The regulation aims to ensure that AI systems placed on the European market and used in the EU are safe and respect fundamental rights. The AI Act sets rules for large, powerful AI models, ensuring they do not present systemic risks to the Union.
It introduces obligations for "high-risk" AI systems used in critical infrastructure, transport, education, jobs, law enforcement, etc. This includes data governance, record-keeping, human oversight and accuracy/robustness obligations. Among others, it bans practices that pose unacceptable risks such as "social scoring" and “real-time remote biometric identification" unless under specific circumstances like searching for a missing child.
It creates a new EU-wide certification framework for "high-risk" AI systems to ensure they meet requirements for health, safety and protection of fundamental rights.
To that end, it establishes transparency obligations for AI systems used to interact with humans, detect emotions, and recommend jobs/education. For example, AI emotion recognition or biometric categorization systems used in employment or education must undergo prior human review.
Overall, Systems posing risks like bias must be tested for accuracy, robustness and cybersecurity before being used. Certified AI systems must display a conformity mark and providers must register in an EU database which will be accessible to enforcement authorities.
Companies that don't comply could be fined up to 7% of their global turnover.
The Artificial Intelligence Act is the first of its kind on a global scale. It is intended to establish a uniform set of rules across the EU to regulate AI and avoid fragmentation of the digital single market.
It was first proposed by the European Commission in April 2021 following extensive consultation with experts and stakeholders over two years. The draft legislation was then debated in the European Parliament and underwent negotiations between representatives of the 27 EU member states on the Council.
The final draft went to a plenary vote of Members of the European Parliament again. Once passed by MEPs, the final text was negotiated between Parliament and the Council in "trilogue" meetings to reach a compromise.
The law still needs to go through a few final steps for approval, but the political agreement means its key outlines have been set. The bans on banned AI will take effect in six months, the transparency requirements in 12 months and the full set of rules in around two years. National supervisory authorities will monitor compliance and be empowered to restrict or prohibit AI systems that don't meet requirements to protect people's rights and safety.
Read more, here
--
CC-BY-4.0: © European Union 2023– Source: EP
