IEEE Technology and Society Magazine - December 2021 - 62

Soft Law Governance of Artificial Intelligence
conduct of professionals. Each of the ethics review
boards in common use assesses research to reduce
the probability or magnitude of harm from actions
with an uncertain risk profile [53], [54]. An organization
with sanctioning authority necessary to support
an AI ethics review board will need to have authority
to speak from the position of a fiduciary to the public
and be able to reasonably respond to the public
desire to reduce harms in the face of uncertain risks
[55]. It will also be necessary that there be a definitive
scope of their fiduciary duties versus duties they
have to their profession or to their firms.
Authority to regulate actions of specific professions
has historically been granted to similar
professionals with distinct training and associated
trappings of professionalism, such as continuing
education, licensure, and certification by recognized
bodies. Presently, no such professional body
exists for AI. Likewise, it is not clear that there is a
fiduciary duty that can be attached to the whole
gamut of professionals that conduct applied AI
research or who develop AI-powered products,
much less market them for public use. The myriad
of professionals that are engaged in this technology's
development or research presents a challenge
for defining the scope of review. If it is AI review
for research, then a distinct set of professionals
should be included: computer science researchers,
domain-specific researchers, high-performance
computing cluster managers, cybersecurity, and the
many professionals involved in managing research
(e.g., import-export experts) along with the existing
research review infrastructure [56]. If it is AI review
in a production setting, then another set of professionals
should be included: marketing and sales,
user interface/user interaction, app developers, data
architects, data engineers, and so on.
Such an expansion of the set of professionals
would lead to concerns about inclusion of appropriate
components for the review. For example,
software engineers will envision review components
through the lens of the Software Engineers' Body of
Knowledge (SWEBOK), but data architects might
envision review through the lens of Lambda or Kappa
architectures. Other components to be considered
by AI review boards include: risk assessment for
physical safety, psychological safety, physical, and
cyber security; data valuation and the concurrent
methods to estimate the risks of data loss; methods
to evaluate privacy intrusions from uses of data; and,
62
finally, assessments of the risks that emerge from
particular AI models (e.g., reinforcement learning).
Once the experts in these fields are also included, it
becomes clear that the professional stakeholders to
AI review make up an unmanageably large group.
Add this to the sheer number of professional stakeholders
whose fields are not even governed by soft
law mechanisms and constitute a set of maturing, or
even ephemeral fields, then it is not clear how certain
problems of legitimacy and authority mentioned
above, such as the problem of constitution, could
be resolved favorably or expeditiously. Finally, this
count of stakeholders does not include each of the
subject matter experts (e.g., oncologists and ichthyologists)
or affected users. When everything and
everyone seems to have a stake in AI, who could
decide and how could they decide who to include
in an AI review board?
The matter of personnel scope is also associated
with the challenge of defining what forms of AI ought
to be reviewed. What is unusual for the particular
case of an AI review board is that the field of what is
recognized as AI is also strongly variable by an even
bigger set of institutions than universities and hospitals,
leading to multiple and overlapping sectors of
uncertainty. An effective soft law governor, whether
supported by discrete hard laws or not, will have to
contend with these overlaps when leading implementation
of review board standards. The global
landscape of emerging hard law shows the challenge
of this-what the European Union has defined as AI,
high-risk AI, and trustworthy AI is different than what
the U.S. state of California is attempting to define as
AI and how Singapore defines AI that is well governed.
As it stands in July 2021, any AI review board
will have to sit quite uncomfortably between Silicon
Valley, Brussels, and Singapore.
Perhaps, chief among the challenges facing any
authority that emerges as a governor capable of
fielding a credible AI review mechanism is enforcement
of soft law mechanisms that have historically
permitted or even encouraged a wide range of institutional
variations. Key questions here are " whose
code of ethics should prevail? " and " is there a limit
to the scope of duties and ethics for a method like
AI? " Assuming that each of the myriad professionals
that could be included on an AI review board
are included, then how will they manage the task
of deciding on the ethical and technical scope
of review? When the code of ethics for computer
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
