US Government Introduces Policy for Responsible AI Governance: Key Points

The US government’s Office of Management and Budget (OMB) recently unveiled a comprehensive policy aimed at ensuring responsible deployment of artificial intelligence (AI) by federal agencies. Here’s a summary of the key highlights:

1. Appointment of Chief AI Officers (CAIOs): Each federal agency is mandated to designate a Chief AI Officer (CAIO) within 60 days to lead AI governance efforts.
2. Accountability Requirements: Agencies must submit plans to meet policy goals within 180 days and provide annual AI use case inventories and compliance reports.
3. Role of CAIOs: CAIOs will oversee AI governance, maintain inventories, support equitable AI projects, manage resources, and ensure compliance.
4. Minimum Practices for Safety and Rights Impacting AI:

• AI Impact Assessment: Agencies must conduct and update AI impact assessments, focusing on purpose, risks, data quality, and appropriateness.
• Testing and Evaluation: Rigorous testing of AI systems is required to ensure functionality and mitigate risks.
• Independent Evaluation: Reviews by CAIOs, oversight boards, or relevant offices are necessary to assess AI systems’ benefits and risks.
• Ongoing Monitoring: Continuous monitoring of AI functionality and performance changes is essential, along with human reviews and intervention when necessary.
• Mitigation of Risks: Agencies must mitigate emerging risks to rights and safety, update AI systems accordingly, and discontinue those posing unacceptable risks.
• Human Training and Oversight: Adequate training for operators and human oversight of AI decisions are crucial, along with accountability measures.
• Public Notice and Documentation: Agencies must provide public notices and plain-language documentation on AI systems, ensuring transparency.

5. Additional Practices for Rights-Impacting AI:

• Algorithmic Discrimination: Agencies must identify and mitigate discrimination in AI systems, incorporating feedback from affected communities.
• Ongoing Monitoring and Feedback: Continuous monitoring and feedback mechanisms are required to address AI-enabled discrimination.
• Notification and Remedies: Agencies must inform individuals about adverse decisions resulting from AI use and provide remedies and appeal options.
• Options to Opt-Out: Agencies must offer opt-out options from AI decisions, providing human alternatives where feasible.

6. Waiver Provision: Agencies may seek waivers from certain AI governance requirements based on risk assessments conducted by CAIOs.
7. Guidelines for Federal Procurement of AI: Measures include transparency, responsible procurement practices, and assessments for environmental efficiency and sustainability.

This policy framework aims to ensure that AI deployment by federal agencies prioritizes safety, rights protection, and accountability, setting a standard for responsible AI governance in the public sector.