CHARLESTON, W.Va. — Yesterday, U.S. Senator Shelley Moore Capito (R-W.Va.) along with U.S. Senators John Thune (R-S.D.), Amy Klobuchar (D-Minn.), Roger Wicker (R-Miss.), John Hickenlooper (D-Colo.), and Ben Ray Luján (D-N.M.), all members of the Senate Committee on Commerce, Science, and Transportation, introduced the Artificial Intelligence (AI) Research, Innovation, and Accountability Act of 2023. The bipartisan legislation – led by Senators Thune and Klobuchar – establishes a framework to bolster innovation while bringing greater transparency, accountability, and security to the development and operation of the highest-impact applications of AI.
“I am glad to partner with my colleagues to introduce a bipartisan first step towards addressing the development of artificial intelligence (AI). Our bill will allow for transparent and commonsense accountability without stifling the development of machine learning. I look forward to working with my colleagues to advance this important legislation,” Senator Capito said.
Provisions included in the AI Research, Innovation, and Accountability Act of 2023 are below:
- Content Provenance and Emergence Detection Standards:To provide clearer distinctions between human and AI-generated content, the bill would require the National Institute of Standards and Technology (NIST) to carry out research to facilitate the development of standards for providing both authenticity and provenance information for online content, similar to the efforts of the Coalition for Content Provenance and Authenticity. The bill would also direct NIST to support standardization of methods for detecting and understanding emergent properties in AI systems in order to mitigate issues stemming from unanticipated behavior.
- AI Definitions: The bill would provide new definitions for “generative,” “high-impact,” and “critical-impact” AI systems. A clear distinction would also be made between “developer” and “deployer” of the AI system for purposes of requirements outlined below.
- Generative AI Transparency: To ensure consumers are properly notified when they are interacting with generative AI, the bill would require large internet platforms to provide notice to users when the platform is using generative AI to create content the user sees. The U.S. Department of Commerce would have the authority to enforce this requirement.
- NIST Recommendations to Agencies: NIST would be required to develop recommendations to agencies for technical, risk-based guardrails on “high-impact” AI systems, in consultation with other agencies and non-government stakeholders. The Office of Management and Budget would be tasked with interagency implementation of such recommendations.
- Risk Management Assessment and Reporting: Consistent with the structure of NIST’s AI Risk Management Framework, the bill would require companies deploying critical-impact AI to perform detailed risk assessments. These reports would provide a comprehensive, detailed outline of how the organizations manage, mitigate, and understand risk. Deployers of “high-impact” AI systems would be required to submit transparency reports to the Commerce Department.
- Critical-Impact AI Certification: The bill would require critical-impact AI systems to be subject to a certification framework, in which critical-impact AI organizations would self-certify compliance with standards prescribed by the Commerce Department. The structure of the certification process is outlined below:
- AI Certification Advisory Committee: The bill would establish an advisory committee, composed of industry stakeholders, to provide input and recommendations on the issuance of proposed critical-impact AI certification standards.
- Submission of Certification Plan: Before any standards for critical-impact AI could be prescribed, the Commerce Department would be required to submit to Congress and the advisory committee a five-year plan for testing and certifying critical-impact AI. The plan would outline not only proposed standards and a timeline for their implementation, but also a process for (1) evaluating whether standards are necessary and appropriate, including whether a prevailing international standard is sufficient and (2) how objective and transparent information will be gathered. The Commerce Department would be required to regularly update the plan and consult with other government agencies, the advisory committee, and non-governmental standards organizations in the process.
- Critical-Impact AI Standards: The Commerce Department, upon submitting a certification plan to Congress and the advisory committee, would have the authority to prescribe testing, evaluation, validation, and verification (TEVV) standards for critical-impact AI. Critical-impact AI deployers would self-certify compliance with the standards. All standards would have to be submitted for public comment.
- Exemptions: The Commerce Department would have the authority to issue exemptions from certain TEVV standards, as deemed appropriate, in response to exemption applications from critical-impact AI deployers.
- Noncompliance and Enforcement: Upon discovering noncompliance with an applicable TEVV standard, critical-impact AI organizations would be required to immediately notify the Commerce Department and provide a plan for addressing the noncompliant aspect of the AI system. In response, the Commerce Department could order additional actions to be taken by the organization.
- AI Consumer Education: The bill would require the Commerce Department to establish a working group to provide recommendations for the development of voluntary, industry-led consumer education efforts for AI systems.
# # #