IEEE Technology and Society Magazine - December 2021 - 60

Soft Law Governance of Artificial Intelligence
the Common Rule is regulatory statute, the ethical
principles, captured in the Belmont Report, are not
law. However, this does not mean that the normative
status of research ethics is any weaker. As a set
of practices that dictate what good professionals are
obligated to do, research ethics norms are robust
[40]. Reflecting on arguments from Campbell and
Glass, " In modern times the state has become the
central authority for the regulation of professional
groups. Nevertheless, a key characteristic of professionalism,
even by today's standards, is the power
to control professional conduct privately. That is,
although the collective authority of the professional
group is now granted and delimited by the state, the
autonomy of the individual practitioner within this
group is regulated by the profession itself. In this
way, it is clear that self-regulation remains an important
reality for many professional groups, including
those in the medical community " [37].
An essential component to the self-regulation of
medical groups and other professional researchers
is that they have established mechanisms to police
one another both ex ante and ex post [41]. Ex ante
review of research through ethical review boards or
clinical trial registration is an established component
of many professions today [42]. Ex post controls,
such as peer review of publications and conference
abstracts, ensure that professional norms are
maintained throughout the research lifecycle. These
ex-post controls are supported by a set of soft law
requirements emerging from publishers and journal
editors in the form of the Committee on Publication
Ethics (COPE) norms. Ex-post controllers
also support the ex-ante controls of review boards
by requiring that research submitted for publication
be accompanied by evidence that it was reviewed.
The 360° accountability established by this ex ante
and ex post normative infrastructure is robust, but
it is fallible. As watch dog organizations, such as
Retraction Watch, show many research outputs do
not follow the professional norms expected. In some
cases, normative violations result in sanction and in
others they result (for a time at least) in wild success
(e.g., Dietrich Stappel). The inability of even a
robust ex ante and ex post series of normative controls
to effectively maintain control of the research
profession is a signal problem of highly distributed
systems controlled chiefly through soft law. Without
a regulatory body that can coordinate the systems'
implementation of research review, variations in the
60
quality of review emerge [43]. Without a hard law
issued by an authority that can cover the increasingly
internationalized space of research, it is hard
to imagine how a regulatory body might emerge that
could enforce tighter control [44].
In the field of AI, the first IEEE Ethically Aligned
Design document " recommend[s] that institutions
set up review boards as a resource to AI researchers
and developers and to evaluate relevant projects
and their progress " [45]. The expectation is that
" ... review boards have served as a valuable tool
in ensuring that researchers are able to work with
security and peace of mind about the appropriateness
of their research ... review boards provide a
valuable function in protecting institutions, companies,
and individual researchers from legal liability
and reputational harm. We recommend that organizations
setup review boards to support and oversee
researchers working on projects that aim to create
very capable and autonomous AI systems and that
AI researchers and developers working on such projects
advocate that these boards be set up " [45], [46].
The IEEE authors suggest that " review boards should
be composed of impartial experts with a diversity of
relevant knowledge and experience. These boards
should be continually engaged with researchers
from any relevant project's inception, and events
during the course of the project that trigger special
review should be determined ahead of time ... Ideally
review boards would adhere to some standards
or best practices developed by the industry/field as
a whole, perhaps through groups like the Partnership
on Artificial Intelligence " [46]. This sentiment
is revisited in Ethically Aligned Design (1st edition)
that suggests " For instance, if an ethics review board
comes in at the right time during the A/IS creation
process, it would help mitigate the likelihood of creating
ethically problematic designs " [46].
Three challenges for soft law
governance of AI review
To effectively institute a review infrastructure for
AI, a legitimate external authority that can establish
the terms of play between all of the relevant actors-
government agencies, companies, research groups,
professional bodies, and the thousands of colleges
and universities-needs to emerge. Identifying who
should hold the sword to enforce the requirements
for AI review is truly the most important question to
answer when arguing for a review system. Without
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
