IEEE Technology and Society Magazine - December 2021 - 88

Soft Law Governance of Artificial Intelligence
likely effects of new soft or hard law on two distinct
domains: commercial FRT uses and government
FRT uses. We describe FRT technology, identify generalized
benefits and costs of FRT, and then explore
risks and potential governance solutions in both the
commercial and government uses. We tentatively
conclude that soft law governance will better enable
benefits and address risks of FRT in commercial
uses, compared to FRT-specific hard law. Our
stronger conclusion, however, is that government
use of FRT will in many cases need additional hard
law constraints.
Unfortunately, many critics of FRT focus on
regulation of commercial uses and fail to distinguish
commercial uses from government uses. This
increases the risk of government abuse of the technology,
but also threatens the beneficial commercial
applications. In fact, unless government FRT uses are
subject to targeted and clear hard law constraints,
soft law approaches to commercial FRT will struggle
to gain legitimacy.
What is facial recognition technology?
FRT is a subset of computer vision that focuses on
processing images of human faces. This technology
can be subdivided into three related categories of
analysis [9], [10]. Facial detection is the ability of
software to recognize that an image contains a picture
of a human face and usually includes the ability
to demarcate where within the image the face or
faces are. Sometimes facial detection includes identifying
facial features as well, such as eyes, noses,
and mouths. They also may build wireframe mesh
models of the 3-D representations of a face from the
input images. Facial detection algorithms may be
trained on large datasets of faces, but when running
they do not require such databases. They can also
operate in real time, even on mobile phones. For
example, Snapchat or Instagram filters that manipulate
video-giving a photograph sunglasses or rabbit
ears-use facial detection algorithms [10].
Facial verification builds on facial detection to
determine whether two different images contain a
picture of the same person. Facial verification algorithms
seek to do this even for photos with different
formats, angles, and lighting. A facial verification
algorithm may rely for its input on a facial detection
algorithm that identifies and extracts features from
faces in two different photos; the facial verification
algorithm then seeks to establish whether the faces
88
are the same. Like facial detection, facial recognition
algorithms do not need large databases of photographs
to operate, although the algorithms may
have been trained on such large databases [9].
Finally, there is facial identification. Facial identification
is the process of, given an image of a face,
identifying which face in a large collection of pictures
of other people that best matches the given
image. In concept, this is as simple as running a
facial verification algorithm against a large library
of recorded faces. Unlike facial detection and facial
verification, facial identification algorithms require a
database of photographs of all potential matches-
or a database of facial models derived from such a
database of photographs. Sometimes this database
of photographs is tied to the identities of actual
humans. This is what many people mean when they
colloquially say, " facial recognition " [9], [11].
Each of these subsets of facial recognition technologies
require different amounts and kinds of
data about individuals and enable different uses
(or abuses). Thus, the specific policy challenges for
each category are different.
FRT offers significant benefits
FRT is already in use and its potential future offers
attractive benefits. While most commercial applications
only emerged in the past five years, FRT is
already a widely deployed technology [11]. Market
research indicates global revenues of $3-5 billion
from 2016 to 2019, projected to grow to $7-10 billion
between 2022 and 2024 [9].
This initial success is because FRT can be useful,
convenient, and fun. For example, Apple's Face ID
uses facial verification to enable iPhone users to
easily unlock their phones [12], [13]. Google Photos
uses FRT to tag photos by individuals photographed,
making it easy to, for example, search for all the pictures
of you and your Aunt Sally together [14]. Los
Angeles International Airport has deployed FRT to
cut boarding times in half and Parker Adventist Hospital
in Colorado used FRT to confirm that cancer
patients are receiving the correct treatment [11]. As
mentioned previously, photo filters on popular apps
like Facebook, Snapchat, Instagram, iMessage, and
many others use facial detection [13]. The same
technologies that can detect a face for security reasons
can be employed to overlay cosmetic changes
for users merely by distinguishing a face from other
parts of the body or the environment within the
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
