IEEE Technology and Society Magazine - December 2020 - 48

and again with precision (i.e., " accuracy " 51%). Associations of " efficiency " neither revealed any reflections of
values related to MHC. Instead, paragraphs deal with
" cost " (34%), " war " (58%), " military " (67%), but also with
" precision " (53%) and " quick " (30%), reflecting current
NCW-ratio.

Conflicts of Interest in Implementation of MHC
Our analysis revealed a variety of new findings, some of
which reside on a higher level of abstraction while others are more closely tied to existing practices and materiality. Independent from their theoretical-practical
position, our results yield various implications with
regards to the question of conflicts of interests regarding the implementation of MHC:
a) Time, predictability, and reliability, associated with
each other to varying degrees, are credited to be the
most important values incorporated into emerging
technologies.
b) Accountability, explainability, and potential for intervention are recognized to support MHC.
c) In contrast, efficiency and precision are suggested to
have an ambiguous relationship with affirmative values.
The following section discusses the results with
regard to the ongoing debate on the regulation of LAWS.
Additionally, it is useful to consider not only this paper's
results but also existing limitations. This allows for an
outlook as well.

Implications
Analysis of recurrent understandings of both autonomy
and LAWS across CCW GGE discussions revealed disagreement among the parties as well as a focus on the
targeting cycle and its " critical functions " (see " Results:
Autonomy and LAWS " section). To retain MHC, it seems
helpful to reach a consensus on working definitions by
concentrating on critical functions of engaging and the
different steps of selecting and targeting. Second, looking at autonomy and LAWS understandings, parties may
need to find some common ground with respect to
empirical examples of LAWS while at the same time
being aware of risks of transparency.
Approaching underlying perceptions of humanmachine interactions again revealed the necessity to be
conscious about automation bias and blinded views
on technology (see " Results: Human-Computer Interaction " section). Even though the establishment of MHC
relies on human-centric arguments, our analysis offers
the added value of reflective positions on interacting
networks. To still be able to hold humans responsible,
a special focus on human machine " touchpoints " is
needed. The research by Brown et al. [78] allows a first

glance at the CCW GGE debate in this respect. In contrast to Verdiesen's work [67], our chosen VSD-perspective poses a first approach to the CCW GGE parties'
understanding of MHC-relevant technological features.
Even though the stakeholders might not agree on
Meaningful Human Control as a concept, the issue of
control is at the heart of the debate [20], [21]. Regarding
human control, a sufficient amount of time to intervene
was found to be one of the most important technological
requirements (see " Results: Socio-Technological Values
of LAWS " section). To incorporate this feature of humancontrolled LAWS, such weapon systems need to be
designed accordingly. It is therefore critical to avoid
automatization of firing when identification is seen as
successfully completed. Options of multi-channel communication between operator and system are already
under research [41]. Yet, with respect to assuring MHC, it
is a) important to ensure the option of intervention, while
b) active confirmation or denial of an attack might
reduce the risk of technology-biased behavior. Thus,
designing an interface offering (obligatory) yes/no-options
with regards to an attack may ensure human authority. At
the same time, values like reliability, predictability, information accuracy, or explainability might be carried by a
usable interface, offering transparent listing of target
detection processes. Situational awareness by commanders, consisting of some of the named requirements, is
crucial for the meaningful human control of autonomous
weapon systems [79], [80]. MHC can only be enforced
with sufficient training of the human operators [24].
The expert group's debate mainly focused on the
fifth phase of the targeting process, in which the mission is planned and executed, entailing the F2T2EA
cycle of " find, fix, track, target, engage, and assess "
[28]. As " [i]t is during this phase that the selected lethal
or nonlethal means will be used[,] [...] this is the phase
[the debate is] focused on the most " [28]. With respect
to the detection of a target, training of deep learning
algorithms should ensure both reliability and predictability by systematically recording and evaluating single
steps. Regarding machine learning, it may also be necessary to discuss implications of diverse degrees of
depth, i.e., representation and abstraction of layers. A
focus on testing indicates, as do our findings referring
to the importance of verification and validation of softand hardware, that procedures apart from the planning
and execution phase, on which the debate has mainly
concentrated, are critical to ensure predictable and reliable execution of the targeting process.
As our work on the debate's surrounding and influencing discourses shows, international humanitarian law
principles are still a dominant reference point for evaluation (see " Results: Influences Discourses " section). With
respect to the principles of distinction and proportionality,

IEEE TECHNOLOGY AND SOCIETY MAGAZINE

∕

DECEMBER 2020

IEEE Technology and Society Magazine - December 2020

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - December 2020