Thread by @ZennerBXL

🚨 DOWNLOAD! The 1st of two #AIA draft reports of @Europarl_EN is available from today. On 15 March at 11h00 - two months before the joint IMCO/LIBE piece - @AxelVossMdEP will present it officially in @EP_Legal. Read more about our plans to promote trustworthy #AI in #EU here 👇

Wait, there are two #AIA reports of the @Europarl_EN? Yes, kind of! @AxelVossMdEP's double role as EPP LIBE Shadows + JURI Rapporteur makes him the 3rd leading MEP, while Rule 57+ (check screenshot) makes JURI piece something much more than an opinion. We amend the whole #AIA.

Let's begin by stating that we find the @EU_Commission's proposal a good starting point. However, the regulation still requires a lot of work in order to cope with the complex & highly dynamic field of AI. The @EUCouncil did a great job in this regard & made a lot of progress!

What is still needed? We tabled in total 309 AMs in 7 key areas:
1️⃣ Clear scope & AI def.
2️⃣ 'Trustworthy AI' concept
3️⃣ High-risk à la White Paper
4️⃣ More data for AI training
5️⃣ Fair responsibilities in supply chain
6️⃣ Strong AI board + HLEG
7️⃣ Efficient consistency mechanism

1️⃣ Reasoned with the connectivity, autonomy, data dependency, complexity, openness and opacity of new #AI systems, the #AIA should consequently focus only on such new systems. This means, only Machine Learning should be in its scope as others are already covered by existing laws

2️⃣ European Standardisation Organizations are waiting for a political signal on "meta-level" that they can then translate into technical standards for the various AI applications. The excellent work of the High Level Expert Group on AI offers a balanced concept for such a signal.

3️⃣ The current ANNEX III puts whole sectors under general suspicion. AI White Paper indicated better solution: AI system is considered high-risk when its intended purpose falls in critical area (new ANNEX III) + in manner that significant harm is likely (new risk assessment).

4️⃣ Today's ML systems need lots of high quality data, which is why we proposed several changes to Art 10. To comply with #AIA, we also suggested to consider training/validating/testing of AI systems a 'legitimate interest' and 'compatible purpose' according to Art 6(1f/4) GDPR.

5️⃣ The supply chain of AI systems is highly complex. A fair distribution of #AIA responsibilities among the actors is thus very difficult. General purpose AI systems make it even harder. We took the great Art 23a from @EUCouncil and added a few things to make it more balanced.

6️⃣ The AI board must have a strong mandate + enough resources. It will be key to coordinate between Member States, review implementation and constantly adjust the #AIA to new technological developments. Reestablished HLEG guarantees that the AI community is involved adequately.

7️⃣ Each national competent authority should act as single point of contact for this Regulation. While we introduce one-stop-shop, we don't want to repeat GDPR shortcomings. Therefore: strengthening of consistency mechanism + allow AI board to make binding decisions if necessary.

Download the full JURI draft report via kaizenner.eu/post/juri-draf…. During his presentation in @EP_Legal committee on 15 March, @AxelVossMdEP will present - on top of our draft - FOUR additional points that are necessary to make #EU a global leader in #AI. Stay tuned! 👀

Last but not least, this is an early piece with our ideas on how to improve the #AIA. Further adjustments are definitely needed. So, what is (not) working and what is still missing? Our post-presentation-consultation starts today. Send your feedback to [e-mail]!