European regulation of artificial intelligence, in 3 keys

In the absence of details on various technical aspects, here we summarize the three keys you need to know about this ambitious project.

initiative known as AI Act Classifies AI systems according to the threats they pose to society. brussels Establishes four distinct stages. Thus, it will impose sanctions on those who pose an “unacceptable risk”. This is the case of cognitive-behavioral manipulation, classification of people based on their behavior or socio-economic status, or real-time biometric identification programs.

One step below are “higher risk” uses, which are permitted but subject to strict obligations. Thus, systems that may negatively impact security or fundamental rights – those implemented in sectors such as education, automobiles or border control – will be evaluated before they are marketed and also during their life cycle.

In third place are AI with “limited” risk of manipulation. This refers to virtual assistants or generative systems such as chatgpt, These must meet transparency criteria such as informing users that they are interacting with a machine or that the content it creates is synthetic.

Finally, there are those that have no or “minimal” risk. They are the AI ​​present in most of the digital services we use, from video games to spam filters. These can be used without requiring compliance with any particular standard.

2023 was the year of its emergence Generative AI, models that are fed by large amounts of Internet data are able to automatically generate all types of text, images or audio. Regulation of these applications was also one of the most discussed aspects among European legislators. The text agreed in December establishes that the companies responsible for these programmes, e.g. Microsoft one of two GoogleThey should be transparent and disclose what data they are trained with to avoid potential copyright violations.

France is the country that has pushed most to reduce these obligations, noting that they would harm European AI startups like French Mistral AI. Companies like OpenAI, the creator of ChatGPIT, have also done this. The final agreement will not include the cuts requested by Paris and Silicon Valley.

The last key is perhaps the most controversial and important: biometric surveillance. In June, the European Parliament banned the use of systems like facial recognitionConsidered by organizations in defense of confidentiality As a threat to fundamental rights. The political agreement sealed in December vetoes real-time identification, but adds several exceptions that specifically benefit Police And armies, Thus, security forces and bodies will be able to use this technology in public places and with prior judicial authorization for limited cases like terrorism, human trafficking or detection of rapists.