IEEE Robotics & Automation Magazine - June 2019 - 74

choice. Next, we discuss the empirical evidence on transparency and transparency effects found in HRI research.
HRI Transparency Realization
Sociological and psychological studies have explored transparency effects [13] and expectations in the context of robotics [4]. This research investigates the user effects of
information provision, in the sense of explanations of how
and why the robot does what it does. It shows that user perceptions of and attitudes toward transparency differ substantially depending on the technologies and services investigated,
the tasks given, and other contextual factors. In some cases,
robot transparency has limited impact, for example, on attributions of credit and blame [13] or on assessments of competence [15]. In other cases, robot transparency leads to poorer
perceptions of the robot [20].
How transparency is enacted can lead to different outcomes [22]. The situational importance of transparency has
also been pointed out, suggesting that technology should be
transparent and able to explain itself in critical states. However, this may not be as advantageous when everything is running as usual [11].
General research on transparency in intelligent systems shows similarly mixed results. It does not permit
drawing generalized design recommendations for
transparency and explainability [7]. While the requirement for transparency has strong ethical and rightsbased support and is now a legal requirement in the
GDPR, these mixed results indicate that, from a purely
pragmatic and user-centered perspective, an increase in
transparency is not always clearly desirable. Accordingly, from an industry point-of-view, investment in transparency by developers could be costly, with unclear
effects and benefits, and there may even be a risk that,
at times, transparency could backfire by decreasing
user trust [6].
Fischer [9], for example, points to the detrimental effects
transparency might have in the context of assistive robots. If
such robots are too transparent about their information processing capacities, that transparency might impede natural
and seamless HRI. The desirability of transparency might
depend heavily on the application domain and particular type
of HRI. Accordingly, when engaging with the demand for
transparency, any transparency measures need to be designed
with due regard for the specific characteristics of HRI in that
use context.
What transparency means for the field of robotics is still
underexplored. This is despite recently intensifying research
efforts in the form of a dedicated IEEE group on the transparency of autonomous systems (IEEE-P7001; see http://sites.ieee
.org/sagroups-7001/) or a recent workshop on explainable
robotic systems, whose contributions are summarized in
Table 2 (see https://explainableroboticsystems.wordpress.com/).
From this table, we can see that HRI research on transparency has focused primarily on the explainability of the systems, either from the robot (intelligibility) or user perspective
74

IEEE ROBOTICS & AUTOMATION MAGAZINE

JUNE 2019

(understandability). Their conclusions vary, with some studies showing no relevant findings and some identifying the
downsides of transparency. Such disadvantages include the
inhibition of seamless HRI and the creation of misunderstandings and unwanted inferences about a robot's capabilities [9]. These findings are in line with other results
highlighting the technical limitations of transparency [14],
that it can create false binaries and be harmful, and that it
could be used to prioritize seeing over understanding [1].
Checklist for Implementing Transparency
in Robot Development
The following checklist translates the general considerations
outlined previously into a step-by-step guide for robot developers (Table 3). Its goal is to provide user-centered guidance
on how to design for transparency.
To implement transparency in a given AI environment, we
suggest the following:
1) identify general transparency obligations
2) identify the different transparency needs and expectations
of the involved stakeholders
3) translate the transparency requirements to the level of
understanding of a target group
4) conduct user testing concerning some transparency-related
parameters
5) guide users concerning available transparency functions.
Conclusions
This article offered an overview of the transparency requirement and explained the main dimensions of the transparency
principle in the context of robotics. The implementation of
legal transparency requirements requires interdisciplinary
collaboration among legal, social science, and technology
experts to avoid overlooking ethical and societal aspects and
create an evidence base that will be essential for engineers and
industry in designing transparency measures that are effective
and meet legal requirements [2], [17].
Future HRI research on transparency should investigate
the situational and contextual value of transparency, user
awareness and needs, and design-related questions, such as
how transparency can best be implemented in assistive
robots. In addition to experiments, HRI research could use
ethnographic and observational approaches, interface studies,
and reverse engineering. Doing so will require applying the
findings of this article to concrete use cases, thereby providing
more practical guidance on how to implement transparency
in a given context. The checklist in the previous section could
offer direction in developing such use cases in practice.
There are multiple takeaways from this subject.
● Transparency is an ethical requirement based on the value
of autonomy and is essential for meaningful informed consent. Data subjects must be informed about how controllers process their data in a concise, easily accessible, and
understandable manner.
● It is also a legal requirement that is binding in the European
Union and for the processing of personal data of European

http://sites.ieee.org/sagroups-7001/ http://sites.ieee.org/sagroups-7001/ https://explainableroboticsystems.wordpress.com/

IEEE Robotics & Automation Magazine - June 2019

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2019