Abstract for: Using System Dynamics to Introduce Artificial Intelligence (AI) to the Operational Environment: Lessons from government-funded

The introduction of artificial intelligence into operational systems has been slowed by its limited capability to address the dynamic complexity of the operational environment. Demonstrations against well-defined problems are impressive, but transitioning from the laboratory to the operational environment remains an ongoing opportunity. This presentation discusses the use of system dynamics (SD) across multiple government-funded research programs to quantify and specify complex problem domains. Often understanding the key problems within an operational domain and its potential contributions to humans working within it is a separate task from engineering the system, one that is often addressed inadequately by engineers. The talk will focus on how SD has been applied in several government-funded research programs both as a core component and as a way to identify requirements for additional AI capabilities. The analysis is broken out into data models and complex models, with SD models acting as complex models identifying data requirements and identifying the complex causal relationships to be modeled. The talk will conclude by identifying discussing multiple research thrusts included the SD reference mode for complex problem definition—e.g., wicked problems--with operational system factors such as trust, verification, validation, and accreditation (VV&A), generalizability (& extensibility, maintainability), templates (& modules, patterns) data availability, and human machine teaming additionally addressed. Fuzzy search, Monte Carlo Tree Search, RL LLMs