金融机构的监管预期：对人工智能监管模型的约束（英）-28页_367kb

报告摘要

Summary of FSI Insights on Policy Implementation No 35: Humans keeping AI in check – emerging regulatory expectations in the financial sector

Core Content

This document explores the evolving regulatory expectations surrounding the use of artificial intelligence (AI), including machine learning (ML), in the financial sector. It highlights the need for a balanced and proportionate approach to AI governance, ensuring that the benefits of AI are maximised while mitigating associated risks. The paper examines existing regulatory frameworks, guidance, and principles across multiple jurisdictions, as well as challenges in implementing these expectations.

Main Points

AI and Financial Services: AI and ML are transforming financial services, offering improved efficiency and consumer access. However, they also introduce new risks, including unintended bias, discrimination, and operational vulnerabilities.
Regulatory Frameworks: Financial authorities across various jurisdictions have started to develop or update AI governance frameworks. These include principles-based guidance, discussion papers, and in some cases, formal legislation. The European Union (EU) is a notable exception with a proposed regulation for harmonised AI rules.
Common Principles: Several common principles underpin AI governance in the financial sector:
- Reliability and Soundness: Ensuring AI models are accurate, reliable, and free from bias that could lead to discriminatory outcomes.
- Accountability: Clear assignment of responsibility, with a focus on human involvement in model development and decision-making. Includes both internal and external accountability.
- Transparency: Requiring explainability and auditability of AI models, as well as external disclosure to data subjects.
- Fairness and Ethics: Addressing biases in AI to prevent discrimination and ensuring ethical use, including respect for consumer rights and societal norms.
Challenges in Implementation:
- Transparency: Ensuring that AI models are explainable and not "black boxes" is a key challenge.
- Reliability and Soundness: Maintaining model accuracy and avoiding harmful outcomes requires continuous monitoring and validation.
- Accountability: Assigning clear responsibility in AI-driven decisions is complex, especially with the potential for machine autonomy.
- Fairness and Ethics: Defining and applying fairness in AI is still evolving, and ethical considerations extend beyond bias to include broader issues like privacy and data protection.
Proportionality and Coordination: Given the varying levels of risk and impact from AI models, regulators should apply a proportional approach. There is also a need for coordination between prudential and conduct authorities to ensure consistent oversight.
International Standards: The paper suggests that international standard-setting bodies could develop global guidance on AI governance, which would be especially useful for jurisdictions in early stages of digital transformation.

Key Information

Existing Regulatory Standards: While some principles are already in place (such as data privacy, third-party dependency, and operational resilience), there is a growing need for more specific AI-related guidance.
EU Leadership: The EU has taken a leading role in AI regulation, with the proposed regulation on AI and the FEAT (Fairness, Ethics, Accountability, and Transparency) principles from the Monetary Authority of Singapore (MAS) being notable examples.
Human Oversight: There is an increasing emphasis on human intervention in AI processes, such as "human-in-the-loop" and "human-on-the-loop" mechanisms, to ensure accountability and prevent ethical or legal issues.
Data Privacy and Protection: AI systems must comply with data privacy laws, including the need for customer consent and the secure handling of personal data throughout the AI lifecycle.
Future Outlook: The paper calls for continued dialogue and exchange of experiences among international regulators to refine AI governance frameworks and potentially develop global standards.

Conclusion

The financial sector is at a crossroads with the increasing adoption of AI. While the technology offers significant benefits, it also poses unique risks that require careful management. Regulatory frameworks must evolve to address these challenges, with a focus on fairness, accountability, transparency, and ethical considerations. A coordinated and proportionate approach is essential to ensure that AI is used responsibly and that its deployment aligns with the core objectives of financial stability and consumer protection.

Annex: Proposed AI Regulation in the EU

The EU has proposed a regulation that aims to harmonise AI rules across all industries, marking a first-of-its-kind effort.
This regulation includes principles for the ethical and trustworthy use of AI, focusing on transparency, accountability, and fairness.
It also addresses issues like data privacy, third-party dependency, and operational resilience, ensuring that AI systems are used in a way that aligns with broader regulatory goals.