HORATIO: Hybrid Online Recognition of Adversary Tactics in Information Ops

Information operations (IO) operators attend to large volumes of data across regions, timescales, languages, platforms, and adversaries.  Complex AI and machine learning tools can detect indicators of influence but the operator -- who may be a novice in AI, ML, and social science -- is left to integrate results into a model of a complex, coordinated, distributed adversary.  SIFT proposes to develop HORATIO (Hybrid Online Recognition of Adversary Tactics in Information Ops) as a machine teammate for IO analysis.  HORATIO will feature a multimodal (diagram- and dialogue-based) interface for operators to specify IO objectives and workflows, and then operators will trigger workflows with a playbook-style human-machine interface.  It will use Hierarchical Task Networks (HTNs) to plan and manage these workflows, running a suite of natural language processing (NLP) tools to detect indicators of influence, and then automatically analyzing and accumulating results into a template-like model of red-force tactics, techniques, and procedures (TTP) to contextualize subsequent interactions and analyses with the operator.  This full-loop human-machine operation will bring high-value IO tools into the fold that were previously out-of-reach due to difficultly of operation or complexity of interpretation. To achieve this vision, SIFT will leverage ongoing work on large language models (LLMs), HTNs, NLP, and IO operations. SIFT's recent work fine-tuning large language models (LLMs) -- with low-rank adaptation and constrained decoding -- will help HORATIO read and generate structured data such as workflows, within a mission context.  SIFT is an industry leader in HTN planning and maintains leading open-source HTN planners, and HORATIO will utilize these tools for representing, running, and managing IO workflows.  Finally, SIFT is involved in the DARPA INCAS program, and has produced robust NLP tools for IO analysis -- for detecting topic-based sentiment, emotions, concerns, moral framing, themes, and more -- that will populate HORATIO's initial suite of NLP tools, and more tools are available open-source from other INCAS performers.  Finally, the HORATIO effort will include a consultant on IO narratives, transition partners (see letter of support), and briefings to government operators who are already using SIFT's INCAS technology, all which will prioritize the workflows and TTP that we implement for HORATIO.  SIFT is well-positioned with transition customers in the DoD to build HORATIO and deliver it to operators who need it.