扩展的东盟人工智能治理和伦理指南——生成型人工智能

AI Governance and Ethics - Generative AI Summary

Core Content

This document provides an expanded guide on AI governance and ethics specifically for Generative AI (Gen AI) within the ASEAN region. It complements the ASEAN Guide on AI Governance and Ethics (2024) by offering policy recommendations and use cases to ensure the responsible adoption of Gen AI technologies. The guide emphasizes the need for a coordinated, pro-innovation approach that balances economic growth, safety, and regional harmonization.

Main Risks of Gen AI

The document outlines six key risks associated with Gen AI:

1. Mistakes and anthropomorphism: Gen AI can produce hallucinated or coherent but incorrect outputs, such as medical advice or software code.
2. Factually inaccurate responses and disinformation: Gen AI can amplify false or misleading information, which may distort public perception and erode trust in reliable sources.
3. Deepfakes, impersonation, fraudulent and malicious activities: Gen AI can be used to create realistic fake content, such as deepfakes and phishing emails, increasing the risk of identity theft and confidentiality breaches.
4. Infringement of intellectual property rights: The use of copyrighted works as training data may lead to legal issues or resemblance to existing content.
5. Privacy and confidentiality: Gen AI may memorize and reproduce sensitive training data, or be used to reconstruct confidential information.
6. Propagation of embedded biases: Gen AI can reflect biases from training data, leading to biased or toxic outputs that reinforce stereotypes.

In addition, the guide highlights frontier and systemic risks that may arise from the long-term evolution of highly advanced Gen AI systems, such as:

• Misalignment of AI systems: Self-improving, autonomous AI systems may pursue goals that conflict with human interests.
• Systemic risks: These include labor market impacts, privacy concerns, and environmental effects, particularly due to the high energy consumption of Gen AI systems.

Key Policy Recommendations

The guide recommends nine policy dimensions to manage Gen AI risks and promote responsible use:

1. Accountability: Establish shared responsibility frameworks across the value chain, including developers, cloud providers, and users.
2. Data: Implement privacy-by-design principles, ensure data traceability, and maintain data quality and integrity.
3. Trusted Development and Deployment: Promote ethical and transparent practices in the development and deployment of Gen AI systems.
4. Incident Reporting: Develop robust incident reporting mechanisms to address misuse and unintended consequences.
5. Testing and Assurance: Create quantitative benchmarks and dynamic testing protocols to ensure system robustness and reliability.
6. Security: Enhance security measures against malicious attacks, including prompt injection, data poisoning, and shadow IT.
7. Content Provenance: Ensure traceability and transparency of AI-generated content to prevent misinformation and fraud.
8. Safety and Alignment Research & Development: Support research into AI safety and alignment, including ethical and technical safeguards.
9. AI for Public Good: Encourage the use of Gen AI for social and economic benefit, ensuring it serves public interest.

Use Cases

The guide includes four use cases from public and private institutions in ASEAN that have implemented AI governance and ethics practices:

• PhoGPT (Vietnam): Demonstrates how Gen AI can be used in a regional context to enhance public services and business operations.
• VinAI (Vietnam): Shows the application of Gen AI in innovation and development with a focus on ethical and responsible practices.
• Project Moonshot (Singapore): Illustrates collaborative efforts to ensure AI safety and alignment.
• AI Verify Foundation (Singapore): Highlights the importance of transparency and accountability in AI governance.
• Responsible AI Internal Programme (Accenture, ASEAN-wide): Provides a framework for ethical AI use across the region.
• ThaiLLM, BDI, NSTDA, VISTEC and collaborators (Thailand): Reflects national and regional collaboration to develop trustworthy Gen AI systems.

Guiding Principles

The guide emphasizes several guiding principles to ensure trustworthy and ethical use of Gen AI:

• Transparency and Explainability: Ensure users understand how AI systems work and the decision-making process.
• Fairness and Equity: Prevent AI from amplifying or reinforcing biases and ensure representative and inclusive data.
• Security and Safety: Implement risk-based approaches and technical safeguards to minimize harm.
• Human-centricity: Prioritize human values and well-being in AI development and use.
• Privacy and Data Governance: Adopt privacy-by-design and ensure legal compliance in data handling.

Conclusion

This guide is designed to be used in conjunction with the ASEAN AI Guide (2024) and provides policy considerations for responsible Gen AI adoption. It highlights the importance of a holistic, coordinated approach to ensure trust, safety, and ethical use of Gen AI across the ASEAN region. Continuous research and policy development are essential to address emerging risks and realize the full potential of Gen AI in economic and social development.