Many principles, limited resources and some more concrete rules but destined to spark discussion, such as the provisions to protect copyright holders or those that update the Criminal Code for the era of ChatGpt. This is the impression you get when reading a first, still preliminary draft of the Bill on Artificial Intelligence of the Meloni government, which should be approved in the next few days – after some delays – in the Council of Ministers. And ideally to stimulate and orient in an “anthropocentric” way the development and application in Italy of the technology that promises to change everything.
Artificial intelligence, the Privacy Guarantor repudiates the Meloni government: “Whoever controls must be independent”
by Filippo Santelli
Palazzo Chigi guides and controls
One thing that is clear from reading the text is the will of the government – and in particular of the Undersecretary for Digital Alessio Butti – to centralize the governance of artificial intelligence at Palazzo Chigi. It will be up to the Innovation Department led by Butti himself, again in the next few days, to publish one National strategy for AI, on the basis of the indications provided by the committee of experts, with concrete objectives which will then be updated every two years. The department itself will monitor its implementation. And the competent authorities to supervise the matter will be two government-appointed agencies, viz the Agency for Digital Italy (Agid) and the Agency for Cybersecurity (Acn): an early decision, permitted by the European AI Act with which the Italian law must integrate, but contested by those who hope for third-party and independent supervision, entrusted to an authority such as for example the Privacy Guarantor. Palazzo Chigi will also be responsible for supervising a new Foundation for research, experimentation and adoption of AI systems, which should be a sort of operational arm of the strategy.
Ia and copyright
A very hot topic, even abroad, is how to protect content protected by copyright used by large AI companies to train their large models. The Bill, in the draft circulated in these hours, updates the 1941 law on the subject with a dedicated chapter, essentially applying the concept of “opt out” already contained in the European standard: the producers of the contents will be able to “reserve” their use, that is, prohibit them from being extracted by technology companies to train their systems, even if it is not specified how. However, if companies do so they will be liable to a fine, and in each jar they will have to explain the sources and authors of the materials they use. The effectiveness of these provisions will have to be evaluated in practice, but one of the objectives is to push Big Tech to enter into agreements with publishers and other content producers that provide adequate compensation.
“Fair compensation to publishers from artificial intelligence giants”
by Aldo Fontanarosa
A stamp on fake content, algorithms in the penal code
From reading the draft, the bill seems more unbalanced on the objective of risk prevention, than on the development of AI. With a regulatory approach that the community of innovators should not like. Article 23 talks about what could vulgarly be defined as “sticker”, i.e. an identifying element or sign that all television and radio service providers should apply to content modified or altered – even partially – with AI. The objective is obviously to avoid voluntary or involuntary manipulation of users, even if it does not concern social media, where a large amount of false content spreads.
By the way, the last part of the bill provides for a new type of crime, 612-quater, immediately after the one on “revenge porn”, to punish anyone who harms someone through the dissemination of “content generated or artificially manipulated.” That’s not the only part of the Penal Code in which AI should be inserted, a sign of the times: the Bill also makes it a common aggravating circumstance for a series of specific crimes such as fraud, money laundering and market manipulation. A questionable prediction, considering the still very low level of maturity of the technology. Children under 14 will only be able to access AI applications with parental consent.
Butti: “More researchers and national champions, with our plan AI will speak Italian”
by Filippo Santelli
Data sovereignty
The text provides principles for the use of AI systems within the healthcare system or to support the activities of the courts. But the draft also contains a provision – marked as “proposed” and therefore yet to be validated – to direct public administrations to favor artificial intelligence services and suppliers that locate and process critical data in “data center places on the national territory”. Behind these hypotheses we can sense the desire to push as many companies to equip themselves with infrastructures in Italy, to attract investments and to “increase the technological sovereignty of the Nation”. But despite referring to critical data, a rule of this type appears incompatible with the single market and with the fact that privacy rules are European.
Actual resources: 150 million
In recent days there has been talk several times, including by Prime Minister Meloni, of a billion that the government would have allocated to AI. In the draft of the bill, the resources actually allocated stop at 150 million euros (recovered from other budget items), with which shares of investment funds in startups dealing with AI created by Cdp Venture Capital. The government’s expectation is that with the participation of private investors it could reach 600-800 million euros, although yesterday when presenting its industrial plan CDP Venture spoke of one billion for AI. The billion to which Meloni referred would instead be a different amount, dedicated not to startups but generically to artificial intelligence projects. It is not clear at the moment if and where it will be allocated.
#law #coming #content #producers #deny #Big #Tech #algorithms #enter #Criminal #Code
– 2024-04-09 22:37:31
