扩展的东盟人工智能治理和伦理指南——生成型人工智能（英）_52页_4mb

AI Governance and Ethics - Generative AI Summary

Core Content

This document provides an expanded guide on AI governance and ethics, specifically focusing on Generative AI (Gen AI), to complement the ASEAN AI Governance and Ethics Guide (2024). It outlines the opportunities and risks associated with Gen AI and proposes a set of policy recommendations to ensure its responsible and ethical adoption across the ASEAN region.

Main Objectives and Target Audience

• Objective: To guide ASEAN policymakers in understanding and addressing the challenges and opportunities of Gen AI.
• Target Audience: ASEAN policymakers, industry stakeholders, and relevant regulatory bodies.

Key Risks of Gen AI

The document identifies six primary risks of Gen AI, which are central to the policy recommendations:

1. Mistakes and Anthropomorphism: Gen AI can produce coherent but incorrect information, often referred to as "hallucinations".
2. Factually Inaccurate Responses and Disinformation: Gen AI can amplify false or misleading information, affecting public perception.
3. Deepfakes, Impersonation, Fraudulent and Malicious Activities: The ability to generate realistic content can lead to identity theft and misinformation.
4. Infringement of Intellectual Property Rights: Use of copyrighted material in training data or generation of content similar to existing works can lead to legal issues.
5. Privacy and Confidentiality: Gen AI systems may inadvertently disclose sensitive information or allow malicious actors to reconstruct private data.
6. Propagation of Embedded Biases: Gen AI can inherit and reflect biases from training data, leading to unfair or harmful outputs.

Additionally, the document highlights frontier and systemic risks, such as the potential misuse of Gen AI in creating harmful content, the controllability of highly advanced AI systems, and the environmental impact of computing power usage.

Policy Recommendations

The document recommends a holistic and regionally interoperable approach to managing Gen AI risks and promoting opportunities. The key recommendations include:

1. Accountability: Develop shared responsibility frameworks across the Gen AI value chain, including developers, deployers, users, and cloud providers.
2. Data Governance: Ensure proper management of data used in training foundation models, including transparency, privacy, and data integrity.
3. Trusted Development and Deployment: Implement measures to ensure that Gen AI systems are developed and deployed in a manner that is safe, transparent, and aligned with ethical standards.
4. Incident Reporting: Establish clear incident reporting mechanisms to address misuse or errors in Gen AI systems.
5. Testing and Assurance: Use testing and evaluation to ensure the accuracy, consistency, and reliability of Gen AI outputs.
6. Security: Enhance security measures to protect against attacks such as prompt injection and data poisoning.
7. Content Provenance: Ensure that content generated by Gen AI is traceable and attributable to its source.
8. Safety and Alignment Research & Development: Invest in research to improve the safety and alignment of Gen AI systems with human values.
9. AI for Public Good: Promote the use of Gen AI in ways that benefit society, such as improving public services and supporting inclusive growth.

Guiding Principles

The document outlines several guiding principles for the Gen AI framework:

• Transparency and Explainability: Ensure that users understand how Gen AI systems operate and the data they use.
• Fairness and Equity: Prevent AI from amplifying existing biases and ensure equitable treatment across demographics.
• Security and Safety: Mitigate risks through robust security measures and risk assessments.
• Human-Centricity: Prioritise human interests and values in the design, development, and deployment of Gen AI.
• Robustness and Reliability: Ensure that Gen AI systems are reliable and consistent in their outputs.

Use Cases

The document includes use cases from various institutions in ASEAN that demonstrate responsible practices in Gen AI implementation:

• PhoGPT (Vietnam): An example of Gen AI in action, highlighting its potential for innovation and practical application.
• Project Moonshot (Singapore): Demonstrates the use of AI for public good and transparency.
• Responsible AI Internal Programme (Accenture, ASEAN-wide): Shows a comprehensive approach to AI ethics and governance.
• ThaiLLM, BDI, NSTDA, VISTEC and collaborators (Thailand): Illustrates collaboration in developing ethical and responsible AI systems.

Conclusion

The document concludes that while Gen AI presents significant opportunities for economic and social development, it also poses unique challenges that require careful governance. A coordinated and proactive approach by ASEAN policymakers is essential to ensure the safe, ethical, and beneficial use of Gen AI across the region.

Appendix and Methodology

• The document includes an Appendix with detailed use cases and a Methodology section outlining the approach taken in developing the recommendations.
• References are provided to support the policy considerations and recommendations.

This guide is intended to be used in conjunction with the ASEAN AI Governance and Ethics Guide (2024) and is a resource for promoting a trusted and responsible Gen AI ecosystem in the region.