IEEE Technology and Society Magazine - Winter 2014 - 23

recommendations here along with our rationale for
putting them forward.

Addressing Concerns about Responsibility
As already mentioned, much attention has been
given to the question whether it will be possible to
hold humans responsible for the behavior of artificial
agents of the future. Matthias [18] refers to this problem as "the responsibility gap." Artificial agents can
be programmed to learn as they operate, and because
of this learning, the fear is that no humans - even
those who program the agents - will be able to understand how some artificial agents arrive at their decisions. Hence, no human can fairly be held responsible
for what the artificial agents do.
Although the problem seems plausible as described,
the responsibility gap is generally framed as an inevitable outcome of the complexity of the technology.
This can be seen, for example, in Matthias' presentation of the problem. However, if a responsibility gap
occurs in the future, it will not be because of the complexity of the technology; rather, it will be because
of human decisions to deploy technologies without knowing how they will behave. The issue is not
whether humans will understand how the technology
operates. At some level, humans will understand; they
just won't be able to directly control or fully predict
how the agents will behave in specific circumstances.
The major concerns in this situation will be reliability and safety. Will the agents have a level of reliability and safety appropriate for the tasks performed?
Humans will make these decisions. So, if agents for
which no human can fairly be said to be responsible
are released, that will be because humans will have
decided both to trust particular learning algorithms,
and to do without social practices that assign responsibility to human decision makers.
Scholars and researchers have taken a number of
approaches to address the responsibility issues posed
by artificial agents. Several have offered rules or
informal laws for those who design and use artificial
agents to encourage the clear allocation of responsibility. For example, Miller initiated and led a collective
effort to develop a set of rules for "moral responsibility for computer artifacts" [19], [12]. The rules
state that the people that design, develop, and deploy
computational artifacts have a shared responsibility.
The rules also indicate that people who knowingly
design, develop, and deploy these technologies can
only do so responsibly when they take into account
the sociotechnical systems in which the artifact is
embedded. Moreover, they specify that people should
not deceive users about the artifact or its foreseeable
effects. Similarly, Murphy and Woods [20] developed
three laws of responsible robotics, intended as alternatives to Asimov's three rules from I, Robot. Central
IEEE TECHNOLOGY AND SOCIETY MAGAZINE

WINTER 2014

to Murphy and Woods' laws is the idea that robots
should be designed to be responsive to humans. For
example, Murphy and Woods offer as a first law that
"A human may not deploy a robot without the human-
robot work system meeting the highest legal and professional standards of safety and ethics."
Other scholars have focused on the design of technology as an important place to address responsibility
issues [16]. In this vein, some try to solve the problem by programming agents to be ethical [2] or programming them to adhere to rules constituting ethical
practices such as the rules of engagement in war [3].
As early as 1994, Weld and Etzioni [33] proposed
the idea of programming agents to avoid doing harm.
Other design approaches take the approach of designing systems to facilitate humans to act responsibly as
they control and interact with a system [9], [10].
Non-technical approaches have focused on adjusting and developing regulations and legal frameworks
for the development and use of these technologies [4].
In the case of military robots, for instance, several
scholars and non-governmental organizations have
argued for strict regulation of these technologies and
even a prohibition on the development of artificial
agents that take the human out of the decision-making
loop [14], [24], [25].
In what follows, we offer a set of recommendations
that differ from those mentioned earlier by targeting
the many different actors involved in the development
of artificial agents. We have in mind not just developers and users, but policy makers, the public, managers, financiers, journalists, manufacturing companies,
and others who influence the development of artificial
agents and the ways in which they are understood.
Addressing the responsibility issues posed by artificial agents involves attention to social conceptions and
social practices as well as the design of artifacts. Even
if developers design a technology to encourage and
facilitate responsible behavior of human users, social
practices still must be developed to reinforce or facilitate the individuals and groups acting in or with the
system to recognize and fulfill their responsibilities.

The Recommendations and their Rationale
One overarching principle precedes and permeates all
of our recommendations: responsibility for the behavior of artificial agents resides with human beings.
We believe this principle should be explicitly and
publicly recognized and should inform and guide the
development of artificial agents. We emphasize this
principle in response to rhetoric suggesting that artificial agents for which no human can be responsible
will inevitably be developed, or suggesting that the
concept of responsibility can be extended to artificial
agents themselves. Of course, we are not the first to
emphasize human responsibility. Others, including
|

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - Winter 2014