SA must refine its own AI future and secure algorithmic sovereignty

SA must refine its own AI future and secure algorithmic sovereignty

4 September 2025 - Benjamin Rosman

The country stands at a crossroads. It is essential that we and other African countries build the capacity to refine our own data.

While it is commendable that South Africa is committed to securing our digital resources through data sovereignty policies such as the Protection of Personal Information Act, or Popia, there is an urgent next step: to attain algorithmic sovereignty.

South Africa and other African countries need to build the capacity to refine their own data. This is what we call algorithmic sovereignty - and it is not merely desirable; it is essential.

Algorithmic sovereignty is the capacity to design, develop, deploy and govern artificial intelligence (AI) systems locally so that they serve domestic needs, laws and values.

This is more than just data security and privacy protection. It is about having the technical skills and institutional resources to build the systems that transform data into meaningful insights. Without these capabilities, we become little more than exporters of raw digital resources, dependent on foreign companies and countries to refine our data for us, which we, in turn, import at high cost and poor local fit.

Take large AI models such as OpenAI's GPT, Google's Gemini, or Meta's LLaMA. These impressive technologies are the digital equivalent of massive foreign-owned refineries. They process vast amounts of data and produce powerful insights, but are overwhelmingly shaped by priorities, values and assumptions from outside the continent.

When used in African contexts, these models frequently reproduce biases, and misunderstand local cultures, languages and social realities.

Consider how content moderation tools developed in the US and Europe misinterpret political speech in African contexts, thereby silencing legitimate voices while missing real threats posed in local languages.

Or how credit-scoring algorithms designed for cashless Western economies exclude millions in cash-based African communities from financial services.

These mismatches occur because our data is being processed by externally developed tools that were never built with our realities in mind.

While these externally developed technologies are showing themselves daily to be revolutionary in many areas, a pure dependency on foreign AI products creates serious economic and geopolitical vulnerabilities. Licensing fees, expensive cloud subscriptions and imported technologies drain resources from local economies, benefiting overseas corporations rather than South Africans.

If political tension rises or sanctions are imposed, we could lose access to critical AI services overnight. Imagine being abruptly cut off from the very technologies now increasingly embedded in healthcare, finance, governance and security.

Or we could continue to merely exist as data collectors whose data can be freely mined by foreign models, thus eroding any autonomy over our own future.

To avoid this fate, South Africa must urgently invest in building and operating its own AI "refineries". If we fail to do so, we will remain dependent on foreign-controlled AI systems, surrendering control over critical aspects of our social and economic future. Our economies will continue to export "raw materials" - our data - and import expensive finished products, the AI services developed elsewhere.

This pattern reinforces the inequality of the global digital economy, extracting wealth from Africa while leaving us vulnerable to external agendas.

Algorithmic dependency also means that decisions crucial to our economy, governance and everyday lives will increasingly be shaped in boardrooms thousands of kilometres away, whether in Silicon Valley, Europe or China. We risk losing agency over how AI interprets our data, decides our creditworthiness, moderates our speech, and even determines our access to healthcare and social services.

Building South Africa's algorithmic refineries

Achieving algorithmic sovereignty requires deep investment in local research institutions capable of fundamental innovation in AI. We must go beyond merely adapting foreign technology or collecting data for others to use. Instead, we must establish our own centres of excellence that contribute to setting global AI standards.

A prime example is the Machine Intelligence and Neural Discovery (MIND) Institute at Wits University, which is already leading the charge. By focusing on foundational AI science, MIND is building the intellectual infrastructure needed to refine our data on our terms.

This demonstrates clearly that we have the talent and ambition to do more than export raw data; we can produce world-class technological breakthroughs right here.

But isolated pockets of excellence will not suffice. The South African government must support these efforts through sustained investment, strategic policy frameworks, and the creation of vibrant ecosystems linking academia, industry and public institutions.

We need clear pathways from research breakthroughs to practical applications, ensuring that AI innovations benefit local industries, communities and public services directly.

Practical steps towards algorithmic sovereignty

Four immediate, practical steps that South Africa must take towards algorithmic sovereignty are:

1. Invest in foundational research: Government and industry must fund strong research institutions that conduct original and foundational scientific research. Long-term grants, dedicated research fellowships and advanced computational infrastructure are essential;
2. Bridge academia and industry: Innovation thrives when universities, start-ups and public sector agencies collaborate effectively, with each playing to their own strengths. Clear mechanisms must be created to translate academic research into viable local products and services;
3. Retain and nurture local talent: Competitive research opportunities, supportive funding environments and strong mentorship networks are crucial in preventing talent drain. Initiatives like the Deep Learning Indaba, a pan-African AI community, provide templates for retaining and cultivating local expertise. Lowering bureaucratic barriers to welcoming foreign talent from around the continent would also greatly bolster our efforts; and
4. Lead ethically and politically: South Africa must actively shape global AI governance debates, advocating ethical and policy frameworks that reflect African realities and priorities, rather than simply adopting external standards that may disadvantage us.

A future shaped by Africans, for Africa

Algorithmic sovereignty is thus not merely about technology; it's about our right to shape our own digital future. It's about refining our most valuable digital resources locally, ensuring that the wealth generated by AI stays within our economies, and aligning technology with our ethical and cultural values.

South Africa stands at a crossroads. We have abundant talent and emerging centres of excellence ready to refine the digital oil we already possess. Now we must commit strategically and urgently to building the infrastructure, skills and institutions that can make algorithmic sovereignty a reality.

Just as a nation with abundant oil reserves but no refining capacity remains economically dependent, so a nation with vast stores of data but no local algorithmic capacity is equally vulnerable.

The choice before us is clear: either invest in the refining capacity of our AI industry today or remain perpetually dependent tomorrow. The future of South Africa's digital sovereignty rests on the decisions we make now.

Professor Benjamin Rosman is a Professor in the School of Computer Science and Applied Mathematics and the Director of the new Machine Learning and Neural Discovery (MIND) Institute at Wits University. He is named in this year's TIME100 most influential people in AI in 2025, and was made a Fellow of the Canadian Institute for Advanced Research (CIFAR). He is the Director of the Robotics, Autonomous Intelligence and Learning (RAIL) Laboratory, founder and organiser of the Deep Learning Indaba, and Chief Science Officer of Lelapa AI.

This article first appeared in Daily Maverick.