Predicting Force Protection Threats against DTRA from Foreign Adversarial Messaging

OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Trusted AI and Autonomy OBJECTIVE: Develop predictive models to determine when online messaging is going to breakout into a different part of the online, foreign information environment or into the physical world as activities directed against DTRA personnel or facilities. DESCRIPTION: Foreign adversaries have employed online messaging to target and undermine DTRA's programs, personnel, or overall reputation. A capability is needed to predict when such messaging may break out of its originating source(s) in the foreign information environment (news sources, social media, and websites) and move to other parts of the foreign information environment, where it may lead to the threatening or incitement of physical action against DTRA personnel or facilities. The output of predictive models can help DTRA posture from force protection and public affairs perspective. PHASE I: PHASE I Proof of Concept Conduct and document a review of prior work to predict physical threats from online messaging (e.g., prior work in counterterrorism, brand management, etc. [1]). Develop proof-of-concept model(s) that use publicly available information (PAI) for a country (or region) in which DTRA operates to predict transitions between parts of the foreign information environment and which of those transitions may lead to threatening physical activity. Note the focus here is not on predicting which individuals or groups that may engage in these behaviors but rather gauging the overall threat level to DTRA based on the content and spread of the messaging. Validate models using historical case studies and subject matter experts. Model will be scoped to specific foreign populations, specific parts or topics of the information environment, and/or a small set of potential threats. PHASE II: Phase II - Prototype Development: Objective: Refine model(s) to include explanatory capabilities and demonstrate generality of the models to other foreign populations, specific parts or topics of the information environment, and potential threats. Assess model accuracy against ground truth data and improve model transparency, explainability, and portability to different applications. PHASE III DUAL USE APPLICATIONS: Apply prototype models to a broader variety of online information environments and media types to predict a broader range of threats. Dual use applications for industrial security are envisioned. REFERENCES: 1. Renshaw, S. L., & Carley, K. M. (2024). Linking online activity to offline behavior: A meta-review of three decades of online-to-offline scholarship with future implications for AI. Emerging Trends in Drugs, Addictions, and Health, 4, 100154. https://doi.org/10.1016/j.etdah.2024.100154 Access, 9, 87663 87685. KEYWORDS: predictive analytics, information environment, force protection