AI’s Hidden Costs: Environmental, Social and the Promise of Soft Law

Following the end of EVE’s Fall School last edition, we are happy to open a new series of blog posts, which will constists in a review the final papers of the fall school’s partecipants.

Oluwatobi Oladipo-Ologundudu & Celeste Viljoen are opening the floor, reporting on their research on AI’s Hidden Costs, to which also Sokhna Fall and Zarreen Xanthe-Khan contributed.

Thank you so much for following along,
EVE’s Team

AI’s Hidden Costs: Environmental, Social and the Promise of Soft Law

By Oluwatobi Oladipo-Ologundudu & Celeste Viljoen 

Research in collaboration with Sokhna Fall & Zarreen Xanthe-Khan

We like to think of Artificial Intelligence as weightless… a magical, invisible engine of the Fourth Industrial Revolution that can solve everything from traffic congestion to climate modelling. The story we are sold is one of efficiency and immaterial progress. But every time we type a prompt into a chatbot or ask an algorithm to generate an image, something very material happens out of sight. Behind the apparent “magic” of a single AI query lies a footprint ten times larger than a standard web search. That one interaction triggers a global chain of extraction, labour and environmental impact. It draws electricity from power grids, pulls water into data centres, relies on minerals extracted from the earth, and depends on the labour of people whose names we will never know. These environmental and social costs are quietly pushed out of sight, hidden behind sleek interfaces and the comforting illusion of effortlessness. But when we peek behind the curtain, the true cost of this so-called miracle technology comes sharply into focus.

Despite the industry’s preference for calling AI “clean” or “digital,” its foundations are deeply physical. Data centres, i.e., the warehouses of servers that make AI possible, now consume up to 1.5% of the world’s electricity, often drawing from fossil-fuelled grids. To keep these servers from melting down requires extraordinary amounts of water (they gulp down water on a scale comparable to small nations), straining local supplies in places already vulnerable to drought.

But the cost isn’t just in the water and power; it’s in the hardware. To build the chips that power AI, we have accelerated a global “scramble” for minerals, like lithium, cobalt, copper, and rare earths. This has birthed what scholars call “AI Colonialism,” where the Global North reaps the digital rewards while the Global South bears the environmental and social scars. Thus, this new colonialism is borne and driven not by plantations or railroads, but by data and code. The bottom line is that AI is not neutral but rather reflects a mirror of old colonial systems, concentrating wealth for a transnational super-rich class while reinforcing global divides.

Yet, the true heart of AI is not only in the machines or minerals. It is also in people. AI systems do not learn in isolation; they are trained by vast numbers of human workers who perform “ghost labour” tagging data and moderating toxic content. Much of this work is done in countries such as Kenya, India, and the Philippines, often for wages that barely support a basic standard of living. These workers are essential to making AI function safely and effectively, yet they remain largely invisible and absent from corporate narratives; excluded from the enormous wealth the industry is generating. 

The distance between the polished product and the conditions under which it is made is perhaps most stark in the Democratic Republic of the Congo. The DRC supplies the majority of the world’s cobalt, a mineral critical to the batteries and processors that power modern AI systems. For the Congolese people, this isn’t a digital revolution but rather a daily struggle for survival as the increased demand has translated into deforestation, polluted water, and the air thick with metal-laced dust. Children grow up with respiratory illnesses; farmland becomes unusable. What appears as “green” or “smart” technology in Silicon Valley is experienced very differently on the ground.

These starkly unequal experiences of AI are not incidental; they are the direct outcome of regulatory and governance gaps that allow technological expansion to outpace accountability. The AI narrative is usually expressed in the language of speed and inevitability, as if innovation naturally moves ahead of any rules or oversight. However, not all is lost – in areas where formal laws are weak or missing (across borders, industries, and unequal power structures) soft laws quietly step in. These non-binding guidelines, principles, and standards may not have legal force, but over time they help us recognise, measure and sometimes even manage the hidden costs of AI on people and the planet, especially in the Global South. 

The inception of soft law is recognition. Global frameworks, including the OECD AI Principles, UNESCO’s Recommendation on the Ethics of Artificial Intelligence, and the UN Guiding Principles on Business and Human Rights, lack the force of courts or sanctions, yet they bear both moral and political weight. They tell of a different AI narrative, one that justifies environmental sustainability, human rights, and social justice as mandatory, and not optional, to technological advancement. For nations in the Global South, who are most times absent from binding regulatory plenary, these frameworks develop a common language to define harm and insist on accountability.

Concerning environmental degradation, soft laws quietly but resolutely change expectations. Voluntary sustainability reporting norms and ESG guidelines now require companies to divulge the carbon footprint of data centres, water usage for cooling systems, and the lifecycle consequences of mineral extraction. These disclosures may not close down a mine or suddenly redirect a pipeline, but they reveal environmental impacts that were once hidden. In so doing, they encourage civil society, investors, and local communities to confront customs that offer ecosystems in place of techno-advancement.

Soft law also extends into labour systems. Ethical AI guidelines progressively tackle working conditions of data workers, such as labellers, moderators, and annotators who train intelligent systems in the background. By promoting reasonable wages, mental health safeguards, and transparency in global AI supply chains, soft standards humanise work that has been labelled “invisible.” For Global South workers, these standards are bargaining tools, promoting advocacy, unionisation, and contractual changes where strong domestic labour laws are absent.

Perhaps most important is the fact that soft laws operate where hard laws cannot. They cross borders without the need for treaties and can adapt to different cultures and economic realities. In jurisdictions with limited regulatory capacity, they provide an entry point, a framework for domestic policies, corporate codes of conduct, and regional cooperation. Progressively, what began as voluntary usually becomes expected, and eventually transforms into real responsibility. 

Even so, soft law is no silver bullet. Its power relies on people, on active participation, careful oversight, and persistent pressure from communities and advocates. These standards risk devolving into abstract rules rather than tools for real change if voices from the Global South are absent at the table. But in a world of uneven governance and so unevenly distributed costs of AI, soft law has an important story to tell. It shows us that the responsibility does not have to wait for formal regulation, and that shared principles (even without legal force) can guide AI toward fairness, sustainability, and care for both people and the planet.