Regulation (EU) 2024/1689 of 13 June, which establishes harmonized rules on artificial intelligence, is probably the most ambitious and demanding EU act of recent times. Its impact on the European regulatory acquis can be gauged from the fact that it involves amending six other regulations and three relatively recent directives. The scale of the functional demands now being placed on the Union’s structure, in order to make the rules aimed at framing and controlling the phenomenon effective, is clearly due to the fact that a new organic unit – the European AI Bureau – has been created for this purpose within the European Commission’s Directorate-General for Communications Networks, Content and Technologies.
This new unit, which aims to be a kind of regulatory authority at European level, has been rapidly recruiting highly qualified staff specialized in artificial intelligence technologies. At the same time, there is a clear intention to encourage member states to set up similar regulatory structures themselves.
The text of the Regulation is daunting. It takes up no less than 195 pages of the Official Bulletin, a good part of which is taken up by endless recitals – useful, in any case, for understanding the European legislator’s intentions in this regard. The preceptive part consists of 113 articles plus 13 annexes. It is not difficult to foresee that it will take a long time just for the many people to whom this regulation is addressed to reach an effective level of understanding.
Perhaps the main virtue of the European Regulation is that it identifies areas of application for AI, ranking them based on the risks involved in each one.
From the outset, various types of use of AI systems are prohibited, namely: those involving the use of “subliminal techniques that bypass a person’s awareness, or techniques that are manifestly manipulative or deceptive, with the aim or effect of substantially distorting the behavior of a person or a group of persons, thereby significantly impairing their ability to make an informed decision”; those aimed at “assessing or classifying natural persons or groups of persons over a period of time on the basis of their known, inferred or foreseeable social behavior or personality or personal characteristics”, with possible disadvantageous consequences for those persons; those aimed at “assessing or predicting the risk of a natural person committing a criminal offence”, based solely on profiling them or assessing their personality traits and characteristics; and several others, described in detail.
Next, the areas of application of AI that involve “high risk” are characterized, i.e. “a significant risk of harm to the health, safety or fundamental rights of natural persons”, namely by significantly influencing the outcome of decision-making. Annex III of the Regulation specifies the areas in which these risks may occur: use of biometric data, where legal; critical infrastructures: management and control of critical digital infrastructures, road traffic or water, gas, heating or electricity supply networks; education and vocational training; employment, management of workers and access to self-employment; access to essential private services and essential public services and benefits; law enforcement; management of migration, asylum and border control; administration of justice and democratic processes.
This focus on critical areas and systems is, in itself, a significant step forward in building an adequate regulatory framework for the various phenomena generated by the use of AI.
However, it is natural that this enormous effort by the European institutions should be viewed with caution, not to say skepticism. The object of the new regulation is a largely immaterial phenomenon, which can take on a multitude of forms and uses. Setting up a cumbersome system of regulation and control, with the main targets being the promoters and designers of AI devices, is the typical response of a highly bureaucratized organization like the European Union. The aim is excellent – “to promote the adoption of human-centered and trustworthy artificial intelligence (AI)”, says Article 1 of the Regulation – but it remains to be seen whether the path chosen follows the most appropriate route to achieve it.
António Monteiro Fernandes @ Of Counsel, DCM | Littler
