IEEE Technology and Society Magazine - December 2021 - 61

enforcement mechanisms, the powerful economic
incentives for development of AI are likely to overwhelm
the consensus and goodwill of those participating
in wholly voluntary structures, just as the local
needs of researchers to triumph in the tenure race
prompt questionable research practices.
Authority
Absent legal controlling authorities with legitimate
and agreed upon sanctioning authority, organizations
managed through adherence to soft law can fall prey
to three maladies (at least). First, they may struggle to
establish themselves as a result of the constitutional
paradox-the absence of a mechanism to keep organizations
from consistently reconstituting themselves
in response to shifting demands and controversies
[47]. In the space of AI, membership in prior normative
organizations such as mathematical congresses
can provide a temporary solution. They can rely on a
loose agreement that the society of mathematicians'
rules are " good enough " to decide how to select a set
of proposed normative principles to guide AI.
Proliferation of statements of principles is a second
malady of soft law: in the absence of a hard law
referent, each organization can claim equal authority
to establish a set of principles or procedures as
being the best or most complete [48]. Absence of
external authorities which establish the principles
of a particular organization as fulfilling the expectations
of a regulator encourages organizations to
issue multiple alternatives. This much has been recognized
by those tracking development of AI principles
statements. As Floridi and Cowls note,
" The key questions are how, where, when, and by
whom the impact of AI will be felt. As a result, many
organizations have launched a wide range of initiatives
to establish ethical principles for the adoption
of socially beneficial AI. Unfortunately, the
sheer volume of proposed principles threatens to
become overwhelming and confusing, posing two
potential problems. Either the various sets of ethical
principles for AI are similar, leading to unnecessary
repetition and redundancy, or, if they differ
significantly, confusion and ambiguity will result
instead. The worst outcome would be a " market for
principles " where stakeholders may be tempted to
" shop " for the most appealing ones " [12].
Third, an organization that does not have a clear
source of authority can succumb to a uniquely
December 2021
pernicious form of scope creep as it grapples with
either problem of self-constitution, application of
the proliferation of principles, or even participation
in the proliferation of principles as such. Without an
authority, it is difficult for an organization to establish
an identity vis-à-vis its customers and its environment.
Without an authority to define and then
manage scope of reviews, boards can be captured
by local interests, or can lose sight of their original
purpose. This latter challenge has been leveled
against human subject ethics committees as they
have become organizations that police methodological
innovation, informed consent wordsmithing,
and check-box exercises. Scope creep is partly
the consequence of ambiguous language in the
soft law and regulatory terms governing conduct of
review boards. These boards, originally constituted
to address clinical research, were asked to review
social and behavioral research. The challenge of
providing adequate reviews for such a variety of
methods caused many of them to pursue the management
of nonphysical, identity, and informational,
risks as aggressively as physical risks. This
led to the multiplication of boards within institutions
and, in some cases, reconstitution of review
boards " in flight " to address a broader scope of
social and behavior research review [52]. As this
occurred without significant ex ante guidance or
oversight from external authorities, the landscape of
review boards has become considerably more varied
than originally envisioned. Attempts to reign in
review boards and to provide better guidance where
boards' actions had outpaced regulations, notably
through the changes to the U.S. Common Rule, took
almost three years and multiple regulatory cycles
to accomplish. Intriguingly, the changes developed
still did not address all of the areas of concern, such
as uses of data for AI, suggesting that the regulators
themselves are struggling with the ever-expanding
scope of data uses and their attendant risk and benefits
[49]-[52].
Scope
In the absence of existing regulatory authority,
which organization(s) should or could provide the
consistent, legitimate, and sanctioning authority for
AI ethics? The purpose of other forms of ethical or
professional review is to provide professional oversight
as a matter of fiduciary duty to the relatively
uninformed public, assumed to be at risk from the
61
IEEE Technology and Society Magazine - December 2021

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