Benifei: “After AI Act, a directive on artificial intelligence in the workplace for stringent obligations”

From the correspondent in Strasbourg – The EU Artificial Intelligence Act is almost ready to enter into force after the final green light from the European Parliament this week (now only the Council’s ok is missing), but work is certainly not finished. Not only in terms of steps for implementing the new Regulation, with deadlines for introducing obligations on the use and development of new technologies, but also the next steps of legislative work. “We could make further choices with a directive that looks at the goals to be achieved on the uses of artificial intelligence in the workplace. There are several elements that we could include,” the co-rapporteur for the European Parliament on the EU Artificial Intelligence Act, Brando Benifei (PD), says in an interview with Eunews,  highlighting the possible directions of the Union’s future action and the main challenges that the Regulation is about to face.

With the agreement between Microsoft and the French start-up Mistral, we have seen the strong pressures that can be exerted regarding high-impact foundation models. Is the new Regulation up to the task of meeting these challenges?.

 “The agreement found is level-headed on this front because it provides a high level of protection concerning what we had set out as a problem. That is, not just protecting people but also the companies that will use the foundation models to develop their systems and need certainty about the safety and operation of these more powerful models. In the face of very strong lobbying pressures for the sake of public interest, it will be necessary to better understand what moved, and also why, while there was an agreement in place, Mistral was asking us to relax the rules to help new European developers.

 It is a question that makes us think that perhaps there were other interests behind this kind of pressure, yes legitimate, but still very strong. It had no effect because the final text is crystal clear in envisaging a path by the EU Office for Artificial Intelligence within a clear outline of principles for the safety of the most powerful models, including the issue of environmental impact, which has seen much opposition. This will go through an oversight of the European Parliament, which in the new legislature will have to ensure the proper implementation of the Regulation and that the Commission takes the necessary executive steps.”

 Could this work also come through new, more specific EU directives on artificial intelligence?

“Yes, more than for foundational models, I would say  generally for the use of artificial intelligence in the workplace, particularly for what cannot be governed by a Regulation since the EU does not have full competence over labor issues. There are several objectives that a directive could consider, such as more stringent obligations to involve social partners in bargaining and choices involving elements of safety and mandatory training.

In the Regulations, however, there are already important rules for workers. Not only the application of the principle of classification of artificial intelligence in the context of workplaces as high-risk or the ban on the use of emotion recognition systems, but also the fact that all workers will have to know that an artificial intelligence system is in use in their workplace, with information and involvement of  social partners. These are all things that were not in the original text and that we have built through parliamentary work.”

What kind of cost impact will the Regulation have on tech companies?”

“We have created special sandboxes to reduce entry costs for new market players and those wanting to develop their own business model. They will be able to enter the market without immediately having to comply with all the rules of the Regulation, but only when they are fully ready in an accompanied manner and with reduced compliance costs especially in the start-up phase. We need to give new companies the opportunity to compete at the same level as the big players already in the market. There is no doubt that the costs for new businesses and start-ups will be contained thanks to these actions, but we also need a policy of economic support. In addition to regulations, the European Union and member countries will also have to build support models, for example, an economic incentive system for early compliance with standards before they become fully mandatory. Increasing the desirability of early and voluntary enforcement is a current issue. It is something that should start now.”

About the Regulation, how worrisome is a possible use by member states of the post-remote and real-time exceptions for the extensive use of biometric recognition systems, especially when it comes to terrorist threats?

“We had to be open to find an agreement in the trilogue to some uses. Although the final text sees a much stronger restriction than the Commission’s original text, it is, therefore, much closer to that of the European Parliament and very far from the wishes of the Council. Objectively, it was a victory for Parliament, but it is clear that it is not the position we had voted for a total ban. In the end, the exceptions provided are limited and can be activated only by a decision of a magistrate and under the supervision of the DPA, with immediate notification.

In the case of the ‘specific and current’ terrorist threat, the context is very well defined. The terms used in the Regulations refer to a narrow definition of ‘terrorist threat’ based on the case law of the Court of Justice of the European Union. Therefore, abuses, which are always possible, are clearly such and can be stopped by the judiciary. This concept can never be used preemptively; it cannot be used to look for potential terrorists. The terrorist act must have already happened or, in a very narrow case study, there must be certainty and imminence of the attack.”

 The Italian government was pushing for the use of biometric recognition programs in public places. Are you concerned that it may still be want to pursue this restrictive approach?

“It could try, but it couldn’t do it because six months after the Regulations come into effect, the use of real-time biometric cameras in public spaces – other than for the very limited cases described – and even predictive justice will be banned. These are all things on which Minister Piantedosi had initiated experiments, but they will have to stop. It is no coincidence that Italy was one of the governments that resisted the most on the EU Artificial Intelligence Act.”