Anthropic has published a policy proposal urging the United States Congress to provide ambitious funding to the National Institute of Standards and Technology (NIST). The proposal emphasizes the importance of investing in NIST to build AI measurement tools, standards, and testbeds. This initiative aims to enable the accurate description and quantification of AI system capabilities and risks, which Anthropic views as a prerequisite for effective regulation.
Anthropic's recommendation follows bipartisan support for maintaining American leadership in critical technologies, expressed during an April 18 budget hearing on the 2024 Request for the Department of Commerce.
Key Points
- Anthropic proposes ambitious funding for NIST to develop AI measurement tools, standards, and testbeds.
- The ability to accurately describe and quantify AI capabilities and risks is considered essential for effective regulation.
- NIST has a history of work in AI measurement, including the Face Recognition Vendor Test, the AI Risk Management Framework, and the MNIST dataset.
- Anthropic notes a general under-resourcing and stagnation in funding for AI-related programs at NIST in recent years.
- Increased funding would allow NIST to build on fundamental measurement techniques, standardize them, and create community resources like testbeds.
- The proposal is presented as a "shovel ready" idea that complements other policy levers for stronger AI governance.
Context
According to Anthropic, rapid technological progress in AI and the rise of publicly deployed AI-powered products have highlighted both significant capabilities and risks. Some risks can emerge abruptly with larger-scale models or only after deployment. While various ideas for governing AI have been proposed, Anthropic believes increasing funding for NIST is a simpler, readily actionable approach.
Why It Matters
This proposal suggests that the ability to effectively evaluate AI systems for performance and risks is a necessary precursor to smart AI regulation. Builders and policymakers can benefit from standardized measurement infrastructure to inform regulation and guide safer AI development.
What To Do
- Note Anthropic's advocacy for increased NIST funding as Congress considers appropriations for FY 2024.
- Review the NIST AI Risk Management Framework to understand existing measurement efforts.
- Watch for further guidance or policy changes related to NIST's role in AI standardization and testing.