a growing problem: the lack of diversity in digital data. “The risk is that the platforms digital channels, such as those for online purchases, always offer the same products, undermining digital diversity”. Pansanella proposed the idea of developing public AI models, controlled by the State, to avoid the concentration of power in the hands of a few companies and guarantee a controlled and positive co-evolution between humans and AI.
If you want to stay updated on the latest news in terms of Artificial intelligence and its implications for society and science, continue to follow us!
#Minerva #Llamantino #Italian #innovates #future