Академический Документы
Профессиональный Документы
Культура Документы
Objective: This work aims to further test the theory that Introduction
trust mediates the interdependency between automation reli-
ability and the rate of human reliance on automation. Human trust in automation has gained sig-
Background: Human trust in automation has been the nificant interest as a crucial human factor that
focus of many research studies. Theoretically, trust has been can influence human–automation interaction
proposed to impact human reliance on automation by medi- within a mission, and ultimately mission per-
ating the relationship between automation reliability and the
rate of human reliance. Experimentally, however, the results
formance (J. Y. Chen & Barnes, 2014). Theo-
are contradicting as some confirm the mediating role of trust, retically, it has been widely accepted that trust
whereas others deny it. Hence, it is important to experimen- plays a significant mediation role in the rela-
tally reinvestigate this role of trust and understand how the tionship between automation reliability and
results should be interpreted in the light of existing theory. the rate of human reliance on it (J. Gao & Lee,
Method: Thirty-two subjects supervised a swarm of
unmanned aerial vehicles (UAVs) in foraging missions in which
2006; Lee & See, 2004; Parasuraman & Riley,
the swarm provided recommendations on whether or not to 1997). However, this role has not yet been
collect potential targets, based on the information sensed by empirically confirmed. Only a few research
the UAVs. By manipulating the reliability of the recommen- endeavors (Bustamante, 2009; Chancey, Bliss,
dations, we observed changes in participants’ trust and their Proaps, & Madhavan, 2015; Chancey, Proaps, &
behavioral responses.
Results: A within-subject mediation analysis revealed a
Bliss, 2013; Merritt, 2011; Wiczorek & Manzey,
significant mediation role of trust in the relationship between 2010) were dedicated to the experimental study
swarm reliability and reliance rate. High swarm reliability of the mediation role played by trust. Moreover,
increased the rate of correct acceptances, but decreased the the results of some of these empirical studies
rate of correct rejections. No significant effect of reliability were contradicting to the corresponding well-
was found on response time.
Conclusion: Trust is not a mere by-product of the inter-
established theory on the topic as well as among
action; it possesses a predictive power to estimate the level of each other.
reliance on automation. The objective of this work is to take a step
Application: The mediation role of trust confirms the toward filling this gap. We investigate the related
significance of trust calibration in determining the appropriate literature and contrast experimental findings to
level of reliance on swarm automation.
the theory on the role of trust. Then, we present
Keywords: human–automation interaction, human–swarm
our human experiments, the purpose of which is
interaction, design of experiments, foraging, within-subject to reexamine the hypothesized mediation role of
mediation analysis, recommender systems trust.
Gramopadhye, & Melloy, 2003; Xu & Dudek, trust in terms of the reliance behavior confounds
2015) found that trust is significantly affected other constructs with trust (Chancey et al.,
by automation reliability, in terms of the prob- 2015).
ability that automation performs its assigned Both overtrust and distrust are believed to be
tasks correctly. Trust is also of a dynamic nature, detrimental to mission performance (Abbass,
as it can drop notably after automation errors Petraki, Merrick, Harvey, & Barlow, 2016; Bou-
(Akash, Hu, Reid, & Jain, 2017; Desai, Kania- bin, Rusnock, & Bindewald, 2017; J. Y. Chen &
rasu, Medvedev, Steinfeld, & Yanco, 2013; Xu Barnes, 2014). According to many researchers
& Dudek, 2015). Hence, an accurate estimation (e.g., Lee & See, 2004; Yang, Unhelkar, Li, &
of trust requires repeated measurements during Shah, 2017), human’s overtrust toward an agent
an interaction (Schaefer, 2016). can result in overrelying on it despite its limita-
The effect of error type (i.e., false alarms vs. tions, which may lead to catastrophic conse-
misses) on trust has been investigated in some quences. For instance, Robinette, Li, Allen,
studies (Chancey et al., 2015; Merritt, Lee, Howard, and Wagner (2016) found that in emer-
Unnerstall, & Huber, 2015) which did not find gency evacuation scenarios, subjects opted to
this effect to be significant. This result can be follow robot instructions to the exit even when
explained using the model in J. Gao and Lee the robot provided circuitous routes. They attrib-
(2006) in which automation characteristics uted this behavior to subjects’ overtrust in the
affect trust through the perception of perfor- robot. On the contrary, distrust has been consid-
mance. Therefore, unless automation character- ered to lead to automation disuse, hence missing
istics are manipulated to affect performance, its potential benefits (Lee & See, 2004). Thus,
they may not affect trust. Thus, unless one type maintaining an appropriate level of trust is
of errors is significantly more detrimental to the deemed important for mission success and opti-
performance than the other, trust may not be mized performance. Although this stance is
impacted by error type. prevalent among human factors researchers, it is
not well backed up by experimental evidence as
Trust and Reliance On mixed results on the mediation role of trust have
Automation been reported in the literature.
Although trust is a personal attitude toward
an agent, reliance is the actual delegation of a The Mediation Role of Trust
task to the agent. Lee and See (2004) suggested The conceptual diagram for a general media-
that trust in automation is a main factor that tion model is depicted in Figure 1. A variable M
affects human intention to rely on that automa- is said to mediate the relationship between an
tion. Other factors may include workload, self- independent variable X and a dependent variable
confidence, and task-associated risks. In turn, Y if two conditions are met (Hayes, 2017): (1) X
the intention to rely is a main factor, but not affects M, which in turn affects Y. For a variable
the sole one, that predicts the reliance behavior, to affect another variable, it needs to not only
which is also affected by other factors such as correlate with it but also temporally precede it.
time constraints. (2) The indirect effect of X on Y through M is
This distinction between trust and reliance statistically significant from zero.
can lead to situations in which a human trusts Trust has been frequently hypothesized to
automation but does not rely on it due, for exam- mediate the relationship between automation
ple, to some inherent risks. Conversely, the reliability and human reliance on automation (J.
human may rely on an agent, albeit not highly Gao & Lee, 2006; Khasawneh et al., 2003; Lee
trusting it, due to other factors such as experi- & See, 2004; Parasuraman & Riley, 1997).
encing a very high workload or the existence of Nonetheless, experimental results disagree with
time constraints. In other words, trust does not each other regarding the mediating role of trust.
necessarily manifest itself into reliance and reli- Bustamante (2009) examined the mediation
ance is not necessarily a piece of evidence of role of trust in the effect of alarm system reli-
trust (Miller et al., 2016). Thus, operationalizing ability on human reliance strategy. In that work,
Trust Mediates Reliability–Reliance Relationship 3
as flocking, dispersion, and foraging. Robot sense only target objects, confusion with non-
swarms have been gaining increasing interest target objects is also possible according to some
due to the usefulness of their behaviors in a probability values associated with these nontar-
variety of applications ranging from agriculture gets. That is, an object with a detection prob-
to space exploration. ability p is expected to be sensed by an average
Many research studies have used simulated of p UAVs that have a sensing range greater than
robot swarms for foraging (e.g., Nunnally et al., their distance to that object (nearby UAVs). The
2012; Pendleton and Goodrich, 2013; Tang numbers of target and nontarget objects in the
et al., 2017) in which robots are deployed in an environment are equal. Once an object is sensed
environment to collect target objects. Foraging by one or more UAVs, its position is visualized
is a benchmark problem in the swarm robotics and the whole swarm starts heading toward it.
literature as it resembles many real-world mis- Upon arrival, the swarm recommends how the
sions including search and rescue, cleaning, har- object should be handled (collected/discarded)
vesting, and land mine clearance (Winfield, based on the number of UAVs that are able to
2009). sense it (see Figure 2c). If the number of UAVs
This research complied with the American sensing the object is at least 10, the swarm
Psychological Association Code of Ethics and recommends collecting the object, otherwise
was approved by the Human Research Ethics it recommends discarding it. The human can
Committee (protocol no. HC180308). Informed accept a recommendation or request to verify it.
consent was obtained from each participant. Accepting a recommendation leads the swarm
to execute the recommended action. On the con-
Scenario trary, verification is implemented by retrieving
A swarm of 20 unmanned aerial vehicles the object’s image so that the human can iden-
(UAVs) is deployed in an environment contain- tify its type himself or herself (see Figure 3). To
ing two types of initially hidden objects: targets simulate the object identification task, images
and nontargets. The swarm autonomously navi- containing rectangles are used, where targets
gates the environment in a predefined path (see have an even number of rectangles, whereas
Figure 2a). During its navigation, a UAV can nontargets have an odd number of rectangles.
sense an object within its sensing range (see Object collection is a time-consuming opera-
Figure 2b), with a probability associated with tion that takes 5 s following which the gained
that object. Although the UAVs are aimed to score is visualized (see Figure 2d). Collecting
6 Month XXXX - Human Factors
Figure 2. Scenario description: (a) swarm navigation through the environment starting from the
yellow circle onward; (b) a swarm agent detecting an object within its sensing range; (c) the swarm
arrives at the detected object and recommends collecting it; (d) the gained score after the swarm
finished collecting a target; (e) an indication of missed target after the swarm discarded a target
object.
targets increases the score, whereas collecting erroneous recommendations (collect nontar-
nontargets does not change it. The role of the gets and discard targets) are equally probable
human is to make the decisions that maximize in each reliability condition to avoid the con-
the score of the mission by collecting as many founding effect of error bias (Guideline 3).
targets as possible. The time constraint on the
mission implies that collecting nontargets Experimental Setup
reduces the time available for finding and col- A within-subject design is employed, in
lecting targets. Frequent human verification can which each subject supervised the swarm in the
also waste the available time as image retrieval two reliability conditions. The within-subject
and object identification are time-consuming. design is chosen as it is beneficial to account-
Image retrieval takes 1 s, whereas object identifi- ing for individual differences that can affect
cation depends on the recognition and judgment subjects’ behavior (Guideline 3). The order of
speed of the human. The task is designed such the missions was counterbalanced to correct for
that proper reliance strategies, with respect to the ordering effects (Guideline 3). Thirty-two sub-
level of reliability, yield higher performance. jects participated in the experiment: 17 females
Two levels of recommendation reliability and 15 males. Their average reported age is
(90% and 70%) have been used. These levels 28.13 years (SD = 5.34).
were chosen to be at least 70% (Wickens & Subjects first read and agreed to the consent-
Dixon, 2007) to ensure that automation is ben- to-participate form, and then they filled in a short
eficial to the performance (Guideline 1). The questionnaire. They were introduced to the task,
gap between the two reliability conditions was given its instructions, and allowed to ask ques-
selected to make a notable difference to the tions before starting the task. Subjects were noti-
performance (Guideline 2). The two types of fied that the accuracy of the recommendations
Trust Mediates Reliability–Reliance Relationship 7
Figure 4. Participants’ response time during the experiments. The dashed line indicates the end of the
2-min warm-up period.
Figure 5. The structure of the data used for the analysis. The number inside the parentheses indicates the
row number to which the observation belongs.
M0, the estimates of the within- and between- 95% CI = [0.24, 0.45]) and trust (βb = 0.17, 95%
subject variance are 0.68 and 0.32, respectively. CI = [0.02, 0.33]) is statistically significant. The
This model will be used as a baseline in com- marginal and conditional variance explained of
parison to the models containing predictors, the M2 are Rm2 = .18 and Rc2 = .52 , respectively. The
random intercept models. Significant differ- AIC values for both M1 (615.7) and M2 (612.7)
ences between the models can be indicated by are significantly lower than that for M0 (681), p <
their AIC values. .001. Moreover, the difference between the AIC
In the first random intercept model for reli- values for M1 and M2 is statistically significant
ance rate M1, only reliability is incorporated (p = .03).
within the model, leading to a reduced within- ICC for the response time is .6. Three mod-
subject variance of 0.51. The decrease in the els have also been fitted for response time. In
within-subject variance reflects the model’s the variance component model M0, the esti-
ability to explain a portion of the variation in mates of the within- and between-subject vari-
subject’s reliance rate, which is not accounted ance are 0.39 and 0.6, respectively. The first
for by M0. The regression of reliance rate on reli- random intercept model for response time M1
ability is statistically significant with βc = 0.4, incorporates the effect of reliability; however,
95% CI = [0.31, 0.49], Rm2 = .16 , and Rc2 = .5 . this does not affect the within-subject variance.
In the second random intercept model for reli- The regression of response time on reliability is
ance rate M2, both reliability and trust are incor- not statistically significant from zero (β =
porated within the model, leading to a reduced −0.02, 95% CI = [−0.1, 0.05]). Furthermore,
within-subject variance of 0.49. The regression adding trust in the second random intercept
of reliance rate on both reliability (βc1 = 0.34, model for response time M2 does not affect the
Trust Mediates Reliability–Reliance Relationship 9
M SD Minimum Maximum 1 2 3 4
Between-
Response Within-Subject Subject
Rm2 Rc2
Variable Model β Variance Variance
Note. Two beta coefficients are listed for model M2, where the first is for reliability and the second is for trust.
Figure 7. The average values of subject’s trust and performance in the two scenarios.
many studies (for instance, Rovira, McGarry, & Akash, K., Hu, W.-L., Reid, T., & Jain, N. (2017). Dynamic model-
ing of trust in human-machine interactions. In 2017 American
Parasuraman, 2007). Control Conference (ACC) (pp. 1542–1548). doi:10.23919/
The limitation of this study is that the role of acc.2017.7963172
trust is considered in isolation to other factors Boubin, J. G., Rusnock, C. F., & Bindewald, J. M. (2017). Quanti-
that can affect trust and reliance. For instance, fying compliance and reliance trust behaviors to influence trust
in human-automation teams. Proceedings of the Human Fac-
individual differences (e.g., propensity to trust) tors and Ergonomics Society Annual Meeting, 61, 750–754.
could affect the strength of the mediation role of doi:10.1177/1541931213601672
trust. Automation transparency could also Bustamante, E. A. (2009). A reexamination of the mediating effect
impact the strength of this role due to its positive of trust among alarm systems’ characteristics and human
compliance and reliance. Proceedings of the Human Fac-
impact on trust (Dorneich et al., 2017; Wang, tors and Ergonomics Society Annual Meeting, 53, 249–253.
Pynadath, & Hill, 2016). Incorporating such fac- doi:10.1177/154193120905300419
tors while studying the role of trust can help Chancey, E. T., Bliss, J. P., Proaps, A. B., & Madhavan, P. (2015).
identify contexts where trust, and thus trust cali- The role of trust as a mediator between system characteris-
tics and response behaviors. Human Factors, 57, 947–958.
bration, is of particular importance to the proper doi:10.1177/0018720815582261
reliance. Chancey, E. T., Proaps, A., & Bliss, J. P. (2013). The role of
trust as a mediator between signaling system reliability
and response behaviors. Proceedings of the Human Fac-
Acknowledgment
tors and Ergonomics Society Annual Meeting, 57, 285–289.
This work was funded by the Australian Research doi:10.1177/1541931213571063
Council’s Discovery Grant (no. DP160102037) and Chen, J. Y., & Barnes, M. J. (2014). Human–agent teaming for mul-
tirobot control: A review of human factors issues. IEEE Trans-
University of New South Wales, Canberra. actions on Human-Machine Systems, 44, 13–29. doi:10.1109/
thms.2013.2293535
Key Points Chen, M., Nikolaidis, S., Soh, H., Hsu, D., & Srinivasa, S. (2018).
Planning with trust for human-robot collaboration (arXivpre-
•• Human factors literature usually considers trust print). Retrieved from https://arxiv.org/abs/1801.04099
to be the main mediator between automation reli- Desai, M., Kaniarasu, P., Medvedev, M., Steinfeld, A., & Yanco,
ability and human reliance on automation. This H. (2013). Impact of robot failures and feedback on real-time
hypothesized role was not empirically confirmed trust. In Proceedings of the 8th ACM/IEEE International
Conference on Human-Robot Interaction (pp. 251–258).
by past studies due to their conflicting findings. doi:10.1109/hri.2013.6483596.
•• By scrutinizing the empirical studies on the medi- Dorneich, M. C., Dudley, R., Letsu-Dake, E., Rogers, W., Whitlow,
ation role of trust, we assessed that their findings S. D., Dillard, M. C., & Nelson, E. (2017). Interaction of auto-
mation visibility and information quality in flight deck infor-
could have been affected by experimental issues
mation automation. IEEE Transactions on Human-Machine
leading to this inconsistency. We designed a few Systems, 47, 915–926. doi:10.1109/thms.2017.2717939
guidelines that need to be followed by studies on Gao, F., Clare, A. S., Macbeth, J. C., & Cummings, M. (2013).
the mediation role of trust. Modeling the impact of operator trust on performance in mul-
tiple robot control. In 2013 AAAI Spring Symposium Series.
•• By adhering to the guidelines in our human–
Retrieved from https://www.aaai.org/ocs/index.php/SSS/
swarm interaction experiment, our empirical SSS13/paper/view/5741
findings support the theoretical position that trust Gao, J., & Lee, J. D. (2006). Extending the decision field theory to
plays a significant mediation role in the relation- model operators’ reliance on automation in supervisory control
situations. IEEE Transactions on Systems, Man, and Cybernet-
ship between automation reliability and human ics, Part A: Systems and Humans, 36, 943–959. doi:10.1109/
reliance on it. tsmca.2005.855783
Hayes, A. F. (2017). Introduction to mediation, moderation, and
ORCID iD conditional process analysis: A regression-based approach
(2nd ed.). New York, NY: Guilford Publications.
Aya Hussein https://orcid.org/0000-0003- Herting, J. R. (2002). Evaluating and rejecting true mediation mod-
3204-0712 els: A cautionary note. Prevention Science, 3, 285–289.
Jian, J.-Y., Bisantz, A. M., & Drury, C. G. (2000). Foundations for
an empirically determined scale of trust in automated systems.
References International Journal of Cognitive Ergonomics, 4, 53–71.
Abbass, H. A., Petraki, E., Merrick, K., Harvey, J., & Barlow, doi:10.21236/ada388787
M. (2016). Trusted autonomy and cognitive cyber symbio- Khasawneh, M. T., Bowling, S. R., Jiang, X., Gramopadhye, A.
sis: Open challenges. Cognitive Computation, 8, 385–408. K., & Melloy, B. J. (2003, November). A model for predict-
doi:10.1007/s12559-015-9365-5 ing human trust in automated systems. In Proceedings of
12 Month XXXX - Human Factors
the Eighth Annual International Conference on Industrial Biology and Bioinformatics, 15, 1943–1950. doi:10.1109/
Engineering—Theory, Applications and Practice (pp. 216– tcbb.2017.2682161
222). Las Vegas. Wang, N., Pynadath, D. V., & Hill, S. G. (2016). The impact of
Lee, J. D., & See, K. A. (2004). Trust in automation: Design- POMDP-generated explanations on trust and performance in
ing for appropriate reliance. Human Factors, 46(1), 50–80. human-robot teams. In Proceedings of the 2016 International
doi:10.1518/hfes.46.1.50.30392 Conference on Autonomous Agents & Multiagent Systems
Li, Y., Bienias, J. L., & Bennett, D. A. (2007). Confounding in the (pp. 997–1005). Retrieved from http://dl.acm.org/citation.
estimation of mediation effects. Computational Statistics & cfm?id=2936924.2937071
Data Analysis, 51, 3173–3186. doi:10.1016/j.csda.2006.10.016 Wickens, C. D., & Dixon, S. R. (2007). The benefits of imper-
MacKenzie, I. S. (2012). Human-computer interaction: An empiri- fect diagnostic automation: A synthesis of the literature.
cal research perspective. Boston, MA: Newnes. Theoretical Issues in Ergonomics Science, 8, 201–212.
Merritt, S. M. (2011). Affective processes in human–automation doi:10.1080/14639220500370105
interactions. Human Factors, 53, 356–370. doi:10.1177/00 Wiczorek, R., & Manzey, D. (2010). Is operators’ compliance with
18720811411912 alarm systems a product of rational consideration? Proceed-
Merritt, S. M., Lee, D., Unnerstall, J. L., & Huber, K. (2015). Are ings of the Human Factors and Ergonomics Society Annual
well-calibrated users effective users? Associations between cal- Meeting, 54, 1722–1726. doi:10.1177/154193121005401976
ibration of trust and performance on an automation-aided task. Winfield, A. F. (2009). Foraging robots. In R. A. Meyers (Ed.),
Human Factors, 57, 34–47. doi:10.1177/0018720814561675 Encyclopedia of complexity and systems science (pp. 3682–
Miller, D., Johns, M., Mok, B., Gowda, N., Sirkin, D., Lee, K., 3700). Springer. doi:10.1007/978-0-387-30440-3_217
& Ju, W. (2016). Behavioral measurement of trust in auto- Xu, A., & Dudek, G. (2015). OPTIMo: Online probabilistic trust
mation: The trust fall. Proceedings of the Human Factors inference model for asymmetric human-robot collaborations.
and Ergonomics Society Annual Meeting, 60, 1849–1853. In Proceedings of the Tenth Annual ACM/IEEE International
doi:10.1177/1541931213601422 Conference on Human-Robot Interaction (pp. 221–228).
Nunnally, S., Walker, P., Kolling, A., Chakraborty, N., Lewis, M., doi:10.1145/2696454.2696492
Sycara, K., & Goodrich, M. (2012, October). Human influence of Yang, X. J., Unhelkar, V. V., Li, K., & Shah, J. A. (2017). Evaluat-
robotic swarms with bandwidth and localization issues. In 2012 ing effects of user experience and system transparency on trust
IEEE International Conference on Systems, Man, and Cybernet- in automation. In Proceedings of the 2017 ACM/IEEE Inter-
ics (SMC) (pp. 333–338). doi:10.1109/ICSMC.2012.6377723 national Conference on Human-Robot Interaction—HRI 17
Parasuraman, R., Molloy, R., & Singh, I. L. (1993). Performance (pp. 408–416). ACM Press. doi:10.1145/2909824.3020230
consequences of automation-induced “complacency.” The
International Journal of Aviation Psychology, 3(1), 1–23.
Aya Hussein is a PhD student in the Computer Sci-
Parasuraman, R., & Riley, V. (1997). Humans and automation:
Use, misuse, disuse, abuse. Human Factors, 39, 230–253. ence program at the University of New South Wales,
doi:10.1518/001872097778543886 Canberra, Australia. She received her bachelor’s and
Pendleton, B., & Goodrich, M. (2013, August). Scalable human master’s degrees in computer engineering from
interaction with robotic swarms. In AIAA infotech@aerospace
(i@a) Conference. American Institute of Aeronautics and
Cairo University in 2011 and 2015, respectively.
Astronautics. doi:10.2514/6.2013-4731
Robinette, P., Li, W., Allen, R., Howard, A. M., & Wagner, A. Sondoss Elsawah is a senior lecturer in the School of
R. (2016). Overtrust of robots in emergency evacuation sce- Engineering and Information Technology, Univer-
narios. In The Eleventh ACM/IEEE International Conference
on Human Robot Interaction (pp. 101–108). doi:10.1109/
sity of New South Wales, Canberra, Australia. She
hri.2016.7451740 received her PhD in computer science in 2010.
Rovira, E., McGarry, K., & Parasuraman, R. (2007). Effects of
imperfect automation on decision making in a simulated Hussein A. Abbass is a professor in the School of
command and control task. Human Factors, 49, 76–87.
doi:10.1518/001872007779598082
Engineering and Information Technology, Univer-
Schaefer, K. E. (2016). Measuring trust in human robot interac- sity of New South Wales, Canberra, Australia. He is
tions: Development of the “trust perception scale-HRI.” a member of the Human Factors and Ergonomics
In Robust Intelligence and Trust in Autonomous Systems Society.
(pp. 191–218). Springer. doi:10.1007/978-1-4899-7668-0_10
Tang, Q., Ding, L., Yu, F., Zhang, Y., Li, Y., & Tu, H. (2017).
Swarm robots search for multiple targets based on an improved Date received: September 17, 2018
grouping strategy. IEEE/ACM Transactions on Computational Date accepted: September 3, 2019