世界经济论坛-AI政府采购指南（英文）-2019.9-18页_726kb

AI Procurement Guidelines Summary

Core Content

The World Economic Forum published the Guidelines for AI Procurement in September 2019, aimed at guiding governments and international bodies in the responsible and ethical use of artificial intelligence (AI) in public sector procurement. The document outlines 10 high-level recommendations to ensure that AI systems are procured in a way that aligns with public benefit, transparency, accountability, and ethical standards.

Main Objectives

• To ensure that AI systems are used in a manner that safeguards public benefit and well-being.
• To address risks and challenges associated with AI deployment, including bias, privacy, and transparency.
• To promote innovation and fair competition among AI solution providers.
• To align AI procurement with national strategies and existing legal and ethical frameworks.

Key Principles and Recommendations

1. Use innovative procurement processes

• Focus on outlining problems and opportunities rather than prescribing specific solutions.
• Encourage iteration and testing of AI technologies through phased procurement or proof-of-concept stages.
• Support early market engagement and provide opportunities for new entrants, including start-ups, to compete.

2. Define public benefit and assess risks

• Clearly articulate the relevance of AI to the problem in the RFP.
• Open to alternative technical solutions and ensure public benefit is a central consideration.
• Conduct an initial AI risk and impact assessment, incorporating findings into the RFP and revisiting at key decision points.

3. Align with governmental strategies

• Consult AI national strategies and related guidance documents.
• Collaborate with other government bodies to share insights and learn from their experiences.
• Use strategic procurement to support broader AI development and deployment goals.

4. Incorporate legislation and codes of practice

• Review relevant laws, rights, and administrative rules governing data and AI applications.
• Reference laws and regulations in the RFP to ensure compliance.
• Consider industry best practices and soft law norms such as data ethics frameworks and freedom of information laws.

5. Ensure technical and administrative feasibility of data access

• Establish proper data governance mechanisms from the start.
• Assess data availability and define data-sharing strategies with vendors.
• Ensure that data access is both feasible and ethically sound.

6. Highlight technical and ethical limitations of data usage

• Address data quality and historical bias in the RFP.
• Require bidders to describe strategies for mitigating data shortcomings.
• Plan for addressing any missed limitations during the procurement process.

7. Work with a diverse, multidisciplinary team

• Encourage diverse teams to develop and evaluate AI solutions.
• Require successful bidders to assemble teams with the appropriate skills.
• Diversity helps avoid inadvertent harms or discrimination against vulnerable groups.

8. Focus on algorithmic accountability and transparency

• Promote a culture of accountability across AI solutions.
• Ensure AI decision-making is transparent and explainable.
• Explore mechanisms for internal and external interpretability of algorithms.

9. Ensure ongoing engagement with AI providers

• Implement a process for continuous knowledge transfer and risk assessment.
• Require providers to offer training and guidance for non-specialists.
• Monitor AI solutions throughout their lifecycle to manage risks and ensure proper use.

10. Create a fair and level playing field for AI providers

• Engage a wide variety of AI solution providers.
• Encourage early and frequent collaboration with vendors.
• Ensure interoperability and open licensing to avoid vendor lock-in.

Key Information

• The guidelines emphasize the importance of innovation, transparency, and ethical considerations in AI procurement.
• Governments should avoid the "black box" approach by ensuring accountability and explainability of AI systems.
• A multidisciplinary team is essential to identify and mitigate risks, particularly those related to bias and discrimination.
• The procurement process should be iterative and flexible, allowing for testing and refinement of AI solutions.
• Risk assessment should be a continuous process, covering all stages from design to maintenance.
• The guidelines are a living document, designed to evolve with new practices and feedback from stakeholders, including pilot programs in the UK, UAE, and Colombia.

Conclusion

The Guidelines for AI Procurement serve as a foundational resource for governments seeking to adopt AI responsibly. They aim to ensure that AI systems are procured in a way that maximizes public benefit while minimizing risks. By promoting transparency, accountability, and innovation, the guidelines help governments navigate the complexities of AI procurement and set a standard for ethical and effective public use of AI.