AI Accountability Policy with Ellen P. Goodman

Episode Summary

Executive Summary: Ellen Goodman discussed NTIA’s 2024 AI Accountability Policy Report, emphasizing that AI accountability requires consequences, not just transparency, and that governance must address risks across the AI value chain. She argued existing law can help but is insufficient, so standards, audits, documentation, and government capacity are needed, especially as AI becomes both a distinct industry and a general-purpose infrastructure.

Main Topics: NTIA role and the AI Accountability Report (Priority: 5/5): Goodman explained her 18-month role as Senior Advisor for Algorithmic Justice at NTIA and how she led work on the March 2024 report, including interagency consultation and policy design around AI and algorithmic governance. What accountability means in AI (Priority: 5/5): The report distinguishes accountability from adjacent concepts like transparency and explainability, arguing that real accountability requires consequences for harmful choices and rewards for beneficial ones. Who AI systems are accountable to (Priority: 5/5): Accountability extends beyond direct users to deployers, developers, affected non-users, and regulators, reflecting the complex AI value chain and overlapping responsibilities. Overlap with existing law and enforcement (Priority: 4/5): The discussion covered how FTC, CFPB, SEC, and other enforcers can apply existing law to AI harms, but need documentation, audits, and disclosures to detect and prove violations more effectively. Standards, audits, and the limits of standard-setting bodies (Priority: 5/5): Goodman supported standards for measurement and audit criteria but was skeptical that private standards bodies can resolve normative tradeoffs like fairness versus robustness without broader public judgment and support for civil society participation. Government use of AI and public-sector accountability (Priority: 4/5): She noted federal efforts to inventory and govern agency AI use, including the Biden executive order and OMB guidance, which seek to make government AI more transparent, audited, and risk-tiered. Copyright, provenance, and the future of AI-generated content (Priority: 4/5): Goodman said her current focus is content provenance, watermarking, and authentication, and she predicted copyright law may eventually need to adapt as AI-generated works become more seamless and harder to separate from human authorship.

Key Arguments: AI accountability should mean enforceable consequences, not merely transparency or explainability. AI harms often arise across a complex value chain, so developers, deployers, users, and regulators all need tailored accountability. Existing consumer-protection and enforcement regimes can address some AI misuse, but opacity makes detection and enforcement harder. Audits and disclosures are useful only if there are standards, criteria, and institutions capable of interpreting them. Private standards bodies can help with metrics and vocabulary, but not with resolving deeply normative tradeoffs. The federal government should use procurement, grants, and guidance to push trustworthy AI practices and build market demand for accountability. AI governance must also address new information-environment issues such as provenance, synthetic content, and authentication. Copyright law will likely need to evolve as AI-generated output becomes more integrated and less separable from human input.

Data Points: NTIA RFC questions: 34-36 - Goodman said NTIA’s request for comments asked roughly 34 to 36 questions to frame the accountability inquiry. Comment submissions: about 1,400 - She said the NTIA proceeding received roughly 1,400 comments. Institutional/organizational comments: about 250 - Of the comments, about 250 came from entities such as industry, academia, NGOs, trade associations, governments, and states. NTIA service duration: about 18 months - Goodman said she served at NTIA for about 18 months before ending in late March 2024. NTIA report release: March 2024 - The AI Accountability Policy Report discussed on the show was released in March 2024. Request for comments issued: April 2023 - NTIA’s AI accountability request for comments was initially issued in April 2023.

Pivotal Quotes: "accountability really only happens when there are consequences for choices" — Ellen Goodman: She defined the core theory behind the NTIA report, distinguishing accountability from softer governance concepts. "we're kind of, you know, building the plane while you're flying it" — Ellen Goodman: She described the speed and uncertainty of AI policymaking after the release of generative AI systems. "I think we need to question ... is AI an industry or is it just like electricity? It's just an infrastructure" — Ellen Goodman: She framed the regulatory challenge of deciding whether AI should be treated as a sector or as a cross-cutting foundational technology.

Implications: The conversation suggests AI governance will rely on layered oversight: law enforcement, standards, audits, and public-sector procurement. For industry, documentation and provenance will matter more; for policymakers, the challenge is building institutions fast enough to regulate a moving target.