The two European Parliament co-rapporteurs finalized the preliminary report on artificial intelligence (AI) on Monday (11 April), covering the points on which they found common ground. The most controversial issues have been postponed.
Liberal Dragoș Tudorache and Social Democrat Brando Benifei led the discussion on the AI law for the European Parliament’s Civil Rights and Consumer Protection committees, respectively.
“There are things that we have already agreed on, and they will be in the draft report, and things that we think we will agree on, but because we have not found the common denominator at the moment, we have not included them in the report”said Mr. Tudorache.
“Our approach has been to ensure that this settlement is truly human-centric”Benifei told EURACTIV. “We didn’t agree on everything, but we took an important step forward. »
A broad definition
The two lawmakers agreed to keep the definition of AI broad, excluding exceptions for general-purpose AI, i.e. AI systems that can be trained to perform various tasks. The definition remained mostly unchanged, but the specification that AI should pursue human-defined goals was removed.
Prohibited practices and high-risk applications
The text added preventive surveillance to prohibited practices, “an important measure to combat highly discriminatory practicessaid Mr. Benifei.
For public authorities, the preliminary report foresees more substantial obligations and transparency requirements when using high-risk applications, including registration in an EU-wide database.
The list of high-risk applications has been expanded to cover AI systems designed to interact with children, medical triage, insurance, deep fakesand algorithms with a potential impact on democratic processes, for example those used for election campaigns or for counting electronic votes.
Responsibilities and Transparency
“The text brings greater clarity on the distribution of responsibilities between providers and users. When users make changes to the algorithm put on the market by a provider, this entails certain obligations for the user”said Mr. Tudorache.
However, Mr. Benifei stressed that the preliminary report did not go far enough and foresees that more will have to be done during the review phase.
Governance and enforcement
The co-rapporteur agreed on a two-tiered approach. In addition to the national level, lawmakers want to add a European level for the application of AI which could have far-reaching consumer implications and widespread societal impact.
“The logic is the same as in the DSA”explained Mr. Tudorache. “Some parts of the enforcement remain with national authorities, but in some cases, triggered by certain criteria, the responsibility of the Commission comes into play.”
While the main MEPs both want to strengthen the role of the European Committee on Artificial Intelligence, which will bring together all the competent national authorities as well as the Commission, they still have to define how this will be done.
Progressive lawmakers have pushed for a new EU agency to take over enforcement of the regulation. However, Mr Benifei admitted there was no clear majority for it in parliament.
The preliminary report includes an amendment to include AI in the Representative Actions Directive, which would give consumer groups the ability to start legal proceedings.
Points of difference
The two lawmakers disagree on conformity assessment, a procedure that will lead to new AI systems being put on the market.
The initial proposal relies heavily on companies carrying out their self-assessments, but Mr Benifei believes this could be too risky from a consumer protection and fundamental rights perspective. On the other hand, Mr Tudorache wishes to avoid any excessive administrative burden for businesses.
An even thornier issue concerns biometric recognition technologies, which the Commission has proposed restricting to a few limited cases, an approach championed by Mr. Tudorache and Conservative lawmakers. Conversely, Mr Benifei and centre-left MPs want a total ban, insisting that even limited exceptions could allow for abuse.
Extending the ban on social rating practices to private entities was also on the table, but no agreement on the exact wording has yet been reached. Similarly, biometric categorization will also be addressed during the review process.
Another point that will not appear in the preliminary report concerns regulatory sandboxes (“regulatory sandboxes”), a device for experimenting with new AI applications under the supervision of a regulator. Tory MPs want to expand regulatory sandboxes, and Mr Tudorache is determined to make this a reality not just nationally but also regionally to be able to better target SMEs.
The two lawmakers also discussed including a fundamental rights impact assessment, but Mr Benifei would like to apply it to all users, while Mr Tudorache is limited to public entities.
The preliminary report was presented to the two parliamentary committees on 11 May. Committees are expected to vote on the final text on October 26 or 27, confirmed by a plenary vote on November 9.
Practical difficulties could arise in finding common ground with so many people involved, especially during the summer, not to mention potential disagreements on highly political points. Nevertheless, the two MPs said they were optimistic that it would be passed by Parliament by the end of the year.
We would love to give thanks to the author of this article for this remarkable web content
MEPs finalize preliminary report on AI
You can view our social media pages here and other related pages herehttps://www.ai-magazine.com/related-pages/