IEEE Technology and Society Magazine - June 2021 - 77

Social perspective
Digital discrimination is not only a technical phenomenon
regulated by law but one that also needs
to be considered from a socio-cultural perspective
to be rigorously understood. Defining what constitutes
discrimination is a matter of understanding the
particular social and historical conditions and ideas
that inform it, and needs to be reevaluated according
to its implementation context. Bias in usage, as
defined above, forms a challenge to any kind of generalist
AI solution.
One complication highlighted by a social perspective
is the potential of digital discrimination
to reinforce existing social inequalities. This point
becomes increasingly pressing when multiple identities
and experiences of exclusion and subordination
start interacting-a phenomenon called intersectionality
[24]. One example is formed by the multiple
ways that race and gender interact with class in
the labor market, effectively generating new identity
categories. From a legislation perspective, antidiscrimination
laws can be applied when discrimination
is experienced by a population that shares one
or more protected attributes. However, this problem
can exponentially grow in complexity when also
considering proxy variables and the intersection of
different features [15].
On a cultural and ideological level, the call for
ever-expanding transparency of AI systems needs to
be seen as an ideal as much as a form of " truth production "
[25]. Furthermore, no standard evaluation
methodology exists among AI researchers to ethically
assess their bias classifications, as the explanation of
classification serves different functions in different
contexts [14], and is arguably assessed differently by
different people (e.g., the way a data set is defined
and curated, for instance, depends on the assumptions
and values of the creator) [26]. Conducting
a set of experimental studies to elicit people's
responses to a range of algorithmic decision scenarios
and explanations of these decisions, Binns
et al. [27] find a strong split in their respondents:
some find the general idea of algorithmic discrimination
immoral, others resist imputing morality to
a computer system altogether " the computer is just
doing its job " [27].
Although algorithmic decision-making implicates
dimensions of justice, its claim to objectivity may also
preclude the public awareness of these dimensions.
Given the differing stances on discrimination in
June 2021
society, providing explanations to the public targeted
by algorithmic decision-making systems is
key, as it allows individuals to make up their own
minds about their evaluations of these systems.
Hildebrand and Koops [22], for instance, call for
smart transparency by designing the socio-technical
infrastructures responsible for decision-making
in a way that allows individuals to anticipate
and respond to how they are profiled. In this
context of public evaluation, it also becomes
important to question which moral standards can
or should be encoded in AI, and which considerations
of discrimination can be expected to be
most readily shared by a widely differing range
of citizens [28]. Although such frameworks can
always be criticized as reductionist approaches
to the complexity of social values, keeping into
account what kinds of values are important in
society can go some way in helping to establish
how discrimination can be defined.
Ethical perspective
Finally, we need to bring in an ethical perspective;
as Tasioulas argues, discrimination does not
need to be unlawful to be unfair [29]. Yet, moral
standards are historically dynamic, and continuously
evolving due to technological developments.
This explains why law and encoded social morality
often lag behind technical developments. In light of
the discriminatory risks (and benefits) that AI might
pose, moral standards need to be reassessed to enable
new definitions of discriminatory impact. It says
that one of the famous attempts to address this question
in robotics derives from fiction: Isaac Asimov's
Three Laws of Robotics. More recently, the AI community
has attempted to codify ethical principles
for AI, such as the Asilomar AI Principles.4 However,
these principles are criticized as being vague, mainly
due to their level of abstraction, making them not
necessarily helpful [29].
More grounded and detailed frameworks for AI
ethics have recently been proposed, such as the
standards being defined by the IEEE Global Initiative
on Ethics of Autonomous and Intelligent Systems,5
which aim to provide an incubation space for new
solutions relevant to the ethical implementation
of intelligent technologies. Another noteworthy
4https://futureoflife.org/ai-principles/
5https://ethicsinaction.ieee.org/
77
https://www.futureoflife.org/ai-principles/ https://ethicsinaction.ieee.org/
IEEE Technology and Society Magazine - June 2021

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - June 2021
