TrailBlazer Decision Maker (TBDM) ALIGNABLE AND EFFICIENT ALGORITHMIC DECISION-MAKERS

Developing fully autonomous, human-off-the-loop, yet trusted decision-makers is challenging. Under difficult scenarios there is often no right answer, yet human decision-makers can navigate this complexity to make critical life-and-death decisions. Algorithmic decision-makers must then consider relevant background knowledge while capturing the goals, intents, and values of human decision-makers. Kitware proposes TrailBlazer Decision-Maker (TBDM), an approach to create alignable and efficient algorithmic decision-makers, based on our Phase I approach. TBDM will advance the state-of-the-art in AI, physiology simulation, knowledge graphs, and large language models to develop technology demonstrated for the medical triage domain. Our work will tackle significant research questions in a variety of areas including computational physiology modeling, document analysis, algorithmic decision-making, and model optimization. The Pulse Physiology Engine will be an important part of this effort and will be enhanced to perform scenario generation and simulation at scale, adding treatments and outcomes to train algorithmic decision-makers for medical triage. Model optimization will address limitations of the latest AI, where large models are not suitable to be run efficiently on smaller devices in the field with limited or no connectivity to the cloud. Our approach will enable algorithmic decision-makers to be easily adapted to different domains and scenarios.