acterized, will behave in dynamic environments (Balleine et al., 2009).

Similarly, tremendous advances have been made in understanding movement generation, including development of skills, habits, and automatic performance (Poldrack et al., 2005; Yin and Knowlton, 2006). Moreover, consider an individual who is learning to execute a complex movement accurately—for example, tracking a moving target with his finger (Poulton, 1974). We understand quite precisely how the incentives and feedback provided shape both performance and learning (Newell, 1996; Trommershäuser et al., 2009). We also have evidence that the fastest way to train a movement is not to provide strictly accurate feedback. The speed and effectiveness of such training can often be maximized by providing feedback regimens that take advantage of this inherent accommodation to variability (Schmidt and Lee, 2005; Kording et al., 2007).

In short, advanced models of reinforcement learning and movement control learning have implications for both training and prediction of learning efficiency. We know in principle how to develop optimal training regimes under many conditions, and we know how to predict agent behavior with precision under a range of conditions. Both assets could be leveraged to improve not only training but also data presentation (for situational awareness) and prediction of threat/enemy behavior.

The combination of neuroimaging tools, cognitive neuroscience, and experimental or cognitive psychology has resulted in the development of models of how the brain may process information. For example, recent accounts of brain processing that occurs in the dorsolateral prefrontal cortex (PFC) are based on interpreting cognitive control as altering ongoing behaviors in order to adjust to the changing context of the environment (Botvinick et al., 2001). The resulting computational models reveal which aspects of cognitive performance are altered as information changes—a field of study called “computational neuroscience”—and can be used to predict performance on behavioral tasks (Brown and Braver, 2007).

Computational neuroscience uses mathematical models to study how neural systems represent and transmit information. The discipline may be roughly divided into two schools. The first school uses detailed biophysical models of individual neurons, detailed models of neuronal networks, and artificial neural network models to study emergent behaviors of neural systems. The second school develops signal-processing algorithms, computational models, to analyze the growing volumes of data collected in neuroscience experiments. In these computational models, adaptation to new information is represented as changes in one or more parameters. By combining the models of both schools with information from neuroimaging tools (sometimes called “systems neuroscience”) and behavioral neuroscience, the possible causes of underperformance and the conditions conducive to improved performance can be quantitatively constrained. Eventually, it may be possible to develop specialized interventions aimed at improving performance in soldiers.

Subject Populations for Army-Specific Studies of Learning and Training

A perennial issue for behavioral and neurological testing is the degree to which experimental findings from a specific sample can be extrapolated to a target population. Many of the activities in which soldiers and their leaders engage depend critically on rigorous preparatory training and task-specific expertise. The majority of behavioral research is performed with subjects who are either patients in a clinical setting or volunteers from a university community (mostly undergraduate students). A critical question is how far results based on these study populations of convenience transfer to a soldier population. In cases where the research hypothesis addresses Army-relevant issues directly, the typical research subject populations may not be sufficiently representative of the population to which the Army wants the results to apply. In short, neither clinical patients nor university undergraduates are good surrogates for a soldier.

As the Army seeks to apply research results from the neuroscientific literature, the extent to which they can be transferred to a military (specifically, Army) population should itself be a subject of Army research. In particular, how far do results from typical civilian samples represent those to be expected from an Army population? Although some of the research reviewed by the committee has used actual soldiers or cadets, for the most part the human subjects in potentially relevant studies do not compare well to the soldier population in cardiovascular fitness, psychological drive to perform, and learning/training experiences that clearly affect neurobehavioral response—e.g., boot camp, intense training for operational performance, and actual operations.

One alternative to constraining Army-usable results to just the few studies that use soldiers (or even military cadets) is to seek subject populations that more closely resemble Army soldiers in such key characteristics as cardiovascular fitness, psychological motivation to perform, and training/learning in immersive, demanding environments. High-performance athletes are one such subpopulation, and there is an extensive literature of behavioral and neuropsychological research on them. Appendix C lists a sampling of the research literature from 2001 through 2007 on training methods for high-performance athletes: performance evaluation/assessment of athletes in training, including under stress; social interaction with other athletes; and issues with performance anxiety and other psychological issues including depression in ex-athletes (references 1-109 in Appendix C). Several studies have investigated the use of mental imagery in training athletes and its effects on performance (references 110-123). Performance after mild concussions and determining when the subject can return to a normal (strenuous) routine is a hot topic for both athletes



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement