COSMO: Human-Guided Machine Learning for Cognitive State Modeling

The proliferation of Army systems enabled with artificial intelligence (AI) has enhanced Warfighter capabilities at strategic, operational, and tactical levels. One limitation of these systems is that they lack awareness of the Warfighter’s cognitive state, which can lead to inappropriate interventions that negatively impact overall human-AI collaboration. Adaptive automation is one approach towards more context-sensitive interventions—however, some insight into the human’s state is still needed. Recent advances in sensor technologies and modeling techniques have created new opportunities in this area, yet more research is needed to mature these technologies and techniques so they can be used in operational settings. To meet this need, Aptima will develop COSMO: Human-Guided Machine Learning for Cognitive State Modeling. COSMO will be a diagnostic toolset for cost-effectively assessing human performance and predicting human cognitive states while integrating those insights into adaptive AI systems. Advanced, precise physiological sensors (e.g., fNIRS) will be used to create models in a lab environment which can then be used to develop and validate similar cognitive state models using less-intrusive sensors (e.g., smartwatch-based sensors). COSMO will accurately detect cognitive state changes in real-time and provide contextually relevant information that allows AI systems to adapt and intervene appropriately.