Congressional Watchdog Finds Gaps in US Federal AI Efforts

Major government agencies in the United States intend to apply artificial intelligence, but the majority of planned use cases are still at the planning stage, a congressional watchdog said. Missing from those efforts is governmentwide guidance on the acquisition and use of AI technologies called for in a 2020 law.

See Also: Entering the Era of Generative AI-Enabled Security

The Government Accountability Office surveyed 23 agencies, including NASA and the Department of Homeland Security, about their AI plans and found more than 1,200 current and future AI use cases.

The Department of Homeland Security said its backpack unit intends to use machine learning to analyze camera and radar data in order to “identify border activities of interest.” Use cases by other agencies include analyzing photographs from drones to count wildlife, tracking wildfires and targeting scientific specimens for planetary rovers. Twenty of the agencies reported a total of 1,241 use cases. They are currently using AI in 228 ways, and 516 use cases are still in the planning stage. The majority of use cases are for scientific purposes, and about one-fifth are for internal management.

NASA and the Department of Commerce reported the greatest number of AI use cases. Three agencies – the Department of Housing and Urban Development, the Nuclear Regulatory Commission, and the Small Business Administration – did not report having any use cases for AI.

The White House across two presidential administrations has issued a slew of executive orders aiming to boost and guide federal embrace of the technology. An executive order issued during the final weeks of the Trump administration encourages agencies to adopt AI and makes it national policy to use the technology “where appropriate” to improve government operations and services. Congress in 2020 approved the AI in Government Act, which directs the Office of Management and Budget to publish a memo for the acquisition of AI, including best practices for mitigating any discriminatory affects or unintended consequences.

“Until OMB issues the required guidance, federal agencies will likely develop inconsistent policies on their use of AI, which will not align with key practices or be beneficial to the welfare and security of the American public,” the GAO wrote.

OMB on Nov. 1 released a draft memo and solicited public comment through Dec. 5. In a response to the GAO, OMB Deputy Director Jason Miller said the agency is “actively working to fulfill its statutory obligations.”

The GAO report also showed that agencies declined to identify specifics of about 30% of the active use cases, as they were “considered sensitive.” The State Department said it was using AI in 71 different ways but only publicly disclosed 10 of them.

The watchdog made 35 recommendations in the report, including asking the agencies to fully comply with AI reporting and implementation requirements.