Task, Workflow and System-level Opportunities for AI-based Tools to Support Defense Enterprise Efficiency

Artificial Intelligence Enabled Systems (AIES) are a key enabler of both future warfighting capabilities and efficient business practices. Since the launch of ChatGPT, significant attention has been focused on the potential of GenAI-based tools to replace “white collar” work, while drastically increasing the efficiency of the lifecycle of acquisition, engineering, manufacturing and operations & sustainment. At the same time, there are serious concerns that the potential of current technology is being over-hyped and that we are not prepared to ensure that when deployed, AI-enabled systems will work as needed, in accordance with ethical principles and without introducing new sources of vulnerability.

This paper leverages insights from a portfolio of Systems Engineering Research Center (SERC) and Acquisition Innovation Research Center (AIRC) research to propose a framework of opportunities and risks for AI adoption to support engineering and acquisition work across the Defense Enterprise. Individual projects sought to developed and evaluate tools to summarize vast corpuses of text and/or answer questions about it, cross-check facts, generate domain-specific content etc. These ideas were applied in functional areas ranging from cost estimation, to contracting, to design, to test and evaluation.

Across the projects we observed patterns of opportunities and risks at multiple levels. At the task-level existing AI tools were effective at identifying patterns, translating, templating artifacts and, to some extent, synthesizing. Reliability concerns were effectively mitigated via workflow integration. At the individual workflow-level we observed that most human-defined tasks include a mix of subtasks that play to both AI strengths and weaknesses. Therefore, effective task allocation must be paired with a rebundling of existing subtasks. At the work system-level, this rebundling can be even more effective. It is important to recognize that traditional organization design was constrained by decision-maker attention, which AI’s ability to scale fundamentally changes. AI support can be cross-cutting and layered, enabling humans to pay more attention to critical synthesis and prioritization tasks. All this speaks to a need for a systems perspective on AI integration in work. Finally, these new tasks and new modes of work result in a commensurate need for upskilling, something that LLM-based tools can support as well.