It seems that the major companies that produce AI software have synchronized their clocks, because in the days around ChatGPT’s first birthday (last November 30th) a bit of everything happened.
Let’s think not only of the twists on the OpenAI board, or the rebranding of Bing Chat into Copilot, but also the presentation of renewed versions of already existing software, or that of brand new artificial intelligences, such as Amazon’s Titan or, in recent hours, Google Gemini.
Towards regulation of AI
Ma These are also days of great excitement for the regulation of artificial intelligence, especially generative intelligence.
First it was the turn of a document, signed by 18 countries including Italy, containing some (non-binding) rules on the use of AI. The document identifies four key areas to pay close attention to: secure design, secure development, secure implementation, and secure operation and maintenance.
After which, in anticipation of the discussion of the AI Act underway in Brussels, 34 Italian cultural associations have signed an appeal asking our government to “support balanced regulation which, by guaranteeing the transparency of sources, favors development of artificial intelligence technologies, while protecting and promoting original human creativity and all the cultural contents of our country.”
The discussion on the AI Act
The appeal, we were saying, was signed on the eve of discussion on the AI Act that is currently taking place in Brussels.
In fact, it will be decided whether or not to approve the important European legal regulation on artificial intelligence. As we read on the official website, the AI Act is “a European law proposal on artificial intelligence, the world’s first AI law from a major regulatory body. The law assigns AI applications to three risk categories.”
Last June, Sam Altman’s attempts (also apparently successful) to have the bill revised, downgrading the level of risk of generative AI, and therefore ensuring the loosening of the clampdown on the law caused a sensation. EU on OpenAI products.
The trilogue on the AI Act begins with black smoke
We were saying that an important meeting is currently taking place in Brussels to decide on the AI Act.
It is a so-called trilogue, i.e. a joint meeting of representatives of the Council, Parliament and Commission, to reach an agreement more quickly than the ordinary legislative procedure.
However, from rumors it seems that, after 22 hours of negotiations, only partial agreements were reached between legislators and governments of the member states. The discussion, which began on Wednesday 6 December at around 2pm, should have brought everyone to an agreement regarding “systemic risk” AI models. Whoever develops them will have to publish a clear list of the contents used for their training. This is the position expressed in the appeal of Italian cultural associations, which contested the idea (also shared by the governments of France and Germany) according to which software producers could rely on self-regulation.
On other points, however, we are still far away.
The node of biometric recognition
The important trilogue on the AI Act will be adjourned again at 9am on Friday 8 December.
The point of greatest disagreement seems to be that of biometric identification techniques. And, together, that of predictive policingi.e. the use of artificial intelligence systems by law enforcement agencies to predict the time and place of possible crimes, and the profile of those who might commit them.
In general, the position of the EU Parliament tends to be more severe, while that of the Council and some countries pushes for less stringent rules.
L’iter dell’AI Act
A European regulation on AI was first proposed by the European Commission in April 2021.
In June there was the important green light from the European Parliament, which led to the trilogue currently in full swing.
We were talking about radical position of Parliament. Which, once given the ok, he immediately made it known that “AI systems that present an unacceptable level of risk for people’s safety will be prohibited”.
Which means a halt to remote biometric identification in public spaces and predictive policing. And a stop to biometric categorizations based on sensitive characteristics such as gender, race, ethnicity, citizenship, religion and political orientation.
In the next few hours we will know if the AI Act will pass. In the event of an agreement, the law will be arrived at in the next few months, which will be made operational within two years.