Agencies should adopt consistent, machine-readable formats (such as CSV) for reporting AI procurement data. Standardized reporting of AI use cases across agencies would facilitate cross-departmental comparisons and enhance transparency. The 2024 Guidance for Agency Artificial Intelligence Reporting per EO 14110 outlines the required formats and criteria to ensure all agencies follow a structured reporting approach. However, requirements for disclosing vendor names remain insufficient. In general, while reporting on procurement data was previously time-consuming and resource-intensive, leveraging large language models (LLMs) now allows for faster, more detailed, and comprehensive data reporting.
Introducing a dedicated AI category in procurement systems, such as within the North American Industry Classification System (NAICS) and the Product and Service Codes (PSC) system, would enable more precise tracking of AI technologies. NAICS, defined by the U.S. Office of Management and Budget (OMB) in collaboration with Canada and Mexico, is reviewed every five years to reflect emerging technologies and economic trends. The next update is scheduled for 2027, with feedback opportunities in 2025.
The PSC system, maintained by the General Services Administration (GSA) and the Defense Logistics Agency (DLA), includes over 2,000 codes covering a wide range of goods and services procured by the federal government. Like NAICS, PSC currently lacks a specific code for AI or machine learning, with related procurements often categorized under broader codes like D399 (Other IT Services) or R425 (Engineering and Technical Services). To improve tracking and reporting, GSA and DLA should consider adding distinct codes for AI and machine learning in federal procurement.
In the absence of specific federal regulations on AI, the public should advocate for agencies to report on relevant regulations and standards, highlighting their importance in AI procurement. Clearly linking AI procurement practices to existing standards – such as the Clinger-Cohen Act for IT governance, as well as labor and environmental responsibility standards – can promote sustainable and accountable AI procurement.
Implementing Performance-Based Acquisition (PBA) in AI procurement ensures that vendors are not only accountable for delivering technical capabilities but are also responsible for meeting ethical standards in areas like fairness, bias mitigation, transparency, and user privacy. PBA involves defining clear, measurable outcomes that vendors must achieve, rather than simply prescribing the processes they should follow. For instance, performance-based criteria could include specific benchmarks for bias testing, explainability, and adherence to non-discriminatory practices in AI models. By setting these standards upfront and requiring detailed contract descriptions, agencies can systematically track vendor performance against ethical metrics.
Extended Analysis for Future Fiscal Years: Applying this index to 2024-2025 data will allow agencies to track changes in AI procurement practices over time, assessing the impact of responsible AI recommendations and guidelines. As administration changes could alter responsible AI reporting requirements – particularly given the current reliance on executive orders from the Biden administration – it's essential to leverage independent data platforms like USAspending.org, the Federal Procurement Data System, and FOIA portals. These tools provide valuable, administration-independent resources to support continued demands for responsible and transparent AI procurement.
Increased Vendor Scrutiny: Future work should emphasize the evaluation of "off-the-shelf" AI solutions, rigorously examining these products for compliance with responsible AI standards. This includes assessing vendors' accountability practices and ensuring their adherence to ethical considerations in AI deployment.
Web-Based Platform Development: The project’s web-based platform, currently in development, will leverage LLM capabilities to enhance procurement data and contract analysis to make the process less time consuming.
Department of Agriculture (USDA)
Department of Health and Human Services (HHS)
Department of Homeland Security (DHS)
Department of Housing and Urban Development (HUD)
Department of the Interior (DOI)
Department of Transportation (DOT)