Network centric military systems (NCW) are planned to involve hundreds to thousands of manned and autonomous entities cooperating to achieve complex joint objectives in incomplete information environments. The introduction of pervasive networking and command architectures offers both exciting new opportunities and the possibility of unintended consequences or unanticipated changes to human roles. Therefore, there is a crucial need to be able to understand, model and predict the behavior of these large human-machine systems. The proposed research explores these issues before they confront us in the field using mathematical models, simulation, and human data to investigate the behavior of large networked human-machine systems. The overall research goal of this multidisciplinary research is to provide validated theories grounded in experiments with humans that allow descriptive and predictive characterization of complex human-machine systems. In particular, we aim to

• (a) create, test and evaluate models of human decision making embedded in large scale complex, dynamic environments
• (b) characterize the interactions of the human and machine system components under different assumptions of environment, system scalability, task scenarios and interaction complexity
• (c) understand and predict aggregate system behavior
• (d) analyze trade-offs between computational tractability and modeling fidelity.

To address this challenge, we adopt a methodology of model building and model characterization based on incremental abstraction, in particular abstraction of cognitive models. The overall process will have 4 interactive levels:

• the first level is comprised of humans interacting among themselves and with automation in different tasks relevant to NCW;
• the second level will substitute high fidelity models for some of the humans;
• the third layer will have abstracted cognitive models interacting with each other and the automation;
• the fourth level will have agents represented simply as a set of possible states interacting at a very large scale.

Each higher level of increased abstraction has increased number of entities. Each level has its own algorithms for characterizing behavior, capable of performing different types of analysis and prediction, at scales ranging from a few nodes to many thousands of nodes. Model characteristics at each level generate a set of hypotheses for the level above. Similarly, results from aggregate behavior at a more abstract level can be validated at a lower level. This approach will allow us to incrementally characterize complex hybrid behavior and provide a “scaffolding for validation”. We will deliver a set of validated cognitive models at different levels of abstraction for different NCW scenarios, a set of mathematically based models that can be used to analyze and predict aggregate system properties, e.g. vulnerability, a set of validated simulations and extendable tools that enable further modeling and study. By identifying potential bottlenecks, challenges to stability, and obstacles to human control we hope to identify potential solutions before problems are built into procured systems.