Nokuthula Olorunju: South Africa AI Regulation Needs Binding Constitutional Framework

• UCT researcher Nokuthula Olorunju advocates for a binding, constitutional AI regulatory framework in South Africa to protect human rights and manage risks.
• Her research highlights that South Africa's current "wait-and-see" approach and non-binding ethical principles are insufficient to address the rapid growth of AI.
• Existing regulatory fragmentation, weak implementation, and legal silos create opportunities for AI misuse and regulatory uncertainty.
• Olorunju proposes a hybrid constitutional approach to clarify mandates, establish interdisciplinary oversight, and avoid duplication in AI governance.
• The framework prioritizes human rights, aiming to mitigate risks such as AI-driven bias, privacy violations, and the amplification of historical inequalities.

The government, therefore, bears the responsibility to implement enforceable laws that close these regulatory gaps and prevent the exploitation of rights through AI.

South Africa must transition from its current "wait-and-see" stance on artificial intelligence (AI) regulation to establish a binding, constitutionally-rooted framework, according to research by Nokuthula Olorunju, a PhD graduand and researcher at the University of Cape Town (UCT). Her thesis, titled "Artificial intelligence, bias and the rule of law: A constitutional approach to the regulation of AI in South Africa," argues that such a framework is essential to safeguard human rights without impeding technological advancement. Olorunju, who will receive her PhD in Law (Commercial Law) on September 8, emphasizes that the rapid expansion of AI means the country can no longer afford to delay implementing an effective regulatory system.

This UCT AI regulation research comes at a critical juncture as South Africa works on developing its national AI policy. The urgency of Olorunju's findings is underscored by a recent setback in April, when Communications and Digital Technologies Minister Solly Malatsi withdrew the nation's draft National AI Policy. This withdrawal occurred after internal reviews revealed that the document's reference list contained fictitious sources, compromising its integrity and credibility.

The current regulatory landscape in South Africa, characterized by a "wait-and-see" approach, is deemed insufficient by Olorunju to manage the escalating complexities and risks associated with AI. She highlights that legislative fragmentation, weak implementation, and persistent legal silos create fertile ground for AI misuse. This fragmentation can lead to overlapping jurisdictions, uncertainty regarding authorization processes, conflicting mandates among governmental bodies, and general confusion about which institutions are responsible for specific aspects of AI governance. As Olorunju succinctly puts it, "if a gap can be exploited, it most probably will be."

Furthermore, the research asserts that existing legislation, non-binding ethical principles, and self-regulation are inadequate to address the full spectrum of AI-related risks. While ethical guidelines and self-regulatory measures have helped fill some gaps, they lack the legal enforceability required for comprehensive protection. Self-regulation, in particular, can allow organizations to selectively adopt principles that align most conveniently with their existing missions and values, potentially overlooking critical areas of concern. The government, therefore, bears the responsibility to implement enforceable laws that close these regulatory gaps and prevent the exploitation of rights through AI.

To overcome these challenges, Olorunju proposes a constitutional approach, envisioned as a hybrid framework designed to bridge existing regulatory gaps and address practical enforcement and implementation hurdles. Key components of this approach include clarifying conflicting mandates among regulatory bodies, establishing an interdisciplinary AI oversight body, and actively avoiding unnecessary duplication of governance structures within the South Africa AI constitutional framework.

Central to Olorunju's proposed framework is the protection of human rights, especially given AI systems' capacity to reproduce and amplify existing societal inequalities. Her research identifies a range of critical risks, including AI-enhanced cybercrime, data breaches, privacy violations facilitated by deepfakes, and the potential for racial and gender bias in lending and employment decisions when AI systems are not trained on representative South African data. Other concerns include copyright infringement and mass surveillance. The study underscores that AI systems can perpetuate historical inequalities if the data used for their training reflects pre-existing societal biases. Given South Africa's unique history, the perpetuation of racial and gender bias and the exponential increase of the digital divide are particularly pressing concerns that demand robust AI human rights South Africa considerations.

Lawyers and compliance officers in South Africa should anticipate the imminent development of a binding AI regulatory framework, moving beyond current non-binding principles. This research signals a shift towards enforceable laws, necessitating proactive review of AI governance strategies to mitigate risks related to bias, data privacy, and human rights, and to prepare for new compliance obligations.

Source: Original reporting based on UCT research by Nokuthula Olorunju.