Innovating and prototyping an AI-powered chatbot to track governance policies in the Space Force: Wingman

Abstract

Employees in the Space Force can spend up to 50 man hours a week maintaining governance, which is an extreme time burden on this aspect of their job. The scientists, composed of a functional and technical team, worked to address this issue by innovating a
prototype of a new system to track governance. Interviews with beneficiaries in the federal defense industry revealed more insight into the issue, highlighting the overwhelming amount of searchable governance without an automated system to retrieve the most relevant data. Using this information, the teams innovated a prototype of an AI-powered chatbot, Wingman, which successfully responds to queries in the form of a checklist extracting information from over 100 governance PDFs from the Air Force E-publishing Website and other external terminology. We find that the use of Natural Language Processing (NLP), which uses machine learning (ML) to rank and tokenize the extracted terms, in conjunction with an open AI key, GPT 3.5- turbo, to access and generate a text response from the data, helps efficiently code this prototype, allowing it to run successfully. While the team found general success in the ability of the chatbot, the team needs more time to assess its full accuracy in ranking and sorting through the PDF terms. Future optimization includes improving the source for more practicality as many documents in the federal defense space are highly classified. Overall, the scientists were able to successfully plan, innovate, and code their chatbot Wingman to improve the current methods of maintaining governance, demonstrating the potential of AI and ML to improve various time-burdening processes.