Help! The Complexities of Regulating Artificial Intelligence

The regulation of AI raises a myriad of questions that touch on ethical, legal, and practical considerations. Dr. Matthew Liao S Matthew Liao from New York University School of Global Public Health spoke about the issues around regulation. he said that to navigate these challenges effectively, we must carefully examine what should be regulated, why, who should regulate, when regulation should occur, where it should be applied, and how it should be implemented. I found this particularly interesting as currently I am working with a local authority as we battle through the design of an AI policy and governance framework.

Below is a brief exploration of these profound questions. It would be really great to get some ideas from people who are also working through these very thorny but essential issues.

1. What Should Be Regulated?

AI systems are composed of multiple layers, and regulation could focus on different aspects, including:

• The Data: Datasets used to train AI are often riddled with biases, inaccuracies, or privacy concerns. Should regulation focus on ensuring data quality, representativeness, and ethical sourcing?
• The Algorithms: Algorithms underpinning AI systems can perpetuate bias or lead to harmful unintended consequences. Should there be oversight on how they are designed, tested, and implemented?
• The Risks: Some advocate for risk-based regulation, targeting AI systems with the greatest potential for harm, such as autonomous weapons or healthcare diagnostics.
• Sector-Based Regulation: Should AI regulation differ by sector, with stricter rules for critical areas like healthcare and finance, while allowing more flexibility in others, such as entertainment?

Key Question: Should regulation focus on specific components of AI or adopt a holistic approach that addresses the entire lifecycle of AI systems?

2. Why Should We Regulate?

The rationale for AI regulation varies depending on the perspective:

• Ethical Reasons: Regulation can ensure that AI systems respect human rights, prevent harm, and promote fairness and equality.
• National Interests: Governments may regulate AI to protect their strategic interests, maintain technological leadership, or ensure national security.
• Legal Compliance: Regulation may be necessary to align AI with existing laws, such as data protection (e.g., GDPR) or anti-discrimination legislation.

Key Question: Is the primary goal of regulation to uphold ethical principles, advance national interests, or ensure legal compliance—or should it achieve a balance of all three?

3. Who Should Regulate?

Determining who should lead AI regulation is equally challenging:

• Experts: AI researchers and ethicists have the technical knowledge to design effective and enforceable regulations. However, they may lack broader societal perspectives.
• Governments: Governments have the authority to create laws and hold organisations accountable, but may lack the technical expertise to draft nuanced regulations.
• The Public: Public input ensures that regulations reflect societal values, but widespread participation can complicate and slow down the process.

Key Question: Should regulation be left to experts, driven by governments, or shaped by public participation—or should all stakeholders work collaboratively?

4. When Should We Regulate?

Timing is crucial in determining the effectiveness of regulation:

• Upstream Regulation: Focuses on the design and development phase of AI systems, ensuring issues like bias and safety are addressed early.
• Downstream Regulation: Focuses on deployment and real-world applications, addressing issues like misuse, harm, or unintended consequences.
• Lifecycle Regulation: Encompasses ongoing oversight throughout the AI system's lifecycle, from inception to decommissioning.

Key Question: Should regulation occur early in the development process, only after AI is deployed, or continuously throughout its lifecycle?

5. Where Should It Be Regulated?

The geographical scope of AI regulation is another complex issue:

• Local Regulation: Focused on community-level impacts, addressing localised concerns such as bias in public services.
• National Regulation: Ensures that AI systems align with a country’s laws, values, and priorities. However, this approach risks creating fragmented standards across borders.
• Global Regulation: AI operates across borders, and many argue for global standards to ensure consistency and fairness. However, international cooperation can be slow and fraught with political disagreements.

Key Question: Should regulation be localised, nationally driven, or developed at a global level to reflect AI's cross-border nature?

6. How Should We Regulate?

The approach to regulation will significantly influence its effectiveness and adoption:

• Voluntary Guidelines: These allow for flexibility and innovation but may lack enforceability and consistency across organisations.
• Hard Laws and Regulations: Legally binding frameworks ensure accountability but are often slow to develop and adapt to rapidly evolving technologies.
• Hybrid Approaches: A combination of voluntary guidelines and enforceable laws could provide both flexibility and accountability.

Key Question: Should we rely on voluntary compliance, hard regulations, or a hybrid model that balances innovation with oversight?

Why We Must Think These Questions Through

AI regulation cannot be rushed or one-dimensional. It requires a nuanced approach that balances competing priorities: fostering innovation while preventing harm, enabling flexibility while ensuring accountability, and promoting national interests while addressing global challenges.

Each question—what, why, who, when, where, and how—must be carefully considered to create a regulatory framework that is both effective and adaptable. With so much at stake, the key lies in collaboration, interdisciplinary dialogue, and the willingness to adapt as AI continues to evolve.

Vivienne Neale is an Honorary Research Associate at Hull University,UK