IEEE Systems, Man and Cybernetics Magazine - July 2019 - 54

the physical appearance of the experimental setup, not to
mention external factors that may affect the participant.
Therefore, when participants are asked if they trust a particular setup, their collective response can become pure
noise. The only way to overcome this problem is by creating an experimental setup that ensures complete trust
or distrust.

measure changes in biomarkers relative to changes in the
psychological process of a human participant. This will
then identify the biomarkers associated with distrust and,
by an inverse relationship, trust.

Biomarkers for Trust Measurement
Historically, measurement of trust has been done through
subjective questionnaires [1], [2]. In 2013, Schaefer created
and validated a scale to measure changes in trust in differHow Experiments Should Be Performed
ent types of robots. A 40-item scale was presented as a preand Associated Assumptions
test-posttest human-robot trust change measure. The
I argue that the creation of a completely trustworthy experproposed scale could also accommodate changes in trust
imental setup is not within the capability of an investigator,
across multiple conditions [2].
as he/she cannot control the impact of external factors
An objective measure of the development of trust, or
affecting human participants. However, it is far easier to
the change in trust, is essential to the progress of human-
create a hostile environment in which the participant will
robot interaction, especially with
have no trust at all or will have
the introduction of autonomous
complete distrust. A distrusting
systems where humans might
experimental setup can be arranged
supervise, monitor, or even use the
by creating a sense of danger even
Understanding
autonomous system. Understandwhen no such danger is present. If
and detecting
ing and detecting change in trust
we start an experiment using an
change in trust
during interactions with autonountrustworthy environment, we
mous systems is crucial to improvcan guarantee that human particiduring interactions
ing their design, which will lead to
pants will be in a particular state of
with autonomous
an overall increase in tasks in
mind, which will then act as a basewhich human-autonomous teams
line for all participants. In consystems is crucial
interact. I argue that subjective
trast, establishing this baseline is
to improving their
questionnaires fail to measure
extremely difficult for an experidesign.
trust objectively, and physiological
ment that attempts to begin with a
signals can pave the way for such
trustworthy setup. Therefore, we
objective measurement of trust.
propose that all experiments wantObjective measures can be based
ing to relate trust with any measuron biomarker signals (i.e., physiological signals) such as
able quantity should start with a distrust scenario instead
brain activity (electroencephalogram), heart rate variabiliof a conventional trust scenario.
ty (echocardiogram), skin temperature, eye movement,
This proposal assumes that having a moderate degree
and body movement, among others (see Table 1). Although
of discomfort, or a small amount of fear, is equivalent to
the measure can be based on a sole parameter, a multivaricompletely removing trust from the scenario in question. It
ate formation using different signals will better accommoalso assumes that if the distrust scenario does not cause
date individual differences and variability in context. A
any harm to the participant over time, he or she will grow
multivariate equation will also reflect the complexity and
relaxed, and the component of distrust will be removed
multidimensionality of the trust construct.
from their physiological signals. He or she will be comfortEye behavior is a very promising biomarker for trust
able, and that will act as a proxy for trust. It is very imporanalysis and was studied during interactions with functant to realize that the type of trust we are exploring here
tioning and malfunctioning human-robot interfaces [3].
is not of a general kind; rather it is very much specific to
the trust between robot/autonomous systems and a particThe study showed evidence of change in eye behavior
ular human participant. The trust that the human places in
related to trust in the interface function. Fixation point,
the robot/autonomous system captures only the interacfixation duration, and fixation density were significantly
tion related to a specific task, and this is precisely what I
different and were aligned with differences in trust meawant to explore in a study of trust. This specific task can
sured through subjective questionnaires. Heart rate varibe an autonomous or remote medical diagnostic proceability was also observed in simulated autonomous-car
dure or a self-driving-car ride. Regardless of the task, the
driving [4]. Percentage heart rate change was used as an
underlying type of trust is the same in all scenarios relatinverse measure of trust; it was augmented with other
ing to humans and robot/autonomous systems, and thereself-reported and behavioral factors to determine a final
fore, all experiments in this particular field should begin
trust estimate. More studies are required to reach a comwith a distrust scenario to ensure a set of proper baseline
prehensive objective measure of trust in autonomous sysbiomarkers. Once we have this baseline correct, we can
tems, especially with autonomous systems working in
54

IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE Ju ly 2019

IEEE Systems, Man and Cybernetics Magazine - July 2019

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - July 2019