AI Report on Southern Africa

		

http://misa.org

the digital revolution space.”
The AI Blueprint includes a crucial chapter on AI and ethics. It recognises that “there is a need
to ensure that ethics are integrated into AI systems to minimise structural discrimination
and bias that can arise from biased training data.”
Data Privacy and Protection Law
Section 14 of South Africa’s 1996 Constitution protects the right to privacy. South Africa’s
data privacy law, the Privacy of Personal Information Act (POPIA), received Parliament’s
approval in 2013. However, following the COVID-19 outbreak and the widespread usage
of personal data in relation to digital services, POPIA went into effect on July 1, 2020, and
became fully enforceable a year later. Also relevant is the Cybercrimes Act, which went into
effect on December 1, 2021. Hacking, unlawful interception, interference or acquisition of
data, and malicious or destructive communications are all considered criminal offences. The
Information Regulator is part of the Global Privacy Assembly (GPA).
A major concern recognised in the 2022 AU Data Policy Framework and highlighted by
several notable researchers, such as Emma Ruttkamp-Bloem and Christiaan Viljoen in the
South African context, is the lack of representation of a large portion of South African society
in data sets. According to a 2022 study, the digital divide is essentially a data barrier that
the social divide exacerbates and feeds into, potentially leading to bias and discrimination.
Emma Ruttkamp-Bloem and Christiaan Viljoen, expressed concern that at least half of South
Africans need a digital imprint in this context. The practice of data-driven AI in a country
like South Africa must be distinct from social power dynamics. Policymakers should not
underestimate the intrinsic South African challenge to fairness policies.
As mentioned earlier in the report, machine learning is influenced by existing bias in training
data and modelling procedures and the socially powered contexts in which predictions are
understood, implemented, or acted upon.
Algorithmic Transparency
Additionally, POPIA says that algorithmic transparency is achieved by “appropriate
measures must (a) provide an opportunity for a data subject to make representations
about a decision” when automated decision-making is the only method used, and “(b)
require a responsible party to provide a data subject with sufficient information about the
underlying logic of the automated processing of the information relating to him or her in
order for him or her to make representations.”
South Africa has been a trailblazer in the development of a human-centered approach at
regional and worldwide levels. South Africa has played an important role in the development
of Smart Africa’s Artificial Intelligence Blueprint for Africa, for ensuring that ethics are
integrated within AI systems and should reduce systemic discrimination and bias. The
2021 AI Blueprint represents a first step towards implementing a continental AI strategy.
Endorsing the 2022 Windhoek Statement on Artificial Intelligence in Southern Africa, South
Africa is among the pioneering countries committed to taking tangible actions to implement
the UNESCO - Organisation for Economic Co-operation & Development (OECD), South Africa
and OECD Recommendations for AI Ethics.

15

Select target paragraph3