Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

EU AI Act: the Regulatory Framework on the Usage of Machine Learning in the European Union

Jul 21, 2023 - infoq.com
The EU Council has begun negotiations on the proposed regulation structure for the operation and governance of machine learning applications, with an agreement expected by the end of 2023. The EU Act takes a risk-based approach, aiming to avoid disproportionate prescriptions when executing regulations. It focuses on the general risk factor of any algorithmic and model-based system, including machine learning models and possible future intelligent systems. The Act has gained attention due to the increased plausibility of generative models' output, and is part of a global effort to set principles for future developments and applications of model-based systems.

The EU AI Act's principles are based on four types of risk factors, with different levels of transparency, rules, obligations, and monitoring for providers and users depending on the potential adversarial effects of the system. Minimal-risk systems are allowed free usage, limited-risk systems require providers to inform users of interaction with a ML system, high-risk systems are subject to the highest level of obligations and monitoring, and unacceptable-risk systems will be prohibited. The Act also proposes stricter assessment, regulation, and monitoring procedures for high-risk systems, including registration in the EU Database and disclosure of content generation by a model.

Key takeaways:

  • The EU has started negotiations for the implementation of a regulatory structure for machine learning applications, with an agreement expected by the end of 2023.
  • The EU AI Act takes a risk-based approach, categorizing systems into minimal-risk, limited-risk, high-risk, and unacceptable-risk, with varying levels of transparency, rules, obligations, and monitoring for each category.
  • High-risk and unacceptable-risk systems are of particular concern, with the latter involving the exploitation of sensitive data or manipulation of cognitive and behavioural actions. High-risk systems, often involving large uninterpretable deep networks, will be subject to stricter regulation and monitoring.
  • The EU AI Act is distinct from the EU Data Act, which focuses on broader goals for the EU's data economy and data sovereignty.
View Full Article

Comments (0)

Be the first to comment!