【asean】东盟人工智能治理与伦理指南报告_87页_3mb

ASEAN Guide on AI Governance and Ethics Summary

Core Content

The ASEAN Guide on AI Governance and Ethics is a practical framework aimed at helping organisations in the ASEAN region design, develop, and deploy traditional AI technologies responsibly in commercial and non-military/dual-use applications. It promotes alignment across ASEAN member states and encourages interoperability of AI governance frameworks. The Guide also provides national and regional-level recommendations for governments to implement AI systems in a safe, ethical, and transparent manner.

Main Objectives

• To provide a guide for organisations to ensure responsible AI design, development, and deployment.
• To foster alignment and interoperability of AI governance frameworks across ASEAN jurisdictions.
• To support the development of a regional policy for AI governance and ethics, in line with the ASEAN Digital Masterplan 2025 (ADM2025).
• To build public trust in AI systems by ensuring transparency, fairness, and accountability.

Key Principles

The Guide outlines seven guiding principles for ethical AI governance:

1. Transparency and Explainability

• AI systems should be transparent about their use, decision-making processes, data sources, and purposes.
• Users should be informed of how their data is used and how AI systems arrive at decisions.
• Explainability allows users to understand the reasoning behind AI decisions, promoting trust.
• Techniques such as AI Model Cards, heatmaps, and outcome-based explanations can be used to enhance transparency.

2. Fairness and Equity

• AI systems should avoid reinforcing or amplifying discrimination and bias.
• Regular testing and adjustments should be made to ensure fairness across demographics.
• Diverse and representative datasets are crucial to prevent unjust outcomes.
• Measures should be taken during data collection, preprocessing, training, and inference to mitigate bias.

3. Security and Safety

• AI systems must be secure against malicious attacks and designed to prevent harm.
• Safety measures include risk assessments, testing, and the ability for humans to intervene in unsafe decisions.
• Technical security measures such as robust authentication, encryption, and incident response plans should be implemented.
• Security testing should include vulnerability assessments and penetration testing.

4. Human-centricity

• AI should be designed to benefit human society, including well-being, happiness, and nutrition.
• It should not manipulate users or be used for harmful purposes.
• Dark patterns (e.g., misleading default settings) should be avoided.
• AI deployment should consider its impact on employment and job prospects, with appropriate job redesign strategies.

5. Privacy and Data Governance

• Data privacy and protection must be ensured throughout the AI system lifecycle.
• Data protocols should govern access, use, and deletion of data.
• Compliance with regional data protection laws (e.g., Singapore, Malaysia, Indonesia, Vietnam) is essential.
• Organisations must obtain necessary consent or have a legal basis for data collection and use.

6. Accountability and Integrity

• Clear roles and responsibilities should be defined for all individuals involved in AI design, development, and deployment.
• Accountability mechanisms should be in place to ensure ethical and responsible AI use.
• Integrity refers to the accuracy and reliability of AI systems and their outputs.

7. Robustness and Reliability

• AI systems should be robust and reliable, capable of operating consistently and safely.
• Measures to ensure repeatability and traceability of AI decisions should be implemented.
• Audit trails and data provenance documentation are important for transparency and accountability.

Governance Framework Components

The Guide outlines four key components of the AI governance framework:

1. Internal Governance Structures and Measures

• Establish a multi-disciplinary AI Ethics Advisory Board to oversee AI governance.
• Develop standards, guidelines, tools, and templates to support ethical AI practices.
• Define clear roles and responsibilities for personnel involved in AI lifecycle activities.

2. Determining the Level of Human Involvement

• Three categories of human involvement: human-in-the-loop, human-over-the-loop, and human-out-of-the-loop.
• The level of human involvement depends on the risk level of the AI system.
• Human oversight is crucial for building trust and mitigating risks.

3. Operations Management

• AI system development is an iterative process with multiple stages.
• Conduct risk-based assessments before data collection, processing, and model development.
• Mitigate risks of unjust bias by ensuring representative datasets and testing for fairness.

4. Stakeholder Interaction and Communication

• Build trust with stakeholders and the public through transparent communication.
• Disclose when AI is used in products or services.
• Support employees in adapting to AI-augmented environments through training and awareness.

Recommendations

National-Level Recommendations

• Nurturing AI talent and upskilling workforce: Collaborate with public and private sectors to prepare the workforce for AI integration.
• Supporting AI innovation ecosystem: Create an environment conducive to AI development, including data access and infrastructure.
• Investing in AI research and development: Encourage research on AI ethics, governance, and cybersecurity.
• Promoting adoption of AI governance tools: Implement tools that enhance AI governance and documentation processes.
• Raising awareness among citizens: Educate the public on AI risks and benefits to promote informed decision-making.

Regional-Level Recommendations

• Setting up an ASEAN Working Group on AI Governance: A multi-state group to oversee and implement AI governance initiatives.
• Adapting the Guide for generative AI: Address specific risks such as deepfakes, disinformation, bias, and intellectual property.
• Compiling a compendium of use cases: Showcase how organisations in ASEAN are implementing the Guide in practice.

Use Cases

Several organisations in ASEAN have implemented AI governance practices, including:

• Gojek: Demonstrates responsible AI use in ride-hailing and financial services.
• Aboitiz Group: Implements AI governance in business operations.
• UCARE.AI: Uses AI responsibly in healthcare.
• Smart Nation Group (SNG), Singapore: Shows how Singapore integrates AI into national strategies.
• Ministry of Education, Singapore: Uses AI in education with ethical considerations.
• EY: Provides guidance on AI governance and risk management.

Conclusion

The ASEAN Guide on AI Governance and Ethics serves as a living document that should be periodically reviewed and updated to reflect new developments and regulations in the AI space. It encourages a holistic, transparent, and accountable approach to AI governance and aims to align AI practices with international standards and ethical norms. The Guide is intended to help both organisations and governments ensure that AI is used responsibly and for the benefit of society.

Annexes

• Annex A: AI Risk Impact Assessment Template – provides a structured approach to evaluating AI risks.
• Annex B: Use Cases – showcases practical examples of AI governance implementation across ASEAN.