IEEE Technology and Society Magazine - June 2021 - 62

Feature
challenge, but in no case should these kinds of applications
be ruled out from an AI regulation.
Governance and regulation
alternatives
The EU regulatory efforts in other digital domains
like platform economy suggest that alternatives to
top-down legislation like self-regulation or coregulation
methods are used. These different types of
approaches are deeply discussed in [22].
Self-regulation might be an important part of a regulatory
process along with conventional legislation
efforts, but this new governance approach should
assure that all relevant actors are invited to participate,
not only industry providers. According to the
human-rights approach, the self-regulatory bodies
and fora should not consist of meetings where industry
actors discuss issues of trust, safety, and security
of citizens without the citizens.
Governance and regulation of algorithms are
important issues that need a broad consensus where
a democratic debate should take into consideration
the opinion of the broader community. Ethics must be
introduced into the governance structure, and command-and-control
regulation should not be excluded.
Activities like profiling and microtargeting, that are
posing serious threats to political and cultural pluralism
[23], would very probably not be tackled in a
pure self-regulated body composed of just corporate
members.
Data privacy regulations, already in place in
several countries worldwide, are already touching
upon algorithmic activity, but not in a complete
manner. Although the GDPR establishes the
rules concerning algorithms using personal data,2
there are some other algorithms using other types
of data, which badly handled, can affect people's
rights. GDPR is not enough for a human-rightsbased
algorithms system.
Self-regulation is one of the alternatives, and this
is where actually we are now. We should think if it is
2 As a matter of fact, Art. 22 (1) GDPR does prohibit automated decision-making
by stating that any " data subject shall have the right not to be subject to a
decision based solely on automated processing, including profiling. " However,
the GDPR room for maneuver in the area of ADM systems is quite restricted.
Regarding group-related and societal goals such as nondiscrimination and
participation, the GDPR has little to offer. For this reason, there is a need to
discuss complementary regulatory tools beyond the GDPR. The application
of the GDPR's prohibition of ADM systems is, for a variety of reasons, closely
restricted. The GDPR prohibits only fully automated decision-making. Systems
which " only " prepare the basis for human decisions and give recommendations
may still very well be used. For the prohibition to come into effect, ADM systems
must make fully automated decisions on the basis of personal data, plus the
decisions must have legal consequences or similarly affect the data subject
significantly. If one of these three criteria is missing, the ADM-specific provisions
in Art. 22 GDPR do not apply.
working. Apparently not: Big Tech American companies
are publishing their AI generic principles without
specifying clearly an agreed upon set of actual
checks to certify compliance to these principles. In
some cases, notable examples were found to contravene
their own set of principles.
These inconsistencies denote the disconnection
between ethicists in corporate committees and
product development and rollout teams. It is essential
that a multidisciplinary approach including
ethics is at the core of the company's decision system.
" Integration of expert knowledge, particularly
from the social sciences and including qualitative
methods, can help overcome some limitations. This
knowledge can be used to, for example, inform the
modeler of variables that might be influential but
unobserved " (two independent sets of unobserved
quantities, structural racism and jury racism, can
unfairly lead to a guilty verdict) [24].
Some very authorized voices (Vinuesa,
Azizpour, Leite, Balaam, Dignum, Domish, Fellander,
Langhans, Tegmark, and Nerini [25]) have
stated " we believe that it is imperative to develop
legislation regarding transparency and accountability
of AI, as well as to decide the being pushed
forward by initiatives such as the IEEE Ethical
Aligned Design and the new EU Ethical Guidelines
for trustworthy AI. "
The EU framework for achieving trustworthy AI
[26] consists in the following:
* An ethical purpose: It has core values and principles
that all those dealing with AI should comply
with. These are based on international humanrights
law, which at EU level is enshrined in the
values and rights prescribed in the EU Treaties
and in the Charter of Fundamental Rights of the
EU. The " ethical purpose " of developers, deployers,
and users of AI should consist of respect for
the rights, principles, and values laid out therein.
* Realization of trustworthy AI: Mere good intentions
are not enough. It is important that AI developers,
deployers, and users also take actions
and responsibility to actually implement these
principles and values into the technology and
its use. Moreover, they should take precautions
that the systems are as robust as possible from
a technical point of view, to ensure that-even
if the ethical purpose is respected-AI does not
cause unintentional harm.
62
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
