What’s in the AI ​​Act, the world’s first law to regulate artificial intelligence?

The European Parliament and the Council have reached an agreement to approve the AI ​​Act, the world’s first law regulating the development and use of artificial intelligence systems. The aim of the law is to indicate permitted and prohibited uses to protect the privacy and other rights of European citizens.

After reaching a political agreement, the text will be refined by the technicians called upon to draft the final version of the law, which will then have to be approved by the Parliament and the European Council.

The proposal was presented by the European Commission in 2021, but it took time, in particular due to the notable technological development carried out by several companies over the last two years: the innovations forced the European Union to take into account d possible new problems in a sector in full expansion and with still unclear contours. The negotiations were also long: European parliamentarians negotiated for three days to reach an agreement described as historic by the Internal Market Commissioner, Thierry Breton.

The Commission’s initiative covers various areas and applications of AI, from systems for hiring new staff in companies to algorithms that operate self-driving cars, including facial recognition by security forces. order and the spread of disinformation online.

One of the central themes, already widely discussed before the official presentation of the proposal, is linked to systems for automatic recognition of individuals in security camera images. During the negotiations, extensive discussions took place about how law enforcement should manage these systems while respecting people’s privacy to avoid the risk of mass deposits. Several countries favored very restrictive measures, while Italy, Hungary and France favored a more permissive stance. Facial recognition has been banned, except in three cases: the obvious threat of terrorist attack, the search for victims, investigations into serious crimes such as murders, kidnappings, sexual violence.

The regulation provides for other prohibitions in the use of AI, for example it will not be authorized to use technologies to calculate the “social score” of each individual, a practice increasingly experienced in China where each Citizen is assigned points based on his behavior, which gives him the opportunity to access particular services that are excluded from people with low scores.

Biometric recognition systems using sensitive data, such as political ideas, religion and sexual orientation, have been banned. Images obtained from the Internet cannot be used to create facial recognition databases. Emotion recognition in the workplace and schools has been banned. It will also be prohibited to develop algorithms that can cause physical or psychological harm to individuals, or have the capacity to manipulate their behavior, even in a subliminal form.

Another important point of the law concerns the technological systems on which chatbot services such as ChatGPT rely. Two levels of rules have been provided to distinguish high-impact AI from all other artificial intelligence systems.

High-impact AI refers to systems with considerable computing power: according to initial information, at the moment only OpenAi’s GPT-4 falls into this category. High-impact AI systems must comply with rules for transparency of AI training processes and sharing of technical documentation before being released to the market. All other systems only have to follow these rules from the moment the services are marketed.

Initially this distinction was based on the turnover of companies, but it was ultimately chosen to favor computing power, therefore the impact on the population and possible development.

Source: ilpost


Leave a Reply

Your email address will not be published. Required fields are marked *