Podcast: Real-World AI Use Cases Emerge in DoD
Experts on the *AI in Defense* podcast are discussing the first real-world test cases of generative AI within the Department of Defense. One guest noted, "We’re seeing the first real test cases—language models augmenting intelligence analysis, but only in tightly bounded, human-in-the-loop workflows." This indicates a move from high-level policy to practical, albeit cautious, implementation.
- The Chief Digital and Artificial Intelligence Office (CDAO) is leading the DoD's generative AI adoption, establishing Task Force Lima in August 2023 to assess and integrate tools like large language models. This task force has since been succeeded by the AI Rapid Capabilities Cell (AI RCC) to accelerate the delivery of AI capabilities with an initial investment of $100 million. - Current unclassified applications of generative AI include summarizing policy documents, drafting procurement materials, and creating internal reports. The Pentagon has partnered with OpenAI to integrate ChatGPT, alongside Google's Gemini, into its GenAI.mil enterprise platform to support these everyday tasks for its personnel. - The DoD's "Responsible AI Strategy and Implementation Pathway" guides the ethical development and use of AI, based on five principles: Responsible, Equitable, Traceable, Reliable, and Governable. The Defense Innovation Unit (DIU) has created practical worksheets to help vendors and program managers align with these principles throughout the project lifecycle. - To address the challenge of evaluating large language models, the CDAO has contracted with Scale AI to create a comprehensive test and evaluation framework to measure model performance and ensure readiness for military applications. - Project Maven, an early and significant DoD AI initiative, uses machine learning to analyze full-motion video from surveillance assets. Having demonstrated success in speeding up data analysis, it has been used to support operations against ISIS and in Ukraine. - For companies looking to partner with the DoD, the Army's SBIR and STTR programs have issued funding opportunities for AI/ML projects, such as the "Context-Aware Decision Support" solicitation aimed at enhancing commanders' decision-making with generative AI. - The DoD is exploring a wide range of future AI applications, including optimizing logistics, drafting military doctrine, coordinating autonomous drone swarms, and creating synthetic training data. - A key challenge in DoD AI adoption is the need for specialized data, as foundational models are not trained on military-specific manuals and tactics. Another significant hurdle is ensuring access to sufficient computing power to scale pilot programs across the department.
