South Africa’s artificial intelligence policy — a small step forward, but gaps remain

Artificial intelligence (AI) continues its exponential growth, capturing the imagination of the public, industry and investors alike. Predictions about AI’s impact range from promises of massive economic growth and efficiency gains to warnings of job displacement and existential threats.

In response, many governments and international organisations have rushed to develop AI policies. Unfortunately, until now, South Africa has been notably absent from this group.

However, without much fanfare, the Department of Communication and Digital Technologies (DCDT) released a National Artificial Intelligence (AI) Policy Framework in mid-August. Described as the “first step” in creating a National AI Policy, this framework aims to lay the groundwork for future AI regulations and possibly an AI Act in South Africa.

The framework serves as a strategic blueprint for harnessing AI, with a focus on ethical guidelines to promote economic growth, technological advancement and societal well-being in line with South Africa’s values and priorities.

Notably, it allows for the development of tailored strategies that leverage AI to drive innovation and efficiency in each sector, ensuring that the unique challenges and opportunities of each industry are considered.

At just 13 pages, the framework is relatively concise for a policy document. It outlines 12 strategic pillars for South Africa’s AI policy: talent development and capacity building; digital infrastructure; research and innovation; public sector implementation; ethical AI guidelines; privacy and data protection; safety and security; transparency and explainability; fairness and mitigating bias; a human-centred approach; professional responsibility; and the promotion of cultural and human values.

The AI policy framework can be evaluated from both a content and procedural perspective. Substantively, the framework’s high-level themes and focus areas are ambitious but comprehensive and logical.

The focus extends beyond the technology and its value to consider its socioeconomic and ethical impact. The document acknowledges both the benefits and risks of AI, and notes South Africa’s marginal position in the global AI landscape. It wisely accepts that a one-size-fits-all approach is not feasible, advocating for sector-specific requirements.

However, much work remains to flesh out the framework into actionable steps, timelines and key performance indicators, which will likely be detailed in the final national policy. Only then can the feasibility of the framework’s strategic pillars be properly assessed.

The more pressing issue, however, lies in the procedural shortcomings of the framework.

First, it lacks a clear methodology and offers no attribution beyond the DCDT. There is no explanation of how the document was developed or how the strategic pillars were chosen.

While these choices are not necessarily problematic, a robust policy document should include an open and understandable methodology, a basic requirement of any piece of academic, scientific or evidence-based writing.

Additionally, the document provides little information about the authors or the consultation process, leaving readers in the dark about the expertise, credibility, biases or motivations of those who drafted it.

The reference list is also relatively outdated, with two-thirds of the sources from 2020 or earlier, a shortcoming in a field that progresses rapidly.

Second, it is unclear how the DCDT plans to solicit and incorporate feedback from a broad range of stakeholders. As of 1 September 2024, the AI Framework is not available on the DCDT’s website or social media accounts, and there is no publicly available plan for how the document will be shared with citizens, academia, civil society or businesses, nor is there a timeline.

Currently, the document was only released via media and appears to be a top-down initiative imposed on key stakeholders, contrasting with research suggesting that a national AI policy should, at least partly, be developed through a bottom-up approach to ensure it reflects the issues of major stakeholders and gains wide legitimacy.

Last, the framework does not seem to align with previous domestic or international policy efforts. Notably, it makes no mention of the DCDT’s AI Planning document released earlier in 2024, which outlined various activities and timelines for developing an AI policy. It is unclear what the status of this document is, if any.

The framework also overlooks the former DCDT minister’s announcement that an AI Expert Advisory Council and the Artificial Intelligence Institute of South Africa (AIISA) would drive the policy forward.

This omission makes the government’s AI policy initiative seem fragmented and disjointed. Furthermore, there is no effort to harmonise South Africa’s policy with international efforts, such as the African Union’s AI Strategy or Unesco’s Recommendations on the Ethics of AI.

While the framework is a commendable milestone towards an AI strategy, its procedural aspects need greater clarity and robustness.

Moreover, there is a need to speed up the process and ensure it is inclusive. Unlike previous general-purpose technologies such as the internal combustion engine, electricity and the internet, which took decades to diffuse across the economy, AI will not give society much time to adapt.

South Africans, including labour, business and civil society, should seize the opportunity to provide input on the framework to ensure that the final AI policy reflects the considerations and values of all sectors of society.

On the other hand, the government needs to ensure the final policy captures AI’s benefits while mitigating risks – for all segments of society. DM