The advent of autonomous vehicles has been hailed by commentators as
introducing an improvement for traffic safety by promising to reduce the
overall number of road accidents as the technology matures. Some
advocates even hint at a moral imperative to structure incentives and
smooth over barriers in order to induce widespread usage of autonomous
vehicle in order to actualize this potential. The orientation towards safety
concerns, however, foregrounds the debate on crash-optimization and
imports the trolley-problem thought-experiments to the question of
autonomous vehicles.

This paper examines the potential for three types of structural
discrimination to be woven into the fabric of these developments. First,
given the emphasis placed upon decisional agency by trolley-problem
scenarios, there will be systematic privileging of the occupant vis-à-vis
pedestrians and other third-parties. Second, there is the prospect for
structural discrimination arising from the coordinated modes of
autonomous vehicle behavior that is prescribed by its code, or which is
converged upon through learning algorithms operating towards similar
goals and within similar constraints. This leads to the potential for
hitherto individuated outcomes to be networked and thereby multiplied
across fleets of vehicles. The aggregated effects of such algorithmic policy
preferences will thus cumulate in the reallocation of benefits and burdens
to certain categories of persons in a relatively stable manner. This in turn
raises the spectre of a more pernicious form of active structural
discrimination where the possibility of crash-optimization casts a
protective shield over certain individuals at the cost of third-parties.
Third, the introduction of autonomous vehicles within the framework of
crash-optimization will likely precipitate infrastructural changes, which
in a literal sense, will introduce or exacerbate structural forms of
discrimination with regard to human access to public space.