Skip to content

AI Policy and Practice:  must be for national autonomy and development, not subservience

This piece follows my recent article on the withdrawn AI policy, in which I argued that civil society groups can still re‑engineer social, environmental, and justice perspectives into the debate. 

Recent reports described how the minister and their staff were grilled by the portfolio committee over issues of AI hallucinations and the withdrawn policy draft. Generally speaking, I believe this is embarrassing, but a minor failure—one that underscores the need for genuine and inclusive regulatory oversight.

Here, I revisit two neglected areas: the governance of the policy‑drafting team, and proactive thinking about waste throughout the AI lifecycle.

(image:https://smarterarticles.co.uk/your-cloud-is-drying-my-river-the-real-cost-of-ai_)

Governance and AI policy making

The new panel of experts are:

The Data Centres, Surveillance, Control Working Group, in a media release sent to the Minister and the portfolio committee responsible, stated: “We have learned through public announcements that an advisory body has now been established, welcoming the panel as having people of the highest calibre in their fields present. The working group advised that the panel lacks expertise that reflects the priorities of the majority of our people and our national interests, affirming that: “We uphold data sovereignty: our national data is not for sale but must serve the public good, with insights and innovations remaining under domestic control for the benefit of Africans, especially South Africans.”

Who is missing from this panel?

To partially remedy these shortcomings, it was recommended that the advisory committee find ways to include experts in various fields: gender activism (algorithmic bias, gendered labour, surveillance harms); children’s rights (privacy, developmental risks, online harms); history (technological transitions, state surveillance, civic memory); Indigenous Knowledge Systems (data sovereignty, traditional knowledge, community protocols); environment (data centres’ impact on water, energy, e-waste, with NEMA compliance pending); heritage (intangible and built heritage, with NHRA compliance outstanding); international human rights law; international humanitarian law; automated weaponry ethics (cross-border and conflict implications); and national security (AI’s role in warfare, as seen in West Asia and Ukraine).

There is more to talk about the corporate or the corporations’ influences over the existing panellists – as we must do whatever we can to avoid corporate capture of our policies.

In a hybrid workshop held on 15 May 2026, the recording is also available here for your perusal: https://www.youtube.com/watch?v=PVzo4GYnSY4 — we discussed the blatant human rights and environmental gaps in the policy direction of the withdrawn draft.

Personally, whilst many of us welcomed the panel and demanded more diverse voices to be included, these intellectuals now appointed must be vetted for their alignment with Big Tech. 

This is a consistent part of our concerns to stem the tide of AI hype that is pushed in the media, academia, and amongst many walks of life in society. Narrative capture is largely engineered by these big corporations which, like power generally, have dominated the discourse whilst belittling the search for non-corporate-led alternatives.

The mainstream media, by all accounts, is being captured where groups like Meta, OpenAI, Google, Microsoft, and others dominate the international media landscape. Civic groups are organising – from mapping to organising communities to track the funding relationships between Big Tech, AI companies, and the media (journalism in particular) – hoping to expose the emerging trends around media capture.

It must be clear that we engage to raise awareness of the dangers posed by “AI corporations and ideologues” – which are more than that!

Their manifestos are scary, anti-democratic, and authoritarian. In addition, they consciously seem to underplay questions of national and personal sovereignty and autonomy, and their form of development has been correctly defined as Empires.

Sadly, they seem to see themselves as empire-builders, and some have resorted to having their own manifestos which appear to undermine or hollow out our national constitutions, as well as the principles and conventions of international human rights law and our institutions.

These “benevolent tech dictators” now present their goals, much like the colonists and empire builders of yesteryear, as being in humanity’s best interests.

Through the immense power they have amassed, they hide the fact that their conduct destroys our ecosystems, centralises information (which in itself contributes towards creating monocultures, even undermining our sense of self), and so on. In this process, they undermine our people’s institutions and movements as they march on towards gaining even more concentrated power – political, economic, and social.

more about our panelists

At first glance, none of these researchers are directly in Big Tech’s pockets. But the institutions where they work and study are heavily funded by AI companies—Google in particular. For example, Google gave one million dollars to Wits University (where Prof. Rosman works), and their tasks included facilitating dialogue among academia, industry, and policymakers; thus, getting Africa to the global AI table would be courtesy of this powerful corporation.

His university outlines Prof Marivate’s 14-year association with Google, from a software engineering internship (2012) to a Google Fellowship during his PhD (2009–2015), and his continuing role as a Google Scholar in 2023. He then received a $20,000 research grant and ongoing collaboration on AI projects focused on African leadership in the field.

One would have hoped for a more sober assessment of the role of the corporation from a knowledge centre, but the university went over the top celebrating the corporation thus: “Tech giant Google is not just an internet search engine whose brand name has entered pop culture as a word. The company also funds academic research and scholarships – and it recently awarded the University of Pretoria (UP) three grants totalling $65,000 (over R1.1 million).”

Marivate’s institution also received Google’s investment at the University of Pretoria, and his institute—the AfriDSAI institute—reportedly also received one million US dollars from Google in 2025. The funding supports, inter alia, fellowships for MSc, PhD, and postdoctoral researchers, among others. It is reported that earlier, Prof. Marivate received Google research grants in public research.

The University of Cape Town (linked to Prof. Gillwald) also partners with Google Cloud and NVIDIA. On the surface, Google funding “responsible AI” research at universities looks independent. But there’s a problem: Google profits directly from weak data rules and unrestricted data centre expansion. So, can researchers funded by Google honestly recommend policies that would limit Google’s power?

These conflicts of interest must be declared upfront, just as MPs are required to do. It is vital that communities, NGOs, and unions remain vigilant to ensure that corporations are not writing our laws—directly or indirectly.

Yet corporate control of the AI conversation is only part of the issue. Even deeper problems remain: How are AI minerals extracted, by whom, and who benefits? As regards the data centres, whose ownership is widely believed to be the same actors as those funding other dimensions of the value chain, this remains opaque to the public and the government. What about the e-waste and the “sacrificial zones”—communities and environments destroyed for resources? That is where policy should really begin.

Waste not, want not

My work is particularly around critical raw materials and mine waste. It suggests that we broaden the discussion when discussing Artificial Intelligence (AI) models, whether they are niche analytical tools or Large Language Models (LLMs) such as DeepSeek, Gemini or ChatGPT. One leg of this discussion lies clearly in the recent rush for critical raw materials, where it is said that Africa holds 30% of the world’s critical mineral reserves, essential for electronics and AI hardware, but receives only 10% of the global revenue.

Many of us simply default to talking about software, programming and data used to push AI forward. We choose to ignore, at our peril,  the fact that the raw materials, the hard stuff, are needed to propel this supposed forward march of society.  These big corporations that are in Silicon Valley and elsewhere have to come to our worlds to negotiate, rob, steal or bargain to get these.

The hard underbelly – extractivism of minerals, disrupting lives

As they are essential for or rely on specific minerals to enable semiconductors in GPUs as well as the massive data centres that train large language models, research reveals these minerals and some of their functions as follows:

  • Cobalt: Essential for the batteries that power the data centres running AI models.
  • Lithium: Also crucial for lithium-ion batteries, which are vital for data centre energy storage and other AI-relevant hardware.
  • Graphite: Another key component of lithium-ion batteries.
  • Gallium: Used to produce high-speed processors and powerful semiconductors, the “brains” of AI computing.
  • Rare-Earth Elements (REEs): This group includes neodymium and dysprosium, which are used to create the strong magnets needed for motors and heat-resistant components in cooling systems for powerful electronics.
  • Ruthenium: A Platinum Group Metal (PGM) whose price has surged due to increasing demand from the AI industry, particularly for use in electronics, semiconductors, and data storage technologies.
  • Tantalum: Derived from the ore tantalite, this metal is used to produce electronic capacitors, a fundamental component in high-performance electronics and advanced computing infrastructure.
  • Copper: A fundamental mineral used throughout electronics and for wiring in the energy systems and data centres that support AI.
  • Platinum Group Metals (PGMs): This group includes ruthenium and others. South Africa is a key source of these metals, which are used in various hardware components.

The minerals come from our communities where people live, farm or struggle to eke out a living. When the bosses come, people have to be “relocated”, displaced or resettled. Many of these minerals are in Africa, as I will show below. It is probably appropriate to mention here that the Critical Minerals and Metals Strategy of South Africa (2025), in all of its 44 pages, has only two references to tailings / mine waste, and none of these touch on the damage and disempowering impact of these on the ecology, people’s lives and livelihoods. They simply speak about mining and re-mining of these dumps without real rehabilitation of the environment and redress of people’s lives being considered. This AI policy cannot make that same mistake.

In organisations working with mining-impacted communities, we often analysed the mine / tailings waste and other impacts of extractivism as an externalisation of costs by corporations, which has devastating effects on communities living adjacent to mines. We used to say that mining/extractivism is a waste-producing industry/sector, confirming that: “To get just one ounce of gold, a typical mine in South Africa has to dig up, crush, and process roughly 20 to 50 tons of rock. That means more than 99.9% of what is extracted ends up as waste. However, the physical waste rock is only part of the environmental cost; processing this massive amount of material requires vast quantities of water to crush, grind, and chemically leach the ore from the surrounding stone.”

In the so-called Platinum province North West, Rustenburg or Kolwezi copper and cobalt mines in the Democratic Republic of Congo (DRC) cannot remain sacrificial zones – just available to be mined and extracted. These top-down relations with host communities / mining-impacted communities only result in mine waste disasters, as we recently experienced in Zambia at the Sino-Metals Leach copper mine in Chambishi (Copperbelt Province) on February 18, 2025. There, a tailings dam collapsed, releasing up to 1.5 million tonnes of toxic, highly acidic waste into the Mwambashi and Kafue River ecosystems.

Thus, we assert that the centrality of waste must be factored in — from the inception of mining and extractivism of these minerals, through to processing and use in constructing data centres and even software. This principle was evident even in the flawed MPRDA (mining law) and also in the environmental law (NEMA), which obligates rights holders to provide upfront financial provisioning to deal with environmental liabilities, rehabilitation,  mine closure, and post-mine closure interventions to address the underlying damage of mining. Unfortunately, compliance is lacking, as is effective and coordinated regulation on behalf of people, the environment, and the wider ecology. This is the lesson — warts and all.

An important challenge for us is that the rush for CRMs and the AI boom — which some have called the AI arms race (not to mention its military uses) — can be disarming. It leaves us insufficient space and time to think of African-centric alternatives that are not solely profit-driven.

I want now, to briefly revisit waste around artificial intelligence and its data centres. The hardware that runs AI (chips, GPUs, servers) has a very short life — only two to five years. Given the urgency, policy must address this problem. Similarly, policy must make tech companies responsible for their own trash, which they dump on wider society.

Any company selling AI hardware in South Africa must pay for its entire lifecycle. They must not be permitted to sell these chips and other equipment without being legally responsible for collecting, repairing, and safely recycling every server and GPU they sell here. These are proposed small steps that will ensure we do not become a dumping ground for the global AI arms race.

GPUs are called the “engine” of AI because they are expensive to manufacture, requiring rare minerals and intensive chemicals, for instance. Moreover, due to the rush and competition, corporations find it easier to dump older machines and replace them with newer ones. The consequences, some have found, are as follows:

  • By 2030, AI could produce up to 5 million tonnes of e-waste.
  • This waste contains toxic lead, mercury, and chromium.
  • It also contains gold, silver, and copper — present at concentrations 50 times higher than in mined ore.
  • However, currently less than 25% is recycled properly.

Principles to guide action

This is an edited version to the initial draft. click here for the full list: https://csos4tailingsjustice.org/ai-principles-to-guide-action/

 To ensure our AI policy yields just and fair outcomes rather than following a growth-greed model, we propose the following guiding lights:

  • The Right to Say No (RTSN): Communities must be able to accept or reject data centre projects.
  • Full lifecycle waste accountability: AI and data centre waste must be accounted for from extraction and processing through construction, operation, and maintenance.
  • Workers’ rights to organise: Unionisation must be protected and promoted.
  • Algorithmic justice and equity: Systems must be designed to eliminate bias and inequality.
  • Global regulation of military AI: Enforceable limits on autonomous weapons systems.

AI is not neutral, it is not human, and it has no agency. The poor and working people—in whose name progress is often championed—have agency. We demand that a national policy be inclusively drafted, inspire national debates, make visible the invisibility of AI, and look beyond the bright lights and quick answers to complex problems. It must dig into the murky underground, beneath where minerals and other natural resources are found, and hear and listen to those who are heavily impacted. 

Hassen Lorgat

Hassen Lorgat is a social justice activist and writes here in his capacity as convenor of The CSOs Tailings Working Group