Are you paying enough attention to the EU AI-ACT? | Part 1
While most organisations are only just coming to terms with the EU’s General Data Protection Regulation (GDPR), the next big thing is already looming on the horizon: The EU AI-Act, or AIA. As we have now entered the trilogue phase (more on that later), any organisation using AI (or planning to use AI) should be paying attention. In this two-part blog we will summarise some of the key information you should know about the AIA. In this first part we will go over the AIA timeline, and where we stand now. In the second part we will discuss key elements of the AIA to take into consideration when developing and using AI. You can read the blogs separately, so, if you’re not too interested in the politics of it all you can skip right to part two where we get into the implications for your organisation.
Part 1: A brief timeline of the EU AI-Act (AIA)
For who is this blog: Anyone interested in the politics of the EU AI-Act and where we stand now
Prerequisites: none
The massive uptake of AI solutions in both private and public organisations has led to many opportunities, but also raises important concerns. Especially given how simple it is becoming for organisations to integrate “off-the-shelf” AI-solutions without even having to hire a data scientist. The first set of concerns were related to privacy issues, as a reaction to the mass collection of personal data. This resulted in stronger data protection legislation in the EU, in the form of the GDPR. More recent concerns are related to algorithmic transparency, bias, and accountability. This established the need for AI regulation. To be sure, the legislation was also spurred by market parties who wanted a clear European framework to innovate responsibly. The AI Act is a proposed European law on artificial intelligence (AI) — the first (serious) law regulating AI by a major regulator anywhere. In the rest of this section we will take you through the timeline of the EU’s AIA, highlighting key moments and explaining where we are now. Spoiler: The legislation is almost adopted!
Figure 1: Timeline of the AIA
The White Paper and the draft AIA
In 2018 the European Commission established the High-Level Expert Group on AI. Their task was to provide recommendations and guidance on the development and deployment of AI in the European Union. The group is composed of 52 members, including experts from academia, industry, and civil society. Following recommendations from the High-Level Expert Group, the European Commission (EC) published a White Paper on AI in February 2020, titled “A European approach to excellence and trust”. This White Paper already set the stage for the European approach to AI, and reveals the overall purpose and spirit of the (draft) legislation: encourage AI innovation in a safe way.
On publication, the EC invited Member States, other European institutions, and all stakeholders, including industry, social partners, civil society organisations, researchers, and the public in general, to react to the White Paper. After this round of stakeholder consultation the EC published a draft EU AIA in April 2021. As with any EU draft legislation, this kicked off the routine procedures of parallel discussions and amendments by the Council of Ministers and the European Parliament. Moreover, other key stakeholders like the European Central Bank, are also invited to provide their (non-binding) opinions.
Let the discussions begin: The Council of Ministers and the European Parliament
The Council of Ministers acted first. They represent all the EU member states. Their job is to study the draft EU AIA and propose necessary changes. Slovenia provided a first draft of amendments in a ‘compromise proposal’ in November 2021, followed by another set of amendments by the French and Czech Presidency in 2022. In December 2022 the Council unanimously agreed on the final version of the compromise text, also known as the ‘general approach’.
Meanwhile, the European Parliament (EP) is also tasked with discussing the draft EU AIA. The EP is made up of 705 directly elected members of parliament (MEPs). MEPs sit in political groups. They are not organised by nationality, but by political affiliation. So, they represent different interests than the Council of Ministers. In Parliament, the discussions are led by the Committee on Internal Market and Consumer Protection (IMCO) and the Committee on Civil Liberties, Justice and Home Affairs (LIBE). The EP unveiled a draft report in April 2022 and subsequently received over 3000 amendments. The EP took these into consideration and reached an agreement on a final text on April 27 2023. The vote is scheduled for June. A date that has been postponed several times because they had not yet considered (the potential high-risk of) large language models like ChatGPT. Is it slightly alarming that that was not on their radar yet? Yes. Politico actually wrote a great blog about this.
Entering the Trilogue Phase: AI legislation is imminent!
This means the “trilogue phase” is imminent; the phase of final discussions between the Council, Parliament, and the Commission. Once a compromise is reached, the final text of the proposed legislation is submitted to the European Parliament and the Council of the European Union for approval. If they both approve the final text, the proposed legislation is adopted and becomes law in the European Union. The EU AI-Act has been long in the making but we now have reached the final stages. It is probable (if not a sure thing) that the AIA will be adopted this year! This means all organisations in the EU using AI will need to be AIA-compliant by 2025. So, organisations need to start paying attention NOW to what exactly will be required under the AIA. For concrete advice, see Part 2 of the blog.
