Understanding the AI Act

Understanding the AI Act

  • Post author:
  • Post category:AI
 Many of us had been eagerly waiting for the Artificial Intelligence Act (AI Act) proposed by the European Commission in April 2021 for nearly a year. After the High-Level Expert Group on AI published its Ethics Guidelines for Trustworthy AI in 2020, this legislative proposal was expected to be the first legal framework for AI technologies, aiming to enshrine human rights in the development and deployment of AI systems. Together with the dedicated policy experts Alexandra Ciarnau and Rania Wazir, we decided to pay tribute to the special occasion by designing a host of creative content – and of course, working on our feedback to the consultation. 

The proposed AI Act in short (and somewhat longer) 

With such a groundbreaking framework on the table, and considering the intricacies of the proposed text, we soon decided to find a way to make the AI Act more accessible, both for our members and for everyone who is not a lawyer.   The AI Act is a lengthy and complex piece of legislation. It creates a number of roles for different types of companies as well as for the institutions tasked with overseeing certain aspects of its enforcement. Its provisions regarding processes put it very close to product safety legislation, but the human rights aspects it considers set it apart from other legislation in this category. How, then, can this proposal be rendered more comprehensible? We answered this question for ourselves by designing a series of posters to help others navigate their way through the text. Of course, co-operation is key: a huge thank-you to our Content Lead, Verena Stanzl, who spent many hours improving the design of the posters!   While the posters are both a helpful first starting point and a fun project, we also decided to summarise some key points of the AI Act in a briefing. This document is designed to help readers find key references in the AI Act and get started with a more in-depth look at the text. By the way: another member of WAI Austria, Jeannette Gorzala, designed a series of information points for AI Austria. Head over to AI Austria’s LinkedIn channel to find out more!  

Applause and concerns 

In addition to our work creating content about the AI Act, the majority of our discussions were related to the strengths and weaknesses of the AI Act, what it would mean for our members and their businesses and how the AI Act would live up to its aim of increasing trust in AI by defining trustworthy AI.   One of our core concerns was that while the AI Act allows for updates of the high-risk use cases, categorised according to areas of application, no mechanism was foreseen for updating the areas of applications with high-risk use cases themselves. This could lead to a severe lag in regulation if applications in new areas emerge which present a serious risk to fundamental rights. Sadly, emotion recognition systems – which are scientifically bogus – are acquiring a veneer of respectability through their inclusion in the legislative proposal. Above all, we believe there was a (so far) missed opportunity to legally define AI subjects – in our view, this definition should include any person or business affected by the outputs of an AI system.   Yet despite some weak points, this legislative proposal is a first step in the right direction, and we think it will yield many benefits for those developing AI systems and those affected by AI systems. Our feedback also includes thoughts on the provisions defining high-risk AI, the Annexes listing high-risk AI applications and specifying the technical documentation, as well as the definitions included in Art. 3.  

Next steps 

As mentioned above, the AI Act is so far only a proposal. We look forward to seeing the end result and in the meantime, we are quite happy to see that some of our proposals have also been brought forward by other organisations. For now, we hope you are looking forward to our updated posters once the final AI Act is out!  Valerie Hafez is a member of the board of Women in AI Austria and leads our policy initiatives. An anthropologist by training, she is currently studying science and technology studies at the University of Vienna.