Why Governments Need To Take a More Active Role in Regulating AI – The Markup

🌈 Abstract

The article discusses the role of public policy in regulating AI systems and the current state of AI legislation in Canada, the US, and Europe. It covers topics such as the definition of AI, the differences between various AI regulatory frameworks, desirable properties of AI systems, and the challenges of ensuring accountability and auditing AI deployments. The article also touches on the funding allocated by the Canadian and US governments for AI development and the gaps in this funding.

🙋 Q&A

[01] AI Legislation and Regulation

1. How do the AI regulatory frameworks in Canada, the US, and Europe differ?

• The EU has a "risk-based" framework that clearly outlines different types of AI systems and their associated risks, while the proposed Canadian framework has an unclear definition of "high-impact" systems and lacks prohibitions on certain types of AI.
• The US approach is decentralized, with multiple initiatives across different agencies and states, but no singular federal-level legislation.

2. What are some key properties that an AI system deployed in the real world should have?

• Accountability: Clear responsibility and complaint mechanisms
• Transparency: Disclosure of the development and procurement process
• Functionality: Ability to achieve the intended goal
• Fairness: Robustness against perpetuating biases

3. How can agencies decide whether an AI technology is ready for real-world deployment?

• Conduct public consultations to ensure the technology is addressing a real need
• Assess the functionality and robustness of the system, following relevant standards and regulations
• Establish clear accountability mechanisms for the use of the technology

4. What does good accountability for AI systems look like, and how can the public raise concerns?

• Complaint mechanisms that are easy to access and ensure timely resolution of issues
• Options for raising concerns beyond just the technology provider
• Empowered regulators who can conduct proactive audits and impose penalties

[02] AI Funding and Regulatory Infrastructure

1. How are the Canadian and US governments allocating funding for AI development and regulation?

• Canada is allocating $2.4 billion CAD for public computing resources for AI models, but only $5.1 million CAD for the office of the AI and Data Commissioner, which is seen as insufficient.
• The US has launched a $2.6 billion pilot for AI computing infrastructure, but the funding for regulatory infrastructure is unclear.

2. What are the gaps in the current funding for AI regulation?

• Insufficient funding for building specialized regulatory teams within the government, with expertise in the socio-technical implications of AI.
• Lack of funding for developing in-house auditing capabilities, rather than relying on industry-provided services.