Italy’s new AI rules. On Wednesday, the Council of Ministers approved a set of new rules to regulate the misuse of artificial intelligence. Subject to parliamentary approval, they are designed to make Italians’ digital lives ready for the impacts of AI – from the danger of deep fakes to fostering development in the sector.
- The new rules will apply to five key areas: the national strategy, national authorities, promotional actions, protection of copyright, and criminal sanctions, explains a note from the Prime Minister’s office.
- They also give the government power to align the national system with EU regulations in matters such as citizens’ AI literacy, professional training, and amending criminal law to adapt offenses and penalties to the illicit use of AI systems.
Rome takes the lead. As Innovation Undersecretary Alessio Butti explained before the press, the Italian government is the first one to legislate on AI. The text is compliant with the EU’s AI Act and was also much awaited by G-7 colleagues at the last ministerial meeting, he stressed. “It is a bill that unambiguously defines who draws up the strategy [and] also defines who monitors, supervises, notifies, and sanctions” all things AI-related, he said.
- The bill sets up two National Authorities for AI at the Digital Italy Agency (AgID) and the National Cybersecurity Agency (ACN), tasked with ensuring the application and implementation of national and EU legislation on AI.
Boosting the AI economy… The new rules aim to attract and retain AI talent by extending tax relief schemes for Italians returning from abroad. They also set aside a starting sum of €1 billion that State lender CDP – through its venture capital arm – will use to support small Italian businesses, as Enterprise Minister Adolfo Urso highlighted.
- The country’s 4 million SMEs “must be put in a position to take full advantage of this technology.”
- The goal, he continued, is “to create an AI market that is open, fair, and without barriers to entry [and] guarantee everyone access to high-quality data.”
- Facilitating the birth of startups and the consolidation of those already operating in the IA, as well as the creation of a national champion, are also core objectives.
… and curbing AI misuse. With the EU elections on the horizon, the Italian government is looking to counter the diffusion of AI-generated damaging material. “Whoever disseminates without consent altered videos or images with AI, causing unjust damage, is punished with imprisonment from 1 to 5 years,” summed up Justice Minister Carlo Nordio during the press conference.
- When the use of AI is carried out “in an insidious manner,” it “constitutes a specific aggravating circumstance for a whole series of offenses, such as substitution of person, fraudulent rising and falling of prices, fraud, computer fraud, money laundering, market rigging,” he explained.
An Italian-born AI? In the backdrop, Sapienza University’s Natural Language Processing research group is working within the Future Artificial Intelligence Research Foundation – a project led by the National Research Council – and leveraging a Leonardo supercomputer to build large Italian language models.
- These new LLMs are dubbed Minerva, and as group leader Roberto Navigli told Il Sole 24 Ore, they have been “built and trained from scratch using open access texts, as opposed to existing Italian models that are based on the adaptation of models such as LLaMA and Mistral, whose training data are still unknown.”
- Each Minerva model has been trained on a vast set of online and documented Italian and English sources totaling over 500 billion words. They are now available to FAIR’s scientific community and will soon be made accessible to the wider public.
