The European Parliament approved its draft proposal for Europe’s landmark artificial intelligence legislation, the AI Act, on 14 June. The legislation is currently in the final stage of negotiations, where different drafts from the European Parliament, Commission, and Council are being reconciled. The final text is likely to be agreed upon and in force by the end of this year, starting with a two-year implementation period.
The European Parliament’s draft, with all of its avoidable and unavoidable weaknesses, is a victory for human rights. It is the culmination of tireless work by human rights defenders, MEPs, and staffers who built bridges and convinced unlikely allies that certain uses of the technology, however “efficient” or “taxpayer-friendly” or “GDP-increasing” they may be, are unacceptable.
Key points of the Parliament’s draft
Liberties and other rights defenders especially welcome that the European Parliament’s draft version of the AI Act:
- bans real-time biometric identification systems utilized in publicly accessible areas, and most uses of post biometric identification systems excluding law enforcement for the investigation of severe offenses and solely upon obtaining judicial approval;
- bans biometric categorization systems that use sensitive characteristics (such as gender, race, ethnicity, etc.).
- bans predictive policing systems based on profiling, location data, or an individual's previous criminal history;
- bans emotion recognition systems in law enforcement, border control, workplaces, and educational institutions;
- bans unauthorized scraping of biometric data from social media or CCTV footage for the purpose of constructing facial recognition databases.
We are equally pleased that the European Parliament’s draft version is a definite improvement over the Commission’s draft when it comes to transparency and due diligence. That is, that:
- public authorities and those deploying AI systems on their behalf have to register the system in a publicly viewable database,
- and deployers of high-risk AI systems need to carry out fundamental rights impact assessments of their deployment before they put their systems to use, and public authorities will also need to publish them.
These are important achievements that have been long sought by human rights defenders. The safeguards put forth in the Parliament’s draft show that its members heard our concerns and understood their importance. Unfortunately, however, their draft is not without serious shortcomings:
- The most fundamental one is that it adopts a "risk-based approach". This means that restrictions will be put in place depending on how risky an AI application is thought to be by the legislators. All drafts of the AI Act, including the Parliament’s, do leave AI systems that are not banned or deemed to be high-risk basically unregulated. This means that our rights are not protected in ways they should be. Fundamental rights impact assessments, for example, should not only be prescribed for deployers of the ostensibly high-risk AI. Ideally, all deployers of all AI systems would need to carefully evaluate whether the system they want to deploy carries risks they should mitigate.
- The European Parliament also missed an opportunity to extend necessary protections to people from marginalized and at-risk groups. MEPs chose not to adopt a ban on AI profiling systems based on a person’s sensitive characteristics for the purposes of migration or asylum.
- The Parliament also voted to reject the requirement to provide explanations to people affected by decisions taken based on AI systems deemed not to be high-risk. For decisions made based on the latter, the Parliament’s draft does give the right to explanation.
- MEPs also did not support the right of people to be represented by public interest organizations in official complaints about the use of AI systems.
The trilogue begins
The future AI Act, despite its foreseeable weaknesses, also represents an important and highly commendable shift in regulatory approach. For years, the narrative has centered on human-centric AI and self-regulatory ethical guidelines for tech companies. By now, European legislators agree that self-regulation is not sufficient and have shifted the discourse from abstract AI ethics to more concrete human rights considerations.
This shift is crucial, but it does not imply that legislators can rest on their laurels. Now they need to deliver the goods. The trilogue negotiations, which began immediately after the Parliament’s draft was adopted, will test lawmakers’ commitment to human rights protections and their fortitude in the face of what has already been aggressive lobbying on behalf of businesses.
Once this is done, they need to make sure that the human rights protections in the AI Act are appropriately enforced in practice. One way of doing this is to revise accordingly the Coordinated Plan on AI - the European Union’s strategy regarding AI technology. First published in 2018 and revised in 2021, it was originally conceived when human-centric AI and self-regulation were still the dominant narrative. Its revision was not thorough enough and failed to include adequate human rights considerations and expectations for member states.
Reworking the Coordinated Plan
Instead, it focuses on increasing competitiveness and increasing deployment. At Liberties, one of our key plans related to AI involves rectifying this omission. We have launched a project to review national AI strategies, and preliminary research indicates that individual member states do not typically devote significant attention to human rights considerations in their AI policies, partially because their strategies were to some extent written as a response to the Coordinated Plan.
Our goal is to persuade the European Commission to incorporate a full scope of human rights considerations when revising the Coordinated Plan, aligning it with both the AI Act and the Charter of Fundamental Rights. The Commission, however, has thus far been unresponsive to our inquiries.
We remain undeterred as we wait for the trilogue to play out. We strongly hope that Parliament’s negotiating team will also remain similarly undeterred in the coming months and will not give up on the achievements the relevant Committees have secured to this point.