Agencies have over 1,200 AI uses planned or in operation, watchdog finds

An ambitious new watchdog report examining artificial intelligence acquisition and use in the federal government looks to help Congress maintain a watchful eye on how the technology plays a role in federal operations.

Released by the Government Accountability Office on Tuesday after more than a year of work, the Artificial Intelligence report attempts to capture the confirmed planned use cases for AI systems in government agencies and ensure existing federal guidance is followed when implementing and leveraging AI solutions.

“Federal agencies have taken initial steps to comply with AI requirements in executive orders and federal law; however, more work remains to fully implement these,” the report reads. “Given the rapid growth in capabilities and widespread adoption of AI, the federal government must manage its use of AI in a responsible way to minimize risk, achieve intended outcomes, and avoid unintended consequences.”

A survey of 23 non-defense agencies uncovered 1,241 current and planned AI use cases at 20 different agencies, though only about 200 were currently in use, while 516 fell into the planned category. GAO’s strategy in gauging agency readiness was contingent on their assessment of such agencies’ adherence to existing federal law and guidelines, such as President Joe Biden’s recent executive order on AI, in addition to how agencies are inventorying their AI plans.

The results from the GAO’s audit, which spanned April 2022 to December 2023, were able to distill which agencies plan to be more aggressive AI adopters — like the National Aeronautics and Space Administration and the Department of Commerce, which far outstripped other agencies in the number of AI use cases at 390 and 285, respectively — but they also showcased areas of difficulty for federal agencies in streamlining AI acquisition and understanding.

The documented use cases are diverse. At NASA, one AI solution the agency adopted uses the technology to enable intelligent targeting and selection of scientific specimens to be collected by planetary rovers. Officials at Commerce, alternatively, are using AI to automate counting seabirds from drone photos.

Given the plethora of opportunities to employ AI software, having government offices come to a consensus on defining an AI system or technology is challenging. The report specifically noted that two federal statutes define AI differently, implying regulatory enforcement could be difficult.

“AI capabilities are rapidly evolving, but neither the scientific community nor industry agree on a common definition for these technologies. Even within the government, definitions vary,” the report says. 

Additionally, GAO also found that many of the categorical data agencies submitted to the Office of Management and Budget showcasing AI uses were incomplete. Of the 23 agencies surveyed, only five submitted comprehensive information on their reported use cases, while 15 had instances of incomplete or inaccurate data. Issues in the data ranged from incomplete fields, to duplications, to inaccurate listings. Two provided inventories were later determined to not actually be AI after submission to GAO.

“Agencies provided inventories that included missing elements or inaccurate data,” the report said. “Maintaining comprehensive and accurate AI use case inventories with quality information is critical for the government to have awareness of its AI capabilities and for agency leaders to make important decisions.”

Though GAO noted that agencies like the Department of Commerce and General Services Administration had fully implemented their selected AI requirements, 12 of those reviewed had tackled some, but not all, of what was required.

For example, the White House Office of Science and Technology Policy had not communicated its “designation of which agencies were expected to fulfill specific AI requirements;” the Office of Management and Budget had not yet developed guidance on the implementation and use of the technology; and the Office of Personnel Management had not yet established an occupational category for feds working on AI.

The report recommends that OSTP, OMB and OPM implement those AI requirements that have implications for the entire government, while encouraging 12 of the reviewed agencies to fully implement the requirements that are specific to them. GAO also recommended that the 15 agencies with incomplete AI use case inventories prioritize updating and maintaining that data.

Though a majority of agencies either agreed, partially agreed or expressed no determination either way on the recommendations made to them, OSTP and OMB disagreed with some of the recommendations, due to the fact that the White House’s AI executive order had been released after GAO completed its draft report, therefore superseding previous guidance on what was required of them.