Abstract
Objective
The purpose of this investigation was to elucidate the role of button-response complexity to targets in a response inhibition task.
Background
Response inhibition is the ability to correctly inhibit an overt response to a target. The U.S. military is actively pursuing development of armed, combat robots as a force multiplier, which may present challenges to operators of combat robots in the form of response inhibition errors.
Method
A total of 15 participants completed two 51-min versions of a modified sustained attention to response task (SART). Participants were outfitted with an electrocardiogram to index heart-rate variability and completed the NASA–Task Load Index (NASA-TLX) to index workload.
Results
The results demonstrated that the complex SART reduced errors of commission (4%) and slowed response times (874 ms) to correct Go targets relative to the simple SART (14%, 739 ms). The NASA-TLX did not show differences between the modified SARTs; however, heart-rate variability did demonstrate that Soldiers had an increased autonomic stress response to the complex SART.
Conclusion
Increasing the behavioral response requirement during a response inhibition task can decrease errors of commission; however, it comes at the cost of slower response times to target stimuli. Heart-rate variability may provide better insight into objective workload relative to subjective measures.
Application
The use of complex behavioral responses may provide a viable option to reduce potential “friendly fire” or collateral damage by Soldiers remotely engaging a target-rich environment.
Keywords
Introduction
The U.S. Army is actively exploring the use of combat robots (e.g., unmanned ground vehicles armed with infantry weapons) in conjunction with ground troops on the battlefield (Feickert, 2005). Indeed, a well-known security consultant recently predicted, “At some point around 2025 or thereabouts the US Army will actually have more combat robots than it will have human Soldiers” (Lockett, 2017). The application of combat robots presents numerous advantages such as force multiplying, faster target acquisition, and less exposure of friendly forces to lethal enemy contact (Bray-Miners, Ste-Croix, & Morton, 2012; Haas, David, & Haug, 1996). Thus, the role of selected Soldiers will shift from boots on the ground to remote supervisor of semiautonomous robots, whereby the Soldier is actively monitoring and manually responding to target engagements (Parasuraman, Barnes, & Cosenzo, 2007). For example, a combat robot equipped with advanced optics could actively acquire multiple targets and calculate optimized ballistic solutions, which would only require Soldiers to make quick “shoot” or “no-shoot” button responses.
The use of combat robots will likely increase lethality by improving accuracy and target acquisition time (Gunn et al., 2005); however, lethality may come at a cost of increased “friendly fire” and collateral damage (Helton & Kemp, 2011). Indeed, there is a myriad of contributing factors that may lead to friendly fire, for example, loss of situational awareness, terrain complexity, misidentification of targets, and poor visibility (Wilson, Head, & Helton, 2013). These aforementioned factors are likely to be compounded if the Soldier is not actually physically present on the battlefield but tele-present. Recently, it has been proposed that an additional contributing factor to friendly fire is response inhibition failures, which may be compounded as a function of the faster target-engagement tempo in modern combat (Wilson et al., 2018; Wilson et al., 2013).
In a high “Go/low–No-Go” task, response inhibition failure is the inability of an individual to inhibit a prepotent motor response to a predefined No-Go stimulus (Head & Helton, 2012). This response inhibition error has been investigated in high Go/low–No-Go target-detection tasks such as the sustained attention to response task (SART; Robertson, Manly, Andrade, Baddeley, & Yiend, 1997). The SART requires participants to make numerous overt button responses to Go stimuli while withholding responses to rare No-Go stimuli. Generally, simple numeric stimuli are used, whereby participants respond to the neutral distractors (e.g., numbers 1–9) but withhold their response to a predesignated target (e.g., number 3; Robertson et al., 1997). Performance on the SART is indexed by errors of omission (inappropriately withholding response to Go stimuli), errors of commission (inappropriately responding to No-Go stimuli), and response time to Go stimuli. Errors of omission are thought to be attributed to strategic forced rest breaks to enhance inhibitory control (Helton & Head, 2012). However, the most prominent finding of the SART is faster response times to Go stimuli at the cost of increased errors of commission (i.e., speed–accuracy trade-off [SATO]; Dang, Figueroa, & Helton, 2018; Ishimatsu, Meland, Hansen, Kåsin, & Wagstaff, 2016; Peebles & Bothell, 2004). The SATO of the SART has been argued to be caused by a prepotent, feed-forward ballistic motor routine as a result of a high ratio of Go responses relative to No-Go responses (Helton, Weil, Middlemiss, & Sawers, 2010). Generally, when participants are presented with numerous sequential Go stimuli (i.e., a target-rich environment), participants adopt a prepotent motor-response routine that is difficult to inhibit when a No-Go stimulus is presented. Thus, participants may be fully aware of the No-Go stimuli but are not able to physically stop themselves from making an overt response (Cheyne, Carriere, & Smilek, 2009; Head & Helton, 2013).
Inhibitory control processing has been investigated with a variety of other tasks besides the SART, including Stroop tasks (e.g., Friedman & Miyake, 2004) and the stop-signal paradigm (e.g., Logan, 1994). Understanding the impact of response inhibition failure is not purely academic but has bench-to-bedside application. For example, measuring response inhibition failure in clinical settings has provided a symptomatic approach to aid in diagnosing various psychopathological and neurological disorders (Bender, Filmer, Garner, Naughtin, & Dux, 2016). In addition, response inhibition failure has been associated with various disorders such as schizophrenia, autism, and attention deficit hyperactivity disorder (Verbruggen & Logan, 2008). Understanding the role of response inhibition errors may also show application in potential work environments with military and law enforcement. Indeed, ongoing research has begun to address the problematic nature of response inhibition errors in relatively more ecological scenarios such as marksmanship (Biggs, 2017; Biggs, Cain, & Mitroff, 2015). Recent research has supported that response inhibition is a significant contributor to friendly fire incidents (Biggs, 2017; Head et al., 2017; Wilson, Head, de Joux, Finkbeiner, & Helton, 2015). For example, Wilson et al. (2015) investigated response inhibition errors in a simulated firearms SART: A stationary civilian marksman using a laser-tag gun engaged confederates (i.e., a mixture of “friends” and “foes”) who stepped in and out of a doorway. Wilson and colleagues found the civilian marksman had a 33% error of commission rate (response inhibition error). Further corroborating evidence for this phenomenon has been exhibited by trained infantry Soldiers completing a live-fire version of the SART, resulting in an overall error of commission rate of 40% (Head et al., 2017). Importantly, the simulated and live-fire SART investigations illustrate that this phenomenon is not limited to a laboratory setting but has ecological validity.
Since the inception of the SART, there has been an effort to mitigate response inhibition errors by increasing intentional response slowing. For example, modulating the task instructions of the SART (i.e., emphasis on slower response) significantly reduces errors of commission (Seli, Cheyne, & Smilek, 2012). In addition, changing the motor-response characteristics of the SART by incorporating a “mouse point-and-click” response has also shown promise of significantly reducing errors of commission at the expense of slower response latency (SATO; Head & Helton, 2013; Wilson et al., 2018). At the trial level, the point-and-click SART requires participants to physically move a mouse cursor that is surrounded on screen by four boxes presented at equal distances that could contain either a Go or No-Go stimulus. The point-and-click SART increases the motor complexity of the response by requiring participants to coordinate a mouse cursor to an adjacent box containing a stimulus and executing an appropriate response. The coordinated effort of moving the mouse and selecting the box tends to increase the time participants take to respond, relative to a traditional single-button response. This extra time allows the participant to inhibit the feed-forward motor ballistic routine, which results in significantly fewer errors of commission—albeit at the cost of increased response time.
The potential assimilation of weaponized combat robots within the U.S. military could be problematic with concern to friendly fire or collateral damage as a function of response inhibition error. Soldiers may face a target-rich environment (as is seen in the SART) that would only require a shoot or no-shoot response. This target-rich environment may yield fast, repetitive button responses by the combat robot’s human operator, which could have dire consequences for incorrectly targeted personnel. A potential solution is to modulate the motor response to the target by changing the response from a single-button response to a multiple-button response, which may provide a “procedural countermeasure” that mitigates errors (Helton & Kemp, 2011).
In the current investigation, we seek to elucidate the effects of modulating the motor-response characteristics of the SART by incorporating a 3-button “complex” (SART-C) versus a 3-button “simple” (SART-S) response paradigm as a function of period of watch. It was predicated the SART-C relative to the SART-S would yield slower response times and thus decrease errors of commission. With concern to period of watch, we predicted the SART-C condition would result in stable (i.e., unchanging) response time and errors of commission over time. Conversely, it was predicted the SART-S condition would result in progressively faster response times and increased errors of commission over time. Given the inherent fine-motor demands of the tasks (i.e., multiple and continuous button response), we postulate that errors of omission should progressively increase as a function of time-on-task and necessitate forced rest breaks during the SART-C and SART-S conditions. Finally, we utilize subjective (NASA–Task Load Index [NASA-TLX]) and objective measures (heart-rate variability [HRV]) to index workload.
Method
Participants
In total, 16 healthy, male Soldiers were originally recruited to participate in the study. However, due to technical difficulties, only 15 Soldiers participated in all phases of this investigation. Soldiers were provided informed consent in accordance with the Helsinki Accord, and ethics permission was obtained from the U.S. Army Research Laboratory’s Institutional Review Board. All Soldiers held military occupational specialty 11C (indirect-fire infantry) or 11B (infantry). The Soldiers’ average age was 23.93 years (SD = 2.93) with a range of 18–29.
Procedure
The Soldiers were recruited from various duty stations within the United States. All Soldiers completed a video baseline, complex button-response SART-C, and simple button-response SART-S. The order of conditions was counterbalanced among Soldiers and randomly assigned. Soldiers completed three separate visits, which took place within 1 week (once a day) at the same time of day. The infantrymen were given specific instructions to sleep at least 7 hr nightly and to be consistent with their daily depressant and stimulant intake (i.e., alcohol, nicotine, and caffeine). During the 3 days of the experiment, Soldiers’ substance use was recorded each day prior to starting the experimental tasks.
Typically, on Day 1 of the experiment, the Soldiers first completed an informed consent and demographics questionnaire. Prior to the video baseline, the Soldiers surrendered time-keeping devices and cellphones. Each was outfitted with a five-lead electrocardiogram (ECG; Shimmer3, Dublin, Ireland) prior to completing the video baseline. The Shimmer3 ECG enabled HRV to be measured to index parasympathetic stress responses. For the video baseline, each Soldier was isolated in a room and viewed a video documentary on trains, “The American Orient Express” (Pegasus-Eagle Rock Entertainment, 2004). The documentary was presented on a video display terminal (53.4 cm × 33 cm, 60-Hz refresh rate) that was mounted at eye level. This type of stimuli has been used in prior investigations because it is neutral in content and helps maintain a stable heart rate and stable mood (Head et al., 2017; Head et al., 2016). The video ran 51 min, during which HRV was measured. Upon the video’s completion, Soldiers completed the NASA-TLX (Hart & Staveland, 1988), a widely used measure of workload comprising six individual scales. The scales can be examined individually or aggregated to derive a generalized workload measure (see Head et al., 2017).
On Days 2 and 3, the Soldiers wore their Shimmer3 ECGs during a practice session (1.67 min) prior to starting the SARTs. Each Soldier received visual feedback (“correct” or “incorrect”) from the computer on response accuracy during the practice session; however, response time feedback was not provided. In addition, a researcher remained in the room only during the practice session to provide guidance on task requirements.
Upon successful completion of the practice session, each Soldier completed one version of the SART each day that involved monitoring number stimuli. The number stimuli, 1 to 9, were individually presented in the center of the screen for 250 ms followed by a mask for 1,200 ms. The Soldier was instructed to respond overtly to numbers 1, 2, and 4–9 and to withhold responses to the predesignated target, 3. The numeric stimuli were displayed in Courier font; however, the font size varied among 48, 72, 94, 100, and 130 (i.e., 12–29 mm). Different font sizes were used to encourage Soldiers to interpret the number rather than rely on peripheral feature mapping (Robertson et al., 1997). Font-size selection was achieved by randomly sampling from a nested list of the five aforementioned fonts without replacement. Once all five fonts were sampled, the nested list would reset with the same values until the completion of the study. In total, the SART’s duration was 51 min and consisted of eight periods (6.30 min in duration) of watch with each period containing 261 trials (232 Go distractors and 29 No-Go targets). Font size was not counterbalanced among periods; however, post hoc visual inspection of frequency distributions of fonts as a function of period of watch and condition did not indicate any systematic differences or interactions.
The Soldier was instructed to use the index finger of his dominant hand to make the response as fast and accurately as possible on a modified serial response box. The response box consisted of five buttons; however, two of the buttons were obscured using a shroud. Above the three visible buttons were labels giving a visual indicator of the button number. Below each number label was a light indicator that would illuminate if that button was pressed. The purpose of the light indicator was to give the research assistant a visual cue of whether the participant was pressing the button correctly during the practice trials. (see Figure 1 for diagram of response box). For the SART-C, Soldiers were instructed to press three buttons separately in numeric sequential order (“1,” “2,” and “3”). To accomplish the button sequence, the Soldier had to press the far left button, far right button, and then the middle button (Figure 1). For the SART-S, Soldiers only had to press the button corresponding to label “1” three separate times with his index finger as fast and accurately as possible. For the purpose of this investigation, the Soldier was instructed that only a correct button sequence regardless of condition would be counted as a full response. Thus, partial responses (e.g., “1” . . . “1” or “1” . . . “2”) or incorrect sequences would be counted as a “no response.” Response time was recorded only for correct button-sequence responses. The last button press of the sequence was recorded for the response time. The SARTs were completed on the same video display terminal as the baseline video.
Graphical representation of the response box with button labels.
ECG Instrumentation and Analysis
A five-lead ECG was placed on the Soldier to record waveforms during the baseline and SARTs. Electrode placement was visually confirmed by inspecting ECG waveforms in real time on a wireless tablet (Samsung Galaxy 10, Suwon, South Korea) prior to the baseline and SARTs. The ECG signal was sampled at 1,024 Hz and recorded for later analysis. The cycle-length variability (i.e., the interval between R peaks within the QRS complex) of the ECG waveform were later determined algorithmically (Goldberger et al., 2000) and visually confirmed by at least one researcher to ensure accurate interval identification. The R peak is the upward deflection of the QRS complex and is the apex of the complex. The entire durations of the baseline and SARTs were recorded and analyzed using the R heart-rate variability (RHRV) package, version 4.2.2 (Rodríguez-Liñares, Vila, Méndez, Lado, & Olivieri, 2008) to calculate the SD of the normal to normal (SDNN) R-R intervals. The R-R interval is the time between successive QRS complexes as measured by the peak of the QRS wave (Ashley & Niebauer, 2004). The SDNN calculation was used to index HRV as a function of the task and period of watch. Generally, higher SDNN values correspond to greater parasympathetic activity (i.e., a low-stress state), whereas lower SDNN values indicate a lower parasympathetic state (i.e., higher stress state; Hjortskov et al., 2004; Thayer, Ahs, Fredrikson, Sollers, & Wager, 2012). Thus, SDNN was used to explore autonomic stress responses to the video control and SART tasks (Head et al., 2017; Luque-Casado, Perales, Cárdenas, & Sanabria, 2016).
Questionnaire
A computer version of the NASA-TLX was used to subjectively measure workload (Hart & Staveland, 1988). The NASA-TLX consists of three items measuring external demand (physical, temporal, and mental) and three internal responses to external demands (performance, frustration, and effort). A global measure of workload, which was the aggregate of the six items, was computed.
Statistical Analysis
All analyses were conducted using generalized estimating equation (GEE) models that account for the repeated measure design (i.e., correlated data structure; Zeger, Liang, & Albert, 1988). For all GEEs, which incorporate a repeated time component (e.g., period of watch), a first-order autoregressive correlation structure was used; other GEEs utilized an independent correlation matrix to account for subject-level correlations. In addition to accounting for correlated data structures, GEEs have relaxed assumptions regarding constant variance and utilize the nonparametric Wald test (χ2) for inferential testing (Ballinger, 2004). Residuals were inspected for normality, and the assumption of constant variance was verified by plotting fitted data against residuals (Yan & Fine, 2004). During a sensitivity analysis and deletion diagnostics on the GEE models, one participant was identified as a potential outlier for errors of commission; however, removal of that participant did not alter statistical significance. Thus, the participant’s data were retained for further analysis. All statistics were performed in R-3.2.3 (R Development Core Team, 2015) using “geepack” (Halekoh, Højsgaard, & Yan, 2006).
Mean and ± SD of the mean from model estimates are reported unless otherwise stated. GEEs were used to test whether accuracy (errors of commission and errors of omission) and response time were influenced by condition (SART-S vs. SART-C) and period of watch. For HRV, SDNN was calculated for the video baseline control, SART-S, and SART-C. The SDNN was then assessed with the GEE to determine changes in autonomic stress response as a function of condition and period of watch. For the NASA-TLX, global workload composite scores were calculated by aggregating the six items of the NASA-TLX. The GEE was used to test whether subjective workload composite scores changed as a function of task type (i.e., SART-S and SART-C). Post hoc analysis of substance use with behavioral and physiological measures did not significantly influence model estimates and thus was not included in the results.
Results
SART Performance
For errors of commission, there was a significant main effect for condition and period of watch, χ2(1, N = 15) = 4.86, p = .03, and χ2(1, N = 15) = 6.61, p = .01, respectively. There was a significant interaction between condition and period of watch, whereby the complex condition appears flat over time, whereas the simple condition has a significantly steeper slope over time, χ2(1, N = 15) = 7.28, p < .05 (see Figure 2a). For errors of omission, the main effect for condition was not significant, χ2(1, N = 15) = 2.28, p = .13 (see Figure 2b). In addition, period of watch and the interaction failed to reach significance, χ2(1, N =15) = 1.92, p = .17, χ2(1, N = 15) = 1.96, p = .16, respectively.
a) Displays errors of commission as a function of condition. b) Displays errors of omission as a function of condition. Depicted means were derived from model estimates. Error ribbons were derived from model estimates, which depict standard error of the mean.
For response time, there was a significant main effect for condition indicating that Soldiers overall were slower on the SART-C relative to SART-S, χ2(1, N = 15) = 37.4, p < .001 (see Figure 3). Conversely, a main effect of period of watch and the interaction failed to reach significance, χ2(1, N = 15) = .15, p = .22, χ2(1, N = 15) = 0.05, p = .49, respectively.
Response time as a function of period of watch and condition. Depicted means were derived from model estimates. Error ribbons were derived from model estimates, which depict standard error of the mean.
Subjective Workload State
The GEE model revealed no significant difference between SART conditions on the global workload measure, χ2(1, N = 15) = 2.24, p = .13 (see Figure 4).
Jitter plot indicating the distribution and average (gray square) global workload measure as a function of condition. Bars depict standard error of the mean.
Autonomic Stress Response
There was a significant main effect for condition, χ2(2, N = 15) = 16.86, p < .001; however, the main effect for period of watch failed to reach significance, χ2(1, N = 15) = 0.20, p = .65. There was a significant interaction between condition and period of watch, χ2(2, N = 15) = 6.85, p = .03. A simple slope analysis revealed that SART-C showed greater change over time relative to SART-S, p = .02 (see Figure 5). All other simple slope comparisons failed to reach significance, p > .05.
Displays standard deviation of the normal to normal (SDNN) scores for complex, simple, and video condition. Error ribbons were derived from model estimates, which depict standard error of the mean.
Discussion
In the present investigation, we examined response inhibition performance as a function of response complexity. Two visually identical SARTs were created but differed with respect to the response complexity (simple vs. complex). Specifically, we were interested in whether modulating the response complexity differentially influenced response inhibition on the SART. Subjective workload was measured with the NASA-TLX after each SART condition. In addition, parasympathetic stress response was indexed by measuring HRV during each SART and a control.
Percent errors of commission on the SART-S (14%) was significantly higher relative to the SART-C (4%) as a function of response complexity (i.e., simple vs. complex). Interestingly, the SART-C condition appeared to remain relatively constant over time, whereas the SART-S condition produced a steady linear increase as a function of period of watch. Moreover, Soldiers exhibited faster response times in the SART-S condition (739 ms) relative to the SART-C condition (874 ms). Although the interaction failed to reach significance for response time, it visually appeared that response time got faster for the SART-S condition as a function of period of watch, whereas the SART-C condition remained relatively stable over time. Collectively, the error of commission and response time results indicate that Soldiers responded faster at the cost of making more errors in the SART-S condition, whereas making slower but more accurate responses in the SART-C condition (i.e., SATO effect). Overall, the errors of omission rate were low, and there was no significant difference between SART-C and SART-S. As stated in the “Method” section, only full response sequences were counted as a response, whereas an absent response or partial responses constituted no response. The error of omission results provide evidence that Soldiers overall were compliant with the task-response requirements.
Traditionally, the dependent measures (response time and accuracy) of the SART are measured by a single button press to Go stimuli and withholding to No-Go targets (Manly, Robertson, Galloway, & Hawkins, 1999; Robertson et al., 1997). As participants frequently respond to Go stimuli, they develop a feed-forward motor ballistic routine that is characterized by speeded responses (Head & Helton, 2012, 2013; Head, Russell, Dorahy, Neumann, & Helton, 2012). The motor ballistic routine becomes difficult to inhibit when a No-Go stimulus is presented. From a theoretical standpoint, it appears that introducing complexity to the button response in the SART-C condition counteracts the feed-forward motor ballistic routine, which facilitates decreases in errors of commission. Conversely, the lack of response complexity of the SART-S condition promotes faster responses at the cost of increased errors of commission. Considering the SART-C and SART-S conditions collectively, modulating the response characteristics by increasing the response from a single-button to a multiple-button response facilitated a 9% error of commission rate and response latency of 807 ms. Conversely, the traditional SART, which uses a single-button response, produces a relatively higher error-rate range (29%–55%) with a faster response latency range of 325–421 ms (Head et al., 2012; Helton, 2009; Helton, Head, & Russell, 2011; Helton, Kern, & Walker, 2009; Manly et al., 2003). As similarly found in point-and-click SARTs (Head & Helton, 2013; Wilson et al., 2018), changing the motor-response characteristics decreases response inhibition errors as measured by errors of commission; however, the cost of improved accuracy was slower response times to Go stimuli.
Both SART conditions were visually identical and required three sequential button presses to register a response. However, the SART conditions differed with concern to complexity of the response and the lateral distances between each button response. Although perhaps negligible, the SART-C condition required a degree of working memory whereby Soldiers had to remember the correct three-button sequence to execute a correct response. Prior investigations suggest that working memory load does negatively influence executive functions such as response inhibition (Clark et al., 2007; Lara, Madrid, & Correa, 2014). However, given the low memory chunk of the button sequence coupled with the repetitive nature (i.e., practice effect) of the task, it is unlikely that working memory had a significant impact on performance in the SART-C condition. Conversely, the differences in performance between the SART-C and SART-S are likely predicated on the physical requirement of the task. In other words, having the response buttons in different locations in the SART-C condition required the Soldier to coordinate his finger to the correct button over a distance, which unavoidably slowed him down to press the button. In essence, the SART-C condition models a Fitts’ paradigm in which the Soldier had to actively slow his response to correctly execute the button response (Fitts, 1954). However, the SART-S condition, which required no lateral movement, encouraged faster albeit less-accurate responses in terms of increased errors of commission. Thus, the lateral movements required in the SART-C condition appeared to afford the Soldier more time to inhibit his response to the No-Go target.
The global workload measure failed to differentiate between the two SART conditions, indicating that increasing response complexity does not appear to influence perceived workload of the task. However, the parasympathetic measures of stress (i.e., HRV) did show changes between SART conditions. The HRV slope of the SART-S and video control condition appears to be relatively flat across the periods of watch. Conversely, the HRV intercept of the SART-C condition appears lower (i.e., lower parasympathetic stress response) at the start of the task and increases across the periods of watch (i.e., increased parasympathetic stress response) converging with the SART-S. The key difference between the subjective workload and HRV measure is in specificity. In other words, the NASA-TLX, being a relatively coarser measure (i.e., single measure at the end), fails to capture the changes in workload over the duration of the task. Conversely, the HRV measure provides a relatively fine-grained measure of parasympathetic stress response that changes during the task. It is possible the SART-C condition required greater workload at the start due to the working memory load and selection and pressing of the correct buttons (i.e., Fitts’ effect). However, as Soldiers repetitively completed the button sequence (i.e., practice effect) of the SART-C, it likely became automatized and thus reduced workload demands toward the end of the task. Future investigations should consider sampling subjective state throughout the task to obtain a relatively finer-grained measure of workload.
Response inhibition errors are not only problematic in laboratory settings but also in relatively more complex, ecologically valid environments (e.g., live-fire shooting). The potential for response inhibition errors may be compounded when utilizing combat robots, which may only require a Soldier to make a single-button response to a high number of predesignated targets. From a practical standpoint, utilizing a multiple-button response could be advantageous in reducing errors; however, it is unclear whether the response time trade-off is acceptable given specific equipment and scenarios. Although a delayed response time could reduce friendly fire incidents, there is a heightened risk of being engaged first by a faster-firing enemy. This trade-off could be met with resistance from many parties—in particular, Soldiers in theater (for obvious reasons). However, as increasingly more weapons are operated remotely, this trade-off may be increasingly welcomed. Essentially, being engaged by a faster-firing enemy may lead to a damaged robot (material cost) as opposed to the loss of a human life. An important and related question is whether response complexity should be used as a function of potential collateral damage, for example, should the level of complexity be increased in congested urban areas relative to an open landscape? Moreover, should different button-response combinations be weapon dependent (e.g., machine gun vs. grenade launcher) to differentiate? Future investigations should consider examining these questions to reduce potential response inhibition errors while maintaining optimal accuracy and response time.
The current investigation provides evidence that introducing complexity into the response criteria may mitigate unintentional response inhibition errors, which could decrease the likelihood of incorrectly engaging the wrong targets (i.e., friendly fire). Contrary to typical Human Factors–usability–design recommendations, to prevent friendly fire in the future, designers may need to intentionally design unfriendly interfaces when designing the controls for lethal systems. This is comparable to information security, where usability and security are often at odds with each other (Kainda, Flechais, & Roscoe, 2010). Making the interface easier to use may actually increase lethal human errors.
Key Points
The present investigation examined the influence of response complexity (simple vs. complex) on the sustained attention to response task (SART).
The results revealed that modulating how the Soldier responds to a target can significantly decrease errors of commission at the cost of slower response time.
The subjective workload measure (NASA-TLX) failed to differentiate between the complex and simple SART; however, the physiological measure (HRV) did show an increased stress response in the complex condition that decreased with time on task.
The results of the current investigation provided evidence that response complexity could provide a solution to mitigate potential engagement errors that could occur with combat robots.
Footnotes
Acknowledgments
The authors declare this research was conducted in the absence of any commercial or financial relationship that could be construed as a potential conflict of interest. The results of the present study do not constitute endorsement by the U.S. Department of Defense. The following personnel provided invaluable assistance during study: William Maslin, Patrick Chou, Ed Bauer, Tom Fry, Frank Morelli, Samson Ortega, and Doug Struve.
Author(s) Note
The author(s) of this article are U.S. government employees and created the article within the scope of their employment. As a work of the U.S. federal government, the content of the article is in the public domain.
James Head received his PhD in psychology with an emphasis in human factors and applied cognitive psychology from the University of Canterbury in Christchurch, New Zealand (2013). His research interests include vigilance, response inhibition, marksmanship performance, and the interaction between physical and cognitive fatigue.
Matthew S. Tenan received his PhD in kinesiology from the University of Texas (2014). His research includes statistical modeling, sex differences, and human control of movement.
Andrew J. Tweedell received his MA in exercise physiology from the University of North Carolina at Chapel Hill. His research interests include neuromuscular physiology, motor control, the interaction between cognitive and physical fatigue, and augmentation technologies for human physical-performance enhancement.
Kyle M. Wilson received his PhD in psychology with an emphasis in human factors and applied cognitive psychology from the University of Canterbury in Christchurch, New Zealand (2016). His research interests include vigilance, response inhibition, automation, user experience, and the interaction between physical and cognitive fatigue.
William S. Helton received his PhD in human factors from University of Cincinnati (2002). His interests include vigilance, response inhibition, disaster and emergency response, the interaction between cognitive and physical performance, wearable interfaces, and working dogs.