IEEE Technology and Society Magazine - March 2015 - 69

particular kind of artificial intelligence (AI) agency to help
solve. When cars have the potential to be "social" (even
between themselves as machine-to-machine) there exists
a high likelihood for fragmentation due to PoSR-related
multiplexing. In hardware terms, the ability to parse multiple messages in an automobile is certainly possible. Sensors could be added to handle needed functionality and a
processor could be dedicated to each subsystem promoting undivided attention other than a few input settings.
Software is another matter entirely.
Agency is the capacity to make nondeterministic choices from a set of options as events unfold. For example,
humans exercise agency in a car when deciding to run a
red light, or not, or to turn left and visit friends on the way
to the store. Since options are known possible contingencies, the capacity to exercise these as choices depends on
having the sufficient skills and knowledge to do so. People
generally can foresee more options than they have ability
or confidence to exercise in any given situation. A potential
benefit of technology is that it can create more options,
and by extension, make it easier for people to exercise
those options. However, as technology changes, so do its
benefits and the options that are created within them.
Technology expands agency in particular directions,
but can lead to loss of earlier enactable options if the
earlier options depend on prior knowledge, skills, and
technology that may be replaced by newer options.
The foundation of a social relationship requires the
presumption of agency on the part of the other. Otherwise, it is not a social relationship, but instead a technical relationship where the right knowledge and action
will result in a given behavior. Social relations require
that each party understand that the other can exercise
agency. The nature of the relationship will impact, but
not determine, how each party's agency is enacted.
The effectiveness of social relationships is based more
on trust and reciprocity than specific actions. In return
people gain access to the knowledge and skills of others,
increasing their exercisable options and thus expanding
their agency beyond the capacity of their own knowledge
and skill, as well as increasing their capacity to contribute to simultaneous tasks.
Presently cars do not have agency. Anti-lock brakes,
airbags, and other seemingly automatic features mimic
agency in cars, but those are based on decision trees,
fuzzy logic, or local rules, and will not scale to the multiplexed environment of a truly social car that can take
over aspects of driving which presently require human
agency to perform.
The capacity for both human and machine to make
genuine choices from a rich set of options adaptively
as events unfold is the ideal outcome for a driving environment where events can be unpredictable. In other
words, in the case of people within cars, the combinaMARCH 2015

∕

tion of humans/cars needs to have relatively successful
outcomes in order to avoid accidents. The discussion of
agency applied to PoSR and the social car comes into the
fray not so much because we are concerned with the specific agency of an individual person or car, but because
to exercise agency in a social context, understanding
that others have agency and a context for that agency is
essential to an individual applying their own agency. We
are suggesting that to be successful, the social car will
require an AI that has operational agency in the sense
that its own decisions are in part based on the presumption of agency in other vehicles (human or mechanical),
and thus that the AI has the capacity to make judgments
based on trust and reciprocity rather than relying on
behavioral extrapolations alone. Bojic et al. [16] highlight
some of the issues and problems relating to integrating machines into social networks [16, p. 89]. However,
although they include agency in their argument, they
assume this is imposed from outside the social network,
whose purpose is to realize this external agency, which is
driving the global process as a series of distributed local
processes. Manzalini et al. anticipate the need for distributed agency-driven context awareness [17]. We argue that
if there is more than one agency at play, necessarily these
local processes must include a presumption of agency on
the part of all interacting systems in order to resolve the
often conflicting goals of different agencies.
Problems can emerge from having to manage any
loss of information that comes about by distributing
messages across too many non-intersecting networks
(an outcome of PoSR). When people do not know
enough of the context relating to the people they are
communicating with, there is a greater likelihood of
creating unwarranted inferences. When communicating
in person, people infer things based on many inputs,
including observations, content of the communication,
past experiences with the other party, and context,

Personal
Social
Physical
Figure 2. Meta-multigraph with individual nodes situated in
contextual sets.

IEEE TEchnology and SocIETy MagazInE

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - March 2015