Military tacticians require practice to learn their craft. Practice requires adaptive opponents capable of responding to trainee actions in ways that are realistic and instructionally productive. Current agents are generally too brittle, too scripted and unresponsive to support adaptive training in this way. What is required to develop adaptive agents are (1) real-time feeds of simulator data that are sufficiently rich and realistic to support agent development and execution; (2) agent architectures capable of generating realistic and instructive behaviors from these data; and (3) a testbed that can deliver data and performance measures in sufficient volume to enable modelers to accelerate agent development by applying emerging analytics and machine learning. The 711th Human Performance Wing/RHA has invested in precisely these capabilities over four years, engaging eight of the leading developers of intelligent agents. In this paper, we describe these capabilities, and, importantly, the data requirements these solutions impose on simulators and operational systems that can employ these technologies in the future.
[1]
L. Vygotsky.
Mind in Society: The Development of Higher Psychological Processes: Harvard University Press
,
1978
.
[2]
Demis Hassabis,et al.
Mastering the game of Go without human knowledge
,
2017,
Nature.
[3]
Shane Legg,et al.
Human-level control through deep reinforcement learning
,
2015,
Nature.
[4]
Webb Stacy,et al.
Training objective packages: enhancing the effectiveness of experiential training
,
2016
.
[5]
A. Tversky,et al.
Judgment under Uncertainty: Heuristics and Biases
,
1974,
Science.
[6]
John E. Laird,et al.
The Soar Cognitive Architecture
,
2012
.
[7]
Mica R. Endsley,et al.
Toward a Theory of Situation Awareness in Dynamic Systems
,
1995,
Hum. Factors.