2024-01-19-麦肯锡-释放生成式人工智能的潜力_政府机构面临的三个关键问题_8页_447kb

Summary

Electronic documents typically begin with a title page or cover page that includes the report's title, author information, date, etc.

McKinsey & Company's article "Public Sector Practice: Unlocking the potential of generative AI" examines the application of generative AI (gen AI) in government agencies, addressing three key questions and proposing an eight-step plan for implementation.

Key Question 1: Addressing the Potential Risks of Gen AI

• Risks include unpredictability, inaccuracy, bias, misuse for spreading propaganda, cybersecurity threats, and transparency issues. Mitigation strategies involve establishing regulations, guidelines, awareness programs, and collaborations with global efforts like EU AI frameworks. For instance, governments should revise policies continually to keep pace with rapid AI evolution.

Key Question 2: Transforming Service Delivery

• Government agencies should use the 4Cs framework (Content generation, Content summarization and synthesis, Coding and software, Customer engagement) to prioritize high-impact, low-risk use cases. Examples include tools like Singapore's GovTech app or the US DoD's Acqbot. This transformation aids in areas like education, healthcare, and urban development, streamlining services through AI.

Key Question 3: Developing National Gen AI Foundation Models

• While some governments aim to create national foundation models (e.g., through partnerships with private companies like Cohere or Google), it requires significant investment in talent, computing power, and risk management. Most agencies are opting to use existing models and fine-tune them with proprietary data, due to barriers like data sovereignty and legal implications.

Eight-Step Plan for Implementation

1. Define your organization’s risk posture by creating a mitigation plan with policies and awareness sessions.
2. Identify and prioritize use cases based on impact, feasibility, and risk tolerance.
3. Select underlying models and upgrade technical infrastructure.
4. Ensure necessary skills and roles, such as AI engineers or prompt engineers, are available.
5. Develop AI apps jointly with end users to enhance feedback and accuracy.
6. Keep humans in the loop to monitor and assess AI systems.
7. Design a comprehensive communication plan with disclaimers on AI limitations.
8. Start with small-scale pilots and scale up based on results.

Potential Applications of Gen AI

• Gen AI can be applied in various government domains, including education (virtual tutors, audit reports), defense (intelligence reports), agriculture (crop management), urban development (planning tools), health (diagnostic assistance), and tax services, with a focus on improving efficiency and public service delivery.

This summary draws from McKinsey's December 2023 article, authored by key figures in the Public Sector Practice team.