IEEE Technology and Society Magazine - December 2021 - 94

Soft Law Governance of Artificial Intelligence
Addressing specific FRT concerns
Soft law approaches such as those discussed
above can address many of the privacy and algorithmic
bias concerns regarding commercial uses of FRT.
Addressing privacy
People are concerned about the effect of FRT on
privacy. There are some lessons from early reactions
to the personal camera in the 19th century. The law
review article most often identified as starting the
U.S. commercial privacy rights discussion argued
that " Instantaneous photographs and newspaper
enterprise have invaded the sacred precincts of
privacy and domestic life; and numerous mechanical
devices threaten to make good the prediction
that 'what is whispered in the closet shall be proclaimed
from the house-tops' " [1]. Yet, social norms
addressed most of the concerns presented by the
camera. For instance, today in the United States,
taking a close-up photo of a stranger without their
permission is viewed as a rude and invasive gesture.
As a result, it rarely happens for non-celebrities. Similar
concerns were raised with camera phones in
sensitive locations like gym locker rooms. But social
norms and rules by private establishments quickly
addressed such concerns.
There is evidence that social norms around FRT
are evolving and driving other soft law approaches.
As individuals and interest groups voice concerns
about the safety, security, and privacy implications
of FRT, businesses are creating and using standards
and best practices that address these concerns. The
Security Industry Association describes how many
biometric-technology companies develop best practice
systems and provide training to prevent users
from abusing FRT technologies. They note that
industries that anticipate using FRT in their services
and products, including the American Association
of Motor Vehicle Administrators, developed FRT
standards to protect the privacy and security of the
information they collect and use [62].
Some emerging tools could augment developing
norms and help obfuscate privacy-sensitive users
from FRT technology. For example, the use of face
masks, necessitated by the COVID-19 pandemic, can
stymie the ability of FRT to identify individuals [63].
Some clothing designers have incorporated patterns
that make it difficult for FRT to detect and identify
individuals' faces [64]. Others have developed more
sophisticated technical attacks on FRT technology
94
[65]. Social norms that allow or encourage the creation
and use of such tools do not offer a comprehensive
solution to concerns but do offer another form
of feedback to shape the use of the technology.
Addressing algorithmic bias
Soft law can mitigate harms while maintaining an
innovative framework, even for critical concerns like
racial and gender bias. Completely unbiased and
accurate systems are not achievable because FRT
is dealing with datasets informed by human beings
who themselves are biased. Therefore, governance
cannot be limited to hard law. As one of the few
surveys of FRT bias research has noted, " Mitigating
algorithmic bias is a difficult task that requires a systemic,
sociotechnical and holistic perspective " [31].
However, unlike human beings, FRT systems can be
empirically tested for many kinds of bias and thus
can be improved over time. Indeed, the NIST Facial
Recognition Vendor Test is a government-funded
effort to spur this type of improvement [66]. Furthermore,
in commercial contexts, bias is usually
an unprofitable trait, and companies typically have
financial incentives to ensure that their algorithms
accurately characterized individuals. But firms will
undoubtedly make mistakes. Soft law can help mitigate
errors inherent in these systems prior to deployment
IEEE Technology and Society Magazine - December 2021

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