Canada has added its voice to the global chorus of data
protection and privacy commissioners calling for fairness, transparency and
privacy by design as “core values” in the development of artificial
intelligence by co-sponsoring the Declaration on Ethics and Data Protection in
Artificial Intelligence.
The declaration was adopted by the 40th International
Conference of Data Protection and Privacy Commissioners on October 23.
The declaration, co-written by data protection and privacy
commissioners from France, the European Union and Italy, was signed by the
Office of the Federal Privacy Commissioner and its provincial counterpart from
Quebec in addition to 12 other regulators from Argentina, Hong Kong, Mexico and
the Philippines.
While lauding the significant benefits that AI systems could
offer for users and society in general, the preamble to the declaration
acknowledged that some data sets used to train machine learning-based and
artificial intelligence systems have been found to contain inherent bias resulting
in decisions that can unfairly discriminate against certain individuals or
groups.
The preamble also noted that the decisions of AI systems
decisions that cannot be explained also raise fundamental questions of
accountability, not only for privacy and data protection law but also for
liability in the event of errors and harm to individuals. Given ongoing
concerns about the possible malicious use of AI and related risks to privacy
and data protection, the IDPPC felt it necessary to urge the adoption of
international standards and created the declaration to endorse some key
“guiding principles” as its core values to preserve human rights in the
development of AI.
These six guiding principles include the following.
Fairness
All AI and machine-learning technologies should be designed,
developed and used in accordance with the fairness principle — consistent with
their original purpose and any data collected for use with such AI systems used
in a way that is not incompatible with the original purpose of their
collection.
AI systems should also be developed in a way that
facilitates human development, rather than obstructing or endangering it and,
if required, boundaries on certain uses may be required.
Continued attention and vigilance
There must be accountability for the potential effects and
consequences of AI systems. Such accountability includes relevant stakeholders
to individuals, supervisory authorities and other third parties as appropriate
as well as the use of audits, continuous monitoring and impact assessments. The
declaration stressed the need to invest in awareness raising, education and
research on AI in order to better
understand AI and its potential impacts on society and "demonstrable
governance processes" for relevant actors, including trusted third parties
and independent ethics committees.
AI systems transparency and intelligibility
The declaration called for improvements on AI systems’
transparency through a variety of means, including investing in public and
private scientific research on “explainable” artificial intelligence, making
organizational practices more transparent (by promoting algorithmic
transparency and the audit-ability of systems and the provision of meaningful
information) and ensuring that individuals are always informed appropriately
when they are interacting directly with an AI system or when they are providing
personal information to be processed by such systems (informational
self-determination).
Ethics by design
The declaration stressed that AI systems have to be designed
and developed responsibly from the very start, applying the principles of
privacy by default or privacy by design. Practically, this includes
implementing adequate technical and organizational measures and procedures
(proportionate to the type of system being designed or implemented) to ensure
that data subjects’ privacy and personal information are respected. Also,
developers should be assessing and documenting the expected or potential impact
on individuals and society at large both at the beginning of any AI project and
during the project’s entire life-cycle and identifying specific requirements
for fair and ethical use of such systems.
Empowerment of individuals
While the use of AI is to be encouraged, it should not occur
at the expense of human rights or the rights of individuals. This includes
respecting data protection or privacy rights — including rights to access, the
right to object to processing and the right to erasure — and guaranteeing an
individual’s right not to be subject to a decision based solely on automated
processing if the decision significantly impacts them. Regardless, individuals
should always have the right to object or appeal and challenge decisions
generated through the use of AI systems.
Unlawful biases or discrimination
The declaration expressly acknowledges concerns relating to
unlawful bias or discrimination that may occur from the use of data in AI and
such unintended results must be reduced and mitigated. Accordingly, developers
should invest in research into technical ways to identify, address and mitigate
bias, taking reasonable steps to ensure that the personal data or information
used in automated decision-making is accurate, up to date and as complete as
possible and providing specific guidance and principles in address bias and
discrimination, promoting the awareness of individuals and stakeholders.
Recognizing that the declaration’s guiding principles are
necessarily pitched at a high level, the ICDPPC also called for the creation of
common governance principles on AI to be established on an international basis
— given the fact that the development of AI is cross-border activity that will
impact everyone.
As part of the declaration, the ICDPPC, therefore, also
established a permanent working group — the working group on ethics and data
protection in AI — that is now tasked with creating additional guidance to
accompany the principles articulated in the declaration.
The group intends to work with all relevant parties involved
in the development of AI systems, including governments and public authorities,
standardization bodies, AI systems designers, providers, researchers, companies
and end users of such systems and will periodically report back to the ICDPPC
on its activities.
While there is no doubt that the principles of the
declaration are couched in lofty language, they represent a reaction to some
very real concerns of individuals that in the rush to seize and capitalize upon
the benefits afforded by the use of AI systems — including the ability to process
large amounts of data, improving efficiencies and the development of
deep-learning technologies — the ethical and human rights must not and cannot
be left behind.
Given the landmark work conducted on ethics and AI in this
country, spearheaded in Quebec (and further discussed in a prior column) Canada
remains well positioned to continue to keep the issues of ethics and data
protection at the forefront of AI development.
Lisa R. Lifshitz
Partner, Torkin Manes LLP
Sonu Dhanju-Dhillon
Partner, Torkin Manes LLP