IEEE Technology and Society Magazine - December 2021 - 77

Soft Law for Unbiased
and Nondiscriminatory
Artificial Intelligence
Emile Loza de Siles
Duquesne University School of Law, Pittsburgh, PA 15282 USA
University of Maryland Global Campus, Adelphi, MD 20783 USA
Technology Law Group, Washington, DC 20052 USA
 Here, tHe task undertaken is to consider and
illustrate whether soft law can be an effective and
credible means by which to ensure that artificial
intelligence (AI) and its uses are unbiased and nondiscriminatory
[1], [2].1
This is a pressing need. The current fact is that
many AI systems are designed, intended, or otherwise
to operate in biased and discriminatory ways
[3]-[5]. The most demonstrable bias and discrimination
flaw in AI today is systemic racism, gender
bias, and other corrosive societal ills that are baked
into the data on which AI operates [6]. Bias in, bias
out. That logic is inescapable.
The errors do not stop there, however. Compounding
these problems, and even more to the
point as it relates to AI soft governance and its potential
effectiveness and credibility, is the question
of the implementations and uses of AI [7]. Distinguished
from the inherent design characteristics of
the AI systems, including their training, testing, and
input data, is that their implementation is not meaningfully
transparent, that adopters and purveyors are
not held meaningfully accountable, and that AI uses
are often deeply biased and discriminatory. Worse
1The term " soft law " contrasts with the positive, or " hard, " law of statutes and
regulations instituted by governments. This article uses the term " artificial
intelligence " generally to refer to all its instantiations from chatbots to compact
neural networks and to encompass all of its data curation and other processes and
methodologies (collectively, " artificial intelligence " or " AI " ). This article refers to bias
as human and social phenomena related to potential discrimination and unfairness
and not as technical terms related to machine learning, for example.
still, such AI uses feed into and extend the reach
of power afforded by systemic racism and other
discrimination pervading human systems [8]-[14].
Beyond the intransigent stain of slavery, there are a
number of deeply unacceptable reasons why this
occurs, including automation bias, the AI " governance
gap, " technological innumeracy on the parts of
many government leaders and consumers, utilitarian
policies, and expediency [15]-[19].
These observations present lines of inquiry that
lead to two different determinations as to whether
soft law can be an effective and credible method of
AI governance.
The first line of inquiry asks whether, as a matter
of technical specification, AI systems can be
designed to be unbiased and nondiscriminatory.
The response is an emphatic " yes " based upon long
experience and an enduring faith in science and scientists,
whether it be software engineers, data scientists,
or whomever.
As privacy-by-design and security-by-design have
been conceptualized, AI systems similarly may incorporate
nonbias and nondiscrimination by design.
Among others, computer scientists Michael Kearns
and Aaron Roth posit this possibility in their book,
The Ethical Algorithm [20]-[22].
Digital Object Identifier 10.1109/MTS.2021.3123729
Date of current version: 7 December 2021.
It will take time to sort this and build workable
systems. Attorneys, sociologists, psychologists, and
others with the ability to speak or at least take the
necessary pains and constant endeavors to understand
the language and work of the relevant scientific,
technological, and mathematical disciplines
will need to be onboarded as integral contributors
December 2021
0278-0097/21©2021 IEEE
77

IEEE Technology and Society Magazine - December 2021

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