Abstract
It is evident that businesses are embracing Artificial Intelligence (AI) technologies in their day-to-day operations to make vital business decisions. However, in adopting these new technologies, there are naturally concerns which may arise that a business needs to consider. One such concern is the governance of ethical dilemmas that companies may face when using AI technologies to make business decisions. It is established in the study that to overcome this ethical governance concern each nation needs to develop its own set of ethical AI governance principles that outline exactly which ethical requirements should be adopted for each ethical concern, what constitutes ethical AI, and by whom ethical AI should be monitored. Accordingly, this study develops an ethical AI governance framework addressing these areas that can be adopted by businesses in South Africa to achieve good governance.
This study contributes to the existing body of academic knowledge by pursuing the following primary objective and three sub-objectives. The primary objective of this study is to develop an ethical AI governance framework for South African business. To achieve the primary objective the study addresses the following three sub-objectives: (1) Identify, define, and discuss AI ethical concerns that arise due to the use of AI through a systematic review of literature; (2) Identify constructs, items, and related governance processes that address each ethical concern, including the identification of who should monitor AI. In addition, develop the initial ethical AI governance framework; (3) Engage business leaders to validate the ethical AI governance framework from a Global South perspective and develop the final ethical AI governance framework for South African business leaders
To achieve the sub-objectives a pragmatist paradigm was applied. Both qualitative and quantitative information was gathered for analysis. The study is conducted in two phases, with Phase One informing Phase Two and accordingly a sequential exploratory mixed methods approach is adopted in the study. The method in Phase One involves the systematic review of literature, embedded within a PRISMA approach. During the second
iv
phase of the study, quantitative data is collated through a questionnaire positioned to business leaders. The study was also positioned firmly within a stakeholder theory approach in the development of the AI governance framework.
The results from Phase One of the study, using a PRISMA approach, revealed that there are six key concerns that need to be addressed for ethical AI governance. These include Transparency, Machine Bias, Privacy, Responsible AI, and Beneficial AI. The results from Phase One also indicate that there are various constructs and governance processes that address each of these ethical concerns. The identification and combination of the governance processes of each ethical concern leads to the development of the initial AI governance framework.
The results from Phase Two of the study validated aspects of what was identified in literature in Phase One of the study. The results indicated that South African business leaders agree that transparency, privacy, and responsible AI are key aspects to achieving ethical AI. The results also indicated that South African business leaders did not agree with some of the governance process to be included in an AI governance framework. These include: (1) Have a full understanding of previous iterations of the AI system, how they worked and why they changed, in order to audit the AI system; (2) Disclose technical limitations of the AI system to stakeholders; (3) Create algorithms with the least amount of invasiveness to data subjects; (4) Minimise the data that is required for training the algorithm. All other governance processes to be included in the final AI governance framework were validated by South African business leaders. It was also established that South African business leaders did not agree with the definition of privacy, with the subsequent removal of the following aspect privacy is the right of access to personal information from the definition. The remaining aspects were validated by South African business leaders and led to the development of the final AI governance framework for the Global South.
The recommendation of the study is that businesses should implement the AI governance framework as part of their governance processes to promote good governance in AI. The implication is that businesses will use the governance processes in this AI framework to develop good governance practices within their AI environment. In addition, it was established that the King Code does not incorporate detailed governance processes to address ethical AI. Thus, this AI governance framework could enhance the King Code by its inclusion, as detailed AI governance guidance could be provided to business leaders through the King Code.
Keywords: Artificial Intelligence, Ethics, Governance framework, Transparency, Machine Bias, Privacy, Responsible AI and Beneficial AI, Stakeholders, Business leaders.