治理人工智能：未来蓝图-42页_1mb

Governance of AI: Key Insights from "Blueprint for the Future"

1. Ethical Foundation & Human Control

   • Brad Smith argues AI governance must address its unprecedented capability to perform human decisions. He emphasizes accountability and ensuring AI remains under human control, drawing parallels to historical tech governance failures (e.g., social media).
   • Core Principle: Accountability is foundational; no technology or entity is above the law.

2. Five-Point Governance Blueprint

   • Governance through Public Policy:

     • Implement and build upon frameworks like NIST’s AI Risk Management Framework.
     • Require safety brakes for AI controlling critical infrastructure (e.g., power grids, transportation).
     • Develop legal/regulatory frameworks aligned with AI’s technology stack (e.g., differentiate infrastructure models vs. applications).
     • Promote transparency (e.g., AI-generated content labeling, national registries).
     • Foster public-private partnerships to address societal challenges (e.g., combatting disinformation, supporting vulnerable groups).

   • Governance within Microsoft:

     • A journey from early commitments to scaling Responsible AI, with 350+ dedicated employees.
     • Framework includes principles, governance structures, risk mitigation tools (e.g., red teaming, fairness classifiers), and actionable standards.
     • Sensitive Uses Program: Provides case-specific scrutiny for high-risk AI applications (e.g., voice banking, facial recognition).

3. Transparency & Access

   • Commitment to releasing annual transparency reports and developing tools like Azure Content Safety.
   • Push for academic/nonprofit access through resources like the National AI Research Resource (NAIRR).

4. International Collaboration

   • Advocacy for global norms, citing examples like the Christchurch Call and efforts to counter AI weaponization.
   • Support for multilateral initiatives (e.g., UNESCO AI recommendations) and interoperable risk mitigation approaches.

5. Conclusion

   • AI’s potential is “extraordinary but double-edged” – balancing innovation with safety requires proactive governance, diverse talent (including ethical AI design), cross-sector partnerships, and foundational accountability.