IEEE Technology and Society Magazine - December 2021 - 63

science professionals' clashes with that of advertising
professionals, whose code trumps the other?
This brings the argument back to the question about
proliferation of principles (see above) but also to a
key problem in soft law: managing overlapping and
competing claims to legitimacy.
Risk assessment
Leaving to the side the challenges of negotiating
ethics briefly, there is another important antecedent
question to ask about the place of review in managing
AI. That is, if AI is a technology such that it
cannot be controlled, then why bother with the complexities
of governing it through soft law negotiation?
The genesis of review boards points to a particular
perspective on the control of research risk: it can
be neutralized if particular controls are inserted at
the start. The regulations of review boards assume
this: only those risks that could be identified and
those that are tractable to external management
should be included in a review boards' effort to manage
risks. This effectively narrows review to ex ante
and in medias modes of review. Those risks which
are intractable to external management include
risks to subsequent generations, to unrelated " secondary "
persons, and to " society " as these would
be an ex post review finding or an imaginative uses
of controls for anticipatory ethics concerns [57]. AI
research is different. For example, when researchers
build a data set to test a novel algorithm, they
invite dignitary harms to those persons whose data
was included before any AI research even starts. Or,
when researchers use existing data sets with novel
libraries (and undiagnosed leaky pipelines), they
invite harms against individual privacy or dignity
[58]-[60]. Thus, before AI research even begins,
there is a chance for harms, begging the question
where should a risk-based review begin?
Thinking through authority via soft law
If a decision is made about when AI research
starts and what it is, then there is a chance that we
can move onto the question concerning selection
of an authoritative body to ensure that AI review
is performed. Because of the many professions
involved, it is unlikely that a single actor can take
on this role-AI review will have to be multidisciplinary.
Furthermore, due to the confluence of professions,
it is unlikely that a single organization could
dominate this task. Instead, a set of coordinated
December 2021
organizations such as a consortia or a trust will
likely have to emerge with bottom-up buy in. There
are good reasons to look to soft law governing
organizations for examples of how this has been
done in the past.
The problems of soft law governance can also be
its solutions. The case of the Forestry Stewardship
Council (FSC) and other environmental soft law governors
provide some insights into the emergence of
actors as global normative powers. The FSC was able
to establish itself as a credible actor in a competitive
space by harnessing its relationship with the World
Wildlife Fund and pursuing a reputation as an independent
organization able to certify other organizations'
compliance with environmental principles
authored and agreed upon by many other organizations
[61]. Whichever organizations emerge to lead
in the AI space will likely model themselves after the
FSC and learn lessons about review principles, policies,
and procedures from established actors in the
research ethics space.
The abundance of organizations generating
principal statements has shown that there is a marketplace
of ideas and power in the AI space. There
are at least four soft law governance organizations
who could, if lent some additional tailoring and
buy-in from conventional state actors with coercive
sanctioning power, emerge as the de facto regulatory
body for AI: ACM, IEEE, Partnership on AI, and
Future of Life Institute. Either of these organizations
would need to be elevated as the preeminent normative
resource by powerful state players to step
ahead of their competitors. Organizations with a
different sort of buy in, such as Future of Privacy
Forum, could emerge having developed related,
niche expertise, that expands to AI. Or, players not
even mentioned here, such as the large consulting
houses (e.g., PriceWaterhouseCoopers) may work
with multiple organizations to set up similar structures
as needed for their vertical. Regardless of how
an overarching and trusted authority emerges to take
the lead on AI review structures, it is clear that they
must overcome the hurdles of acquiring substantial
capital investment and willingness to participate
from a set of players not accustomed to working
together. They will need to be willing to indemnify
themselves and their early adopters against some
potential reputational losses if the system fails [59].
These are significant, but not insurmountable, asks
that did not exist for previous research ethics review
63
IEEE Technology and Society Magazine - December 2021

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