The US government is intensifying its efforts to ensure AI systems are accountable and ethically sound, demonstrated at the AI World Government event. Key figures from the Government Accountability Office (GAO) and the Defense Innovation Unit (DIU) presented their respective frameworks designed to translate ethical principles into practical guidelines.

Taka Ariga, GAO’s chief data scientist, outlined the agency’s internal AI accountability framework, emphasizing an auditor-centric approach to verification. This framework aims to bridge the gap between abstract ethical aspirations and the tangible realities of AI implementation across its lifecycle – design, development, deployment, and monitoring. The framework’s pillars are Governance, Data, Monitoring, and Performance, each subjected to thorough evaluation.

Governance assesses the organization’s oversight structure. Data examines the training data’s representativeness and functionality. Performance focuses on the AI’s societal impact, including potential civil rights infringements. Ariga stressed continuous monitoring to address model drift and algorithm fragility.

Bryce Goodman, DIU’s chief strategist for AI and machine learning, detailed their approach to embedding ethical principles into actionable engineering guidelines. DIU evaluates project proposals against ethics standards: Responsibility, Equitability, Traceability, Reliability, and Governability. Proposals failing this test are rejected. Goodman highlighted the necessity of collaboration and transparency, particularly with commercial partners, to ensure responsible AI development.

The DIU’s guidelines offer a series of pre-development questions encompassing task definition, benchmark establishment, data ownership assessment, responsible stakeholder identification, and rollback process creation. Goodman emphasized the importance of metrics beyond just accuracy, especially in high-risk applications that should use low-risk technology. AI, according to Goodman, should be deployed only when necessary and advantageous. The DIU plans to release its comprehensive guidelines, illustrative case studies, and supporting resources on its website.

Photo by Big Bag Films on Pexels
Photos provided by Pexels