A provisional agreement was reached upon after a final negotiation round between the EU Parliament, EU Council, and EU Commission meaning this milestone legislation is nearing implementation, making it the very first law that regulates Artificial Intelligence in the world. This ‘historic’ agreement as described by Thierry Breton will open the doors to a new era of AI innovation.
The underlying priority of the European Parliament is to facilitate safe use of AI systems and enable systems used in the EU to be non-discriminatory, and environmentally friendly. Accordingly, the proposed legislation contains an emphasis on transparency and accountability.
In an effort to implement the overall objective, the new regulations will introduce a risk-based classification system. The risk-based approach will establish obligations for providers and users depending on the risk the AI system presents and will determine through a set of comprehensive rules the level of risk an AI technology would pose so as to classify it under respective tiers.
The following paragraphs will highlight the different classifications and will indicate important takeaways from the framework:
AI systems that are considered to be a threat to people’s safety and rights will fall under this category and be banned. This includes AI systems or applications that make use of cognitive manipulations with the aim to circumvent people’s free will, real-time and remote biometric identification systems such as facial recognition and systems that allow social scoring.
Systems that negatively affect the safety of people or their fundamental rights, will be classified as high-risk. These systems will, amongst others, be subject to a rigorous assessment before being released on the market. Any organisation employing an AI system that is deemed as high risk, will, amongst others, have to comply with to record-keeping requirements, transparency obligations imposed and be subject to an impact assessment.
Generally, AI systems must be designed to operate in a way which prevents generating illegal content and must disclose that the content was generated by AI. Moreover, publishing of any copyrighted data that was used for training purposes will become mandatory.
If classified as limited risk, AI systems would have to comply with minimal transparency requirements where users should be informed that they are interacting with AI (such as AI enabled chatbots) thus allowing users to make informed decisions about their involvement with AI.
With the aim of supporting AI innovation, exemptions to the rules were enabled for research activities and AI components provided under open-source licenses. The legal framework also fosters the use of regulatory sandboxes or controlled environments in order for AI to be tested before it is deployed.
In regards to compliance, the EU AI Office will have the responsibility of monitoring how the AI rulebook is being implemented and citizens will also have the right to file complaints about AI models that drastically impacted their rights.
Undoubtedly, the enactment of the EU AI Act will be the cornerstone of harmonizing the use, embodiment and regulation of artificial intelligence and a pivotal moment for setting a global precedent. As the co-rapporteur of the Committee, Dragoş Tudorache, stated “the AI Act is very likely the most important piece of legislation in this mandate. It’s the first piece of legislation of this kind worldwide, which means that the EU can lead the way in making AI human-centric, trustworthy and safe”.