I just came across two reports from feeds I monitor, one from LinkedIn, the other from Canlii. The first is the report, 2024 State of the AI Regulatory Landscape. This report provides an overview of the evolving AI regulatory frameworks across major jurisdictions, particularly the United States, European Union, and China. As AI technologies advance, governments are increasingly focused on implementing effective governance structures to mitigate potential risks while fostering innovation. The report highlights various regulatory approaches and policy choices, each addressing different aspects of AI development and deployment.
A key focus of the report is the classification of AI systems. Different regulatory approaches emphasize various factors such as risk, compute power, and application. The EU’s AI Act adopts a risk-based approach, classifying AI systems into unacceptable, high, limited, and minimal risk categories, with corresponding levels of oversight and regulation. In contrast, the U.S. tends to regulate hardware and computational power, particularly targeting the export of high-end AI chips to control China’s access to AI-enabling technologies. China’s approach emphasizes the regulation of algorithms broadly, with significant attention on content control and social alignment.
Another significant topic in the report is AI evaluation and risk assessments. Governments are eager to develop tools to assess AI models, particularly in terms of safety, capability, and alignment. These assessments are crucial for understanding how AI systems might harm society, from perpetuating bias to posing security threats. The report outlines that while some regulatory frameworks, such as the U.S. AI Bill of Rights and the EU AI Act, include mandates for AI risk assessments, the methodologies for these evaluations are still in development. New initiatives in the U.S. and the UK are working on establishing safety evaluations for AI models, with the ultimate goal of creating standardized assessments that can be mandated globally.
A notable innovation in AI governance is the creation of model registries, which function as centralized databases of AI systems used in real-world applications. These registries serve as foundational tools for AI governance, allowing governments to track AI models, their capabilities, and their compliance with safety regulations. China’s algorithm registry, for example, requires detailed reporting from AI developers, including security assessments, while the EU’s AI Act mandates that high-risk AI systems must undergo conformity assessments before public release.
The report also delves into emerging issues such as open-source AI models, incident reporting, and cybersecurity concerns related to frontier AI models. It discusses how open-source AI, while fostering innovation, poses challenges in terms of security and intellectual property. Incident reporting mechanisms are gaining traction as a way to track and mitigate unexpected outcomes or malfunctions in AI systems, providing critical feedback for improving governance frameworks.
If you want more information on the different AI regulatory frameworks, particularly in the EU, Uk, and the U.S., I invite you to read by blog post, AIDA’s regulation of AI in Canada: questions, criticisms and recommendations.
The other report is Report on Artificial Intelligence and Civil Liability, prepared for the British Columbia Law Institute. The BCLI report examines various models of liability that could be applied to AI, ultimately advocating for a fault-based approach rather than strict liability. It is a good complement to the AI Regulatory Landscape report, which deals with regulatory models rathe than the existing common law remedies for dealing with risk based AI systems.
** Note, the first draft of this post was generated by ChaGPT.
