Military Review English Edition May-June 2014 | Page 17

ROBOTIC WARFARE

U.S. Army Sgt. Benjamin D. Parker, an explosive ordnance disposal team leader, and Spc. Chase Donnelly, a robotics operator, prepare their robot to inspect
a suspected improvised explosive device in eastern Afghanistan’s Nangarhar Province. (U.S. Army, Sgt. Tracy J. Smith)

2011 update of the Roadmap, “an autonomous
system is able to make a decision based on a set
of rules and/or limitations. It is able to determine
what information is important in making a decision.”10 By contrast, the DOD argues, automatic
systems are fully preprogrammed. They can “act
repeatedly and independently of external influence or control,” but they “follow a predefined
path,” and their behavior has to be fully specified
in advance.11
Machine autonomy, from this perspective,
refers to robotic systems that would somehow
be more flexible and unpredictable, compared to
automated systems, in deciding how to operate—
given predefined goals, rules, or norms. Those
that make this distinction about autonomy tend to
point to artificial intelligence technologies—such
as machine learning or probabilistic reasoning
methods—as technologies that would enable
these kinds of robotic systems.12 Robots equipped
with these kinds of technologies would be able
to learn from experience and adapt to changing
MILITARY REVIEW

May-June 2014

circumstances as well as deal with uncertain or
missing data. Such descriptions of autonomy
seem to suggest that human operators as well
as developers would have less control over the
behavior of the system. The machine would not
only operate independently of the human operator,
but also, to a certain extent, independently of its
human creators.
Nevertheless, even here, autonomy does not
mean that machines are free in the decisions they
make; the conditions for making a decision are
carefully set by humans. As the DOD’s 2011 conception of autonomy shows, laws and strategies
provided by humans will still govern the behavior
of autonomous systems. The envisioned systems
could vary their behavior as long as they stayed
within these predefined constraints. Note that this
would be a remarkable feat, as it would mean
these robots could interpret laws and strategies,
applying them appropriately in ever-changing
sociotechnical contexts.
Regardless of whether this is possible, devel15