IEEE Technology and Society Magazine - December 2021 - 92

Soft Law Governance of Artificial Intelligence
Commercial FRT should primarily rely
on soft law approaches
For commercial uses, soft law alternatives (in the
context of existing, technology-neutral hard laws
on discrimination and privacy) have provided and
promise to offer a better path forward, even for concerns
about privacy and bias. Indeed, building on the
foundation of general privacy and antidiscrimination
laws (discussed below) and other than the few
state BIPA laws (mentioned above), soft law is the
biggest influence on commercial FRT deployment.
Soft law provides flexibility for use cases over
time. FRT is rapidly evolving and changing as it is
developed, refined, and deployed by individuals all
over the world. The FRT available in the next ten
years-and the uses to which it is put-will differ significantly
from the FRT available today. When hard
law fails to anticipate technological developments
and new use cases, it can hamper beneficial applications
while completely missing harmful outcomes.
Soft law allows for positive evolution because it does
not try to rigidly anticipate all uses and outcomes;
instead, it evolves alongside the technology.
Hard law also often focuses on preventing worsecase
scenarios. As humans, we are typically better at
imagining future harmful applications of something
new than we are at grasping all the many ways different
people might apply a technology to benefit their
lives. But as Adam Thierer, a Senior Research Fellow
at the Mercatus Center at George Mason University,
has said, " We should avoid basing policy interventions
on hypothetical worst-case scenarios, or else
best-case scenarios will never come about " [51]. Soft
law can better develop mechanisms to anticipate
potential harm without closing off potential benefit
and in many cases can also more quickly react to
real world harms if they occur.
Types of soft law for FRT governance
Soft law has many variants. Rather than listing
the full menu of soft law options and how they each
apply to FRT, we focus on the approaches already
governing commercial FRT today, which include
social norms, self-censorship, education, labeling,
education, voluntary standards, private certifications,
and public pressure.
Social norms
Social norms are the default governing structure
for most new technologies. Social norms and public
92
accountability are currently the dominant governance
structures for commercial FRT. As companies
and individuals deploy this technology, they rely on
cultural frameworks of acceptability. Technology
companies are integrating FRT when they work well
and have a reasonable chance of being met with a
positive consumer response. When errors do occur
or consumers negatively respond, the technology is
pulled or amended.
For example, Apple added FRT to the iPhone
X, enabling users to unlock the device, verify payments,
and use various entertainment features.
Apple announced the FRT capability well ahead of
launch with an emphasis on usability and privacy.
Today, the system works well for most people and
has generally been accepted [13]. Compare the
deployment and subsequent removal of FRT in 200
urban Rite Aid stores. Rite Aid used a system of cameras
with FRT to match customers against a photo
database of criminal suspects. When a match was
made, the security agent in the store would receive
a notification. Rite Aid did not publicize the deployment,
appeared to have thought little about the privacy
downsides and failed to make the case for how
this benefited consumers when the potential harm
to consumers was considerable. When reporters
uncovered the system, Rite-Aid disabled it to deal
with the bad press and public outcry [52].
The downsides of social norm soft law are the
delayed timeframe and the lack of clear leadership.
When concerns run high and calls for " something to
be done, " a wait-and-see-as-culture-adapts approach
is neither a satisfying way of redressing wrongs nor
politically satisfying for politicians who demand a
fix. Responsibility is diffused as well. No one specific
organization, institution, or individual is responsible,
as social norms are processes that emerge based on
the interactions of many individuals [53].
Self-censorship
Given the name, people might think that soft
law is timid or lacks real-world effect. But soft law
responses can be quite severe. Large companies
have generally shown caution in deploying FRT,
publicly declaring their intentions and working to
educate consumers on how data is collected and
used [54], [55]. Some firms delay deploying this
technology until it reaches an acceptable standard
of accuracy. For example, Microsoft is funding
efforts to include people with disabilities in AI and
IEEE Technology and Society Magazine
IEEE Technology and Society Magazine - December 2021

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