|
MEP Dragos Tudorache (Renew, Romania) explained that the law needs to be as European as possible, which implies three things:
-human rights and freedoms
-resilience and health of our democracies
-success, competitiveness, and innovation capacities
He further underlined that the definition should remain as broad as possible. The Regulation follows a risk-based approach. There are prohibited AI systems, high-risk systems, and limited risk AI systems (transparency requirements only). All the other AI systems are not subject to the requirements.
The two leading MEPs decided to include predictive policing practices in the list of prohibited practices. They justify this amendment by the fact that predictive policing is discriminatory and does not work.
In addition, MEP Dragos Tudorache and MEP Brando Benifei modified the list of high-risk AI systems (Annex III). They added:
-AI systems intended to be used by political parties, for instance, to influence voters or count ballot votes
-AI systems intended to be used for children (e.g. personalised education) and that can shape the development of children
-AI systems intended to be used for making decisions on the eligibility of natural persons for health and life insurance
-deep fakes relating to existing persons and AI authors
The two rapporteurs did not agree with the product safety approach and amended it.
Manufacturers need to register the AI high-risk systems in a special database. The two MEPs want to avoid having an AI-powered Pegasus system. Therefore, they strengthened transparency requirements on the intended use.
The two rapporteurs considerably reinforced the role of the AI Board (e.g. for enforcement) and removed most of the part on Member States’ enforcement.
MEP Brando Benifei stressed that EU policymakers have a double responsibility to shape the rules for Europeans and give an example to the world. He explained that he and MEP Dragos Tudorache could also agree on strengthening the alignment with GDPR. The way of processing data is and shall remain GDPR.
The two rapporteurs could also agree on deleting the possibility of derogating from conformity assessment for the purpose of security or national interest. Moreover, the two MEPs strengthened the obligation of human oversight for high-risk AI systems.
They propose a rebalancing of the obligations falling on the different actors of the value chain. Sometimes providers do not even have access to the datasets, users are those having access to the datasets. Therefore, the draft report also places some obligations on users.
The Commission is given the role of a European enforcer and market surveillance authority.
MEP Brando Benifei plans on introducing further amendments to the draft report. He wants to mandate a fundamental rights impact assessment for high-risk AI systems. He also wants to do more on AI used in the workplace and emotional recognition.
Additionally, he would like third party assessment to be made mandatory for high-risk AI systems. In his view, such an amendment would reduce the burden on SMEs, which do not have the resources to conduct self-assessments.
MEP Axel Voss (EPP, Germany), shadow rapporteur in LIBE, highlighted that AI is of strategic relevance, and that is why policymakers need to strike the right balance not to hinder AI. For the scope, the AI Act should focus on autonomous AI. MEP Axel Voss sees the need for clarifications when it comes to high-risk issues. The shadow rapporteur emphasised the importance of standards, in particular, to assess the quality of data. The AI Act must be doable, and the only way to do this is with technical standards.
MEP Deidre Clune (EPP, Ireland), shadow rapporteur in IMCO, supports the issue of sandboxes to boost innovation, in particular for health. Regarding the definition of AI, she believes that MEPs need to keep receiving feedback. The definition should not be too broad.
MEP Petar Vinatov (S&D, Bulgaria), shadow rapporteur in LIBE, explained that the S&D group would like to strengthen further some provisions for a complete ban of remote identification solutions in public spaces with no exceptions both for real-time and ex-post. They also would like to introduce mandatory human rights assessments for high-risk AI.
MEP Svenja Hahn (Renew, Germany), shadow rapporteur in IMCO, stressed that Europe needs to have an ecosystem that favours innovation. Policymakers must support SMEs and start-ups by lowering the regulatory burden. The Renew group wants to clarify provisions to make it easier for businesses to comply with the Regulation.
MEP Kim van Sparrentak (Greens, the Netherlands), shadow rapporteur in IMCO, also spoke on behalf of MEP Sergey Lagodinsky (Greens, Germany), shadow rapporteur in LIBE. The green shadow rapporteurs consider that, for high-risk, the AI Act must consider not only risks for individuals but also risks for society and the environment. There must be a mandatory fundamental rights impact assessment. In addition, third-party conformity assessment should be mandatory. Self-assessment and presumption of conformity are not enough to assess these important criteria.
The two green rapporteurs also want to introduce explanation by default when decisions affect people and have been made with AI. Collective action should be an explicit possibility as well.
Regarding prohibited AI systems, they support a ban on facial recognition in public spaces and want to stop emotional recognition and surveillance of people at work or school.
MEP Kim van Sparrentak does not want to repeat the GDPR situation for the governance part. Ireland cannot supervise the big tech companies again. Therefore, the European Parliament should strengthen the role of the AI Board in the enforcement of the AI Act.
Kilian Gross, Head of Unit at the European Commission, reacted to these comments. He stressed the need to be very precise for the list of high-risk AI systems. Operators should easily know if their system falls within the list or not and hence whether the system needs to undergo a conformity assessment.
He explained that the AI Board cannot take binding decisions in the current shape of the legislation. If policymakers want to give this power to the Board then they need to allocate sufficient resources for enforcement as well. At the moment, no such resources are foreseen.
He further clarified that the European Commission did not include the obligation to perform a fundamental rights impact assessment because the GDPR impact assessment already includes a fundament rights impact assessment. Obligations should not be duplicated.
Next steps
MEPs from the IMCO and LIBE Committees can amend the draft report until 1 June 2022.
If you have any questions on this topic, please contact Camille Dornier - Policy Manager: camille.dornier@eurosmart.com
|
