Abstract

We will perform a secondary analysis of data from a study that investigated the impact of extended duration close quarters confinements as well as acute sleep loss and communications delays on human performance outcomes. These secondary analyses will attempt to use individual difference, manipulation, and intervention data to predict outcomes specified by DARPA sponsors. The outcomes to be predicted will be held out until after predictions are made. Our primary approach for making predictions under Attune is to build high-fidelity, agent-based models of individuals’ cognitive and social performance based on the Apprentice Learning Architecture (ALA) that can then be employed to simulate the impact of actual, hypothetical, or counterfactual HPO interventions on target individuals. Rather than taking an abstract, statistical approach, Attune models will be causal and computational in nature; simulated Attune agents will actually do/experience the tasks and interventions and make predictions based on their computational implementations of the underlying cognitive and social mechanisms that come into play. This theory-driven approach provides a unique capability that purely data-driven approaches cannot; mainly, Attune models will be able to generate hypothetical and counterfactual predictions of how particular HPO interventions will impact performance, even when previous data is not available for the target tasks. While these models can make predictions in the absence of any data, the Attune effort will explore how data about specific individuals and teams (ie, individual residuals) can be leveraged to make improved predictions. In …

Date

February 10, 2021

Authors

Christopher MacLellan, Kimberly Stowers, Adam Russell, Jeremy Gottlieb, Sonja Schmer-Galunder, Noshir Contractor, Gretchen Knaack, Ion Juvina, Brandon Minnery, Srikanth Nadella, Kristina Lerman, Keith Burghardt, Ryan Wohleber

Publisher

OSF