IEEE Technology and Society Magazine - December 2021 - 38

Soft Law Governance of Artificial Intelligence
are all highly uncertain, " traditional legislative and
regulatory approaches do not work well [1]. Governments
are also reluctant to enact laws and regulations
that might impede innovation and hinder their
ability to attract businesses. Additionally, AI broadly
impacts many industries, products, jurisdictions, and
stakeholders in varying contexts, further complicating
determinations of not only how to regulate AI,
but who will regulate it.
Soft law is intended to fill the AI governance
gap, at least in the short term. It creates substantive
expectations that are not directly enforceable by
governments [2]. Soft law programs vary widely in
form and format. Examples include codes of conduct,
ethical statements, professional guidelines,
statements of principles, certification programs,
private standards, public-private partnership programs,
and voluntary programs.
The benefits of soft law include that it can be rapidly
adopted and revised because it resides outside
of traditional government bureaucratic processes.
This fosters simultaneous experimentation with multiple
approaches. Additionally, soft law programs
are not limited by narrow delegations of authority or
restricted to specific legal jurisdictions.
The weaknesses of soft law begin with it being
unenforceable, at least in a traditional sense. Instead,
soft law relies on other mechanisms to achieve
desired activities and outcomes. It is also problematic
when soft law is developed behind closed
doors, lacking the transparency and opportunities
for participation afforded by traditional governmental
processes. Another risk is that some organizations
may engage in " ethics washing " or " greenwashing " -releasing
or agreeing to ethical principles but
not changing their behavior. Similarly, " good actors "
might comply, while " bad actors " might not. Additionally,
because soft law is often expressed in vague
terms, it may be difficult to measure compliance.
Governing AI for legal services: Start
with evaluating effectiveness
A significant number of objectionable uses of AI
could be prevented if effectiveness evaluations were
widely adopted and implemented during all project
phases: design, development, deployment, and
maintenance. Effectiveness evaluations could help
detect various AI problems. For example, bias can
be revealed by evaluating effectiveness across categories
of people, as illustrated by Buolamwini and
38
Gebru's [3] " Gender Shades " study of facial analysis
technology. AI effectiveness evaluations can also
expose violations of fundamental best practices,
such as using training data that is not representative
of the environment in which the AI will be deployed.
Early effectiveness testing could head off AI projects
based on junk science, such as attempts to breathe
life into phrenology.
Effectiveness evaluations do not end the inquiry
in all cases, of course. For example, opinions
about effective facial analysis technology may differ
depending on whether the proposed use is by
a person who is blind to identify friends or by governments
for surveillance. Other principles, such as
fairness, accountability, and transparency, should
also be considered. At the same time, effectiveness
testing can inform our application of these principles
and help rid society of AI that does not work
and could be harmful. Thus, the evaluation of AI for
legal-services delivery ought to begin with an evaluation
of its safety and effectiveness.
Responsible technology adoption for
legal services: Increasing access to
legal services and justice?
About 5.1 billion people worldwide are deprived
of justice [5]. Of these, 4.5 billion people are
excluded from social, economic, and political
opportunities that the law provides; 1.5 billion people
have justice problems they cannot solve; and 250
million people live in extreme conditions of injustice.
These problems are not limited to developing
countries. In the United States, it is estimated that
more than 80% of the impoverished and more than
50% of the middle-class lack access to legal services.
Even many businesses lack efficient, effective solutions
for their legal needs, especially as complexity
grows in our interconnected, global, and increasingly
digital economies.
The
rapid
advancement of
computational
technologies presents abundant opportunities to
improve legal services, legal systems, and the law
itself. Additionally, the COVID-19 pandemic has
accelerated legal-industry technology adoption.
Michigan Supreme Court Chief Justice Bridget
McCormack says that the pandemic should cause us
to ask: " Why is our system of justice held together
with the threads of 20th century technology and 19th
century processes? " [11].
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
