The rapid acceleration of artificial intelligence growth and the opportunities it presents to warfighters and civilian staff could “outpace” the Defense Department developing “appropriate and sufficiently broad guidance” for AI acquisitions, the Government Accountability Office warned in a report to the Senate Armed Services Committee.
“It is especially important that DOD and the military services issue guidance to provide critical oversight, resources, and provisions for acquiring AI given that the U.S. will face AI-enabled adversaries in the future,” GAO said. “Without such guidance, DOD is at risk of expending funds on AI technologies that do not consistently address the unique challenges associated with AI and are not tailored to each service’s specific needs.”
In February 2022, GAO noted that DoD recognized the differences in developing and deploying artificial intelligence capabilities compared to software — something the department “has historically faced challenges in developing, procuring, and deploying.”
“Data and computing power, among other things, are especially important for AI compared to traditional software,” GAO said. “…Given the need for data, training, and testing for operational usefulness, it may take longer to get to a minimally viable product for AI than for traditional software Agile projects. Once deployed, the AI capability should continue to get better at its task as it is exposed to more data, but needs to be continuously monitored to ensure the new data are not negatively affecting performance.”
In January 2020, the department released the Adaptive Acquisition Framework to help address the ways in which DoD’s traditional hardware and software acquisition wasn’t well-suited to acquire and deploy AI technologies. The following year, the new position of the Chief Digital and AI Officer was established to oversee the department’s path forward on artificial intelligence and oversee the AI marketplace known as Tradewind, which is intended to quickly award Federal Acquisition Regulation-based contracts and other transaction agreements.
DoD has also released various AI strategies, including the AI Education Strategy released by the Joint AI Center and the Ethical Principles for AI guidance released in 2020 followed by the Responsible Artificial Intelligence and Implementation Pathway released a year ago, which defined guiding principles for the process of creating and using AI for combat and non-combat purposes.
GAO recommended in March 2022 that DoD take steps toward more effective AI collaboration to avoid “a fragmented approach that could lead to unnecessary duplication and overlap in the future” including issuing guidance “that defines outcomes and monitors accountability for AI-related activities, and includes AI key performance indicators.” DoD has not yet completed the seven recommendations.
“Although numerous DOD components are acquiring, developing, or already using AI, DoD has not issued comprehensive department-wide guidance for AI acquisitions,” said the GAO report. “Similarly, the military services have not issued service-specific guidance that is tailored to their individual needs. This lack of guidance could result in acquisition of AI capabilities across the services that does not consistently address the unique challenges associated with AI or the specific needs of the acquiring service.”
“To fill this gap, various military services and other DOD components have individually developed or plan to develop their own informal AI acquisition resources,” GAO continued. “Some of these DOD component resources reflect key factors also identified by the selected companies for AI acquisitions.”
GAO said it interviewed “many officials” from DoD components who stated that “authoritative AI acquisition guidance would be helpful to navigate the AI acquisition process.” But an official from the Office of the Under Secretary of Defense for Acquisition and Sustainment said that the establishment of the Chief Digital and AI Officer delayed that guidance as that new office works out a path forward.
The Office of the Chief Digital and AI Officer is “in the process of planning to develop AI acquisition guidance and standing up an acquisition policy office to do so, but has not yet defined concrete plans,” the report notes, adding that an official from that office gave no timeline for this effort.
“Until DOD establishes department-wide guidance to act as a framework, it cannot ensure that its components are acquiring AI capabilities in a manner that accounts for the unique challenges and considerations associated with AI as they navigate the acquisition process,” GAO said. “Additionally, given that DOD is investing considerable effort and funds toward developing and acquiring AI tools and capabilities and requested $1.1 billion for core AI in fiscal year 2023, establishing guidance would better position DOD to effectively spend funds on AI acquisitions consistently across its components.”
While AI guidance has not yet been issued by the Pentagon, the service branches have been finding ways to help guide their AI programs.
“A Navy official told us that the Chief Digital and AI Officer has been helpful in promoting engagement with AI capabilities, but that they do not depend on support from this office. Marine Corps officials stated that DoD guidance would be helpful for the AI acquisition process and explained that having headquarters-level AI acquisition guidance standards is critical to get ahead of this new technology. In the absence of such guidance, these officials told us that they look to other military departments and their work, as they benefit from others’ lessons learned,” the report says. “According to Air Force officials, because of these gaps in guidance, the Air Force-Massachusetts Institute of Technology AI Accelerator developed its own guidebook with subject matter experts to compile lessons learned. Space Force officials told us that they leverage Air Force resources and do not have their own guidance or resources.”
Service-specific complementary guidelines reflecting the unique needs of military sectors will be critical to address in a department-wide AI-acquisition strategy, GAO said. “Guidance tailored to the types of capabilities each service acquires could include, for example, a decision-support AI capability to help operators plan and execute Navy undersea warfare missions, or a target-detection capability using satellite imagery,” the report noted.
Some DoD components have developed “informal AI acquisition-related resources” that “do not constitute guidance that establishes component-wide policy,” GAO added, and the Office of the Chief Digital and AI Officer is building a catalog of acquisition services for components’ use. The Defense Innovation Unit also created several AI worksheets that “walk developers through questions to help them align their AI projects with DOD’s ethical AI principles.” Several DoD components are also in the process of developing additional AI resources.
“Officials from the Intellectual Property Cadre shared their plans to publish an Intellectual Property Guidebook that will be applicable, but not specific, to AI acquisitions,” the report says. “Defense Acquisition University officials told us that they plan to provide commercial AI training to their own acquisition workforce, including plans to develop several AI-focused courses on AI reliability and system safety. Air Force officials told us they plan to add a course for acquisition personnel on building and designing AI applications, as well as a class for field grade officers and senior noncommissioned officers. Officials from the Navy also shared their plans to create the Naval AI Accelerator, a training and education Center of Excellence.”
GAO made four recommendations: that the Chief Digital and AI Officer prioritize establishing department-wide AI acquisition guidance, followed by the Army, Navy, and Air Force issuing service-specific AI acquisition guidance “that includes oversight processes and clear goals.”
The Office of the Under Secretary of Defense for Acquisition and Sustainment told GAO that DoD concurs with the recommendations.
