Computer Vision of AI. Generative AI

Advancements in the EU’s AI Act- Meeting of October 24th by Liliana Kotval 

The EU’s AI Act framework was originally presented by the European Commission in April of 2021, with the European Parliament approving its draft in May of this year. The final draft is currently still under negotiation, and once approved, it will be the world’s first comprehensive AI law. If successful upon implementation, it will set a global standard for AI policy. 

This year, the EU Council, Parliament, and Commission have been attending regular trilogue meetings to discuss the AI Act and to establish regulations for manufacturers and users to ensure that AI systems are safe, transparent, traceable, non-discriminatory, and environmentally friendly. So far, throughout the previous three meetings held in June, July, and September, the Commission has categorized the levels of risk posed by AI systems- unacceptable, high, limited, and minimal- and has proposed corresponding regulatory measures.  

At the latest meeting on October 24th, EU policy makers met for some final, intense negotiations regarding the Act. The talks have furthermore clarified exemptions of AI systems from falling under the high-risk category if they prove to be purely accessory (Article 6 of the Act). This includes AI systems that: 

  • Are capable of performing a narrow procedural task 
  • Are able to detect deviations from decision-making patterns 
  • Do not influence a decision, such as whether to provide a loan or to make a job offer 
  • Only improve the quality of a work, such as a smart grammar checker or a translator 

Deciding whether or not an AI system falls under the high-risk category may be ambiguous, and the European Commission wants to retain the authority to modify these exemptions as time passes.  

Furthermore, regarding AI in law enforcement, the negotiators have proposed text that addresses the use of foundation models and of general-purpose AI, however the final decision has not been agreed upon. It may prove difficult to implement, as law enforcement requires some AI tools that the European Parliament may want to ban, such as facial recognition or remote biometric identification systems.  

Overall, many questions are still left unanswered and the trilogue is scheduled to meet again on December 6th, with the hopes of passing the Act by the end of the year. According to Kai Zenner, Adviser for MEP Axel Voss, the negotiations will expand upon the unresolved stumbling blocks of prohibitions in Article 5, foundation models and generative AI, the definition of AI, AI Office, and real-world testing. 

Approval of the Act by the end of this year may be tight, however, Zenner additionally noted that it would be most beneficial to complete negotiations under the current Spanish presidency and before the new 2024 EU presidency. 

Komentarze są wyłączone.