IEEE Technology and Society Magazine - June 2021 - 82

Feature
approach to responsible AI. For example, Google's
AI principles include " Be socially beneficial, " " Avoid
creating or reinforcing AI bias, " " Be built and tested
for safety, " and " Be accountable to people, " among
other principles [6]. OpenAI discusses its focus on
" Broadly Distributed Benefits, " " Long-Term Safety, "
" Technical Leadership, " and its " Cooperative Orientation "
[7]. These high-level responsible AI principles
are often vague and may host a multitude of
possible interpretations.
Yet, some companies have established relatively
clear strategies following from their principles,
such as changes to training, hiring, algorithm development
frameworks and tools, and governance
strategies. For example, Vodafone's AI Framework
provides some detail on specific actions it will take,
such as adhering to its Code of Conduct and privacy
commitments [8]. SAP proposes as part of its
Guiding Principles an AI Ethics Steering Committee
and an AI Ethics Advisory Panel [9]. IBM's Everyday
Ethics for AI provides a set of recommended actions
and questions for its employees to address key concerns
[10]. In contrast, some principles are yet to
be accompanied by clear expressions of changes
to practice. For example, documents from Tieto,
Futurice, and Salesforce emphasize abstract principles
and commitments.
The question then is why do aspirations not translate
straightforwardly into effective and responsible
practices? In fact, the process of translation is neither
automatic nor obvious. According to Mittelstadt
[11], " norms and requirements can rarely be logically
deduced... without accounting for specific elements
of the technology, application, context of use,
or relevant local norms. " Barring practical guidance
and absent " empirically proven methods... in realworld
development contexts, " claims of responsible
AI may amount to no more just that-claims.
In the best case, companies may still be in the
process of working out the details or may have
communicated their intended strategies in other
venues. Nevertheless, remaining at the " mission
statement " level and the lack of practical detail are
worrisome, and engaging in responsible AI is, in any
case, no simple task. The task of closing the principles-to-practices
gap is critical for the future of AI,
worthy of attention by companies developing AI, by
governments that might procure and deploy AI systems,
and by other stakeholders and the public more
broadly. To assist with this effort, this article reviews
82
possible explanations for the gap, depicted in
Figure 1, along with preliminary recommendations.
The six explanations we review are: 1) the incentives
dilemma; 2) the complexity of AI's impacts; 3) the
disciplinary divide; 4) the many hands problem; 5)
the governance of knowledge; and 6) the over-abundance
of tools.
The incentives dilemma
The first and perhaps most widely discussed explanation
is that the values, motivations, and incentives
that guide firms are not sufficiently aligned with
responsible uses of AI. The strongest criticisms deeply
impugn the motives of firms. For example, Greene
et al. [12] argue that companies attempt to shift
responsibility onto designers and experts to minimize
scrutiny of business decisions. On this account, companies
may be strategically promoting their principles
merely to ameliorate customer trust and reputational
concerns, thereby engaging in " ethics washing, " " ethics
shirking, " " ethics shopping, " or similar [13]. In this
way, firms can appear actively engaged regarding
AI's ethical risks in the public eye, but while framing
issues so as to minimize genuine accountability.
Alternatively, organizations need not have the
motives stated above, but may still be driven by an
Figure 1. Six explanations for the
principles-to-practices gap.
IEEE Technology and Society Magazine
IEEE Technology and Society Magazine - June 2021

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