Teaching Machines to Recognize Neurodynamic Correlates of Team and Team Member Uncertainty

Abstract

We describe efforts to make humans more transparent to machines by focusing on uncertainty, a concept with roots in neuronal populations that scales through social interactions. To be effective team partners, machines will need to learn why uncertainty happens, how it happens, how long it will last, and possible mitigations the machine can supply. Electroencephalography-derived measures of team neurodynamic organization were used to identify times of uncertainty in military, health care, and high school problem-solving teams. A set of neurodynamic sequences was assembled that differed in the magnitudes and durations of uncertainty with the goal of training machines to detect the onset of prolonged periods of high level uncertainty, that is, when a team might require support. Variations in uncertainty onset were identified by classifying the first 70 s of the exemplars using self-organizing maps (SOM), a machine architecture that develops a topology during training that separates closely related from desperate data. Clusters developed during training that distinguished patterns of no uncertainty, low-level and quickly resolved uncertainty, and prolonged high-level uncertainty, creating opportunities for neurodynamic-based systems that can interpret the ebbs and flows in team uncertainty and provide recommendations to the trainer or team in near real time when needed.