Racism, Tap Water & AI
It has never been a secret that many diasporas of racial minorities across the globe have had to face discrimination from their peers. More famous examples include the apartheid of the Rohingya People of Myanmar, the ethnonationalist genocide of the state of Palestine, and segregation under Jim Crow faced by ethnic groups in America. But in the 21st Century, a new generation of racism is being brought about; the danger faced by communities of colour due to environmental hazards such as landfill, incinerators, and hazardous waste disposal.
Environmental racism is a deeply insidious disease, one of which combats many communities today more than ever. And in this digital age, the cycle of harm only entrenches itself in society further.
A time before the beast:
The textbook definition of environmental racism includes “the disproportionate impact of environmental hazards, pollution, and ecological degradation experienced by marginalised communities, as well as those of people of colour.” It is important to remember that environmental racism has existed in the Pre-Information Age, not just with the advent of AI. One of the most famous examples is in Flint, Michigan; Flint had one of the most infamous public health crises due to the water’s lead contamination issue and Legionella outbreak. But did you know that Flint is a predominantly black city? At the time of the outbreak, the city was made up of 56.09% Black or African American people. While it may seem like a coincidence that this public health issue affected a deeply black community, environmental racism as an issue runs a lot deeper. The fact that marginalised communities have a lack of political power and representation in office means that marginalised communities, such as the people of Flint, are unable to resist governmental bodies and cannot access political power to motivate change.
That is the core of the issue, the fact that minorities are impacted by environmental hazards and then lack the representation or power to fight it.
How to feed a beast:
In order to understand how the recent rise of AI carries on a lineage of disproportionate attacks on people of colour, we must understand how AI datacentres work in the first place. A datacentre is a large building or dedicated space that houses the infrastructure required for computer systems. They have many requirements for functionality, however, when specifically considering AI datacentres, the main environmentally taxing requirements are cooling facilities & energy demands.
Datacentres can be cooled in many ways, the most prevalent however is water cooling. To prevent delicate computer components from overheating and therefore breaking, water can be used to remove heat from the component through evaporation. This cooling method requires vast quantities of clean water, of which is returned to the environment contaminated or returned to the atmosphere as steam. This renders the water unsafe for human consumption and therefore is removed from the supply of potable water available locally.
One 100-word ChatGPT query consumes around 40ml of water to cool the components, which is around one-shot glass (an image-based response increases this dramatically). While not seeming like a lot at first, this totals to mean that a single 100-megawatt data centre can use from between 2,000,000 to 5,000,000 litres of water per day. Which is the equivalent to the daily consumption of 6,500 to 16,250 households.
Data centres also guzzle electricity exponentially. Data centres in the U.S. consumed approximately 176 terawatt-hours of electricity in 2023, accounting for about 4.4% of the country's total electricity consumption. However, globally with the recent surge of AI, it is expected that the amount of electricity used could inflate to the energy requirements of the entire country of Japan annually, by 2030. It is difficult to imagine this extensive scale of energy consumption, so, in order to demonstrate the energy consumption of an average person’s AI habit imagine this: a person running a bake sale uses ChatGPT to produce around 15 recipe ideas, 10 visual aids for flyers and a short 5 second video for Instagram. You would require 2.9 kilowatt-hours of electricity. This is enough to ride 100 miles on an e-bike, 10 miles in an electric car, or to run a microwave for over 3 and a half hours nonstop.
On the topic of electricity, data centres must be left running 24 hours every day nonstop or else all data stored there is lost. Therefore, the majority of datacentres have a failsafe mechanism in place if power was to be lost, being diesel generators. A smaller datacentre may not require many, however larger datacentres such as the ones purpose built for AI require dozens and dozens. For example, the company, Quantum Loophole, proposed installing 168 diesel generators capable of delivering 504 MW of power running all the time in their datacentres. These generators emit significant amounts of particulate matter (PM), oxides of nitrogen (NOx), sulphur dioxide (SO₂), and carbon dioxide (CO₂). These pollutants degrade air quality and pose serious health risks to nearby communities.
When the beast attacks:
As previously mentioned, the core of the issue of environmental racism is that marginalised communities have a lack of political power and representation in office meaning they are unable to resist governmental bodies and cannot access political power to motivate change. But now with the rise of AI (facilitated by government investment), datacentres are sprouting up everywhere, coincidentally, in neighbourhoods of a predominantly non-white ethnic majority.
In America right now, African American communities are forced to live next to huge, toxic particulate diffusers and breathe highly contaminated air, leading to heightened levels of asthma and lung cancer from all the pollutants emitted by large scale data centres, while simultaneously having to compete with the datacentre itself for basic resources such as water, electricity and even the space that the datacentre takes up. The price for these basic commodities skyrocket as supplies run low.
These low-income communities certainly cannot outcompete these goliaths, meaning basic resources are monopolised and the local community is held hostage for their own resources.
An infamous example in the news right now is the residents of Boxtown, Memphis fighting against Elon Musk’s company; xAI and their large-scale AI data-centre ironically called ‘Colossus. A local climate activist described that, “You could almost see the sky pulsing because of all the chemicals just being spewed at such an accelerated rate.” He describes AI datacentres to be “the new Frankenstein of environmental racism” due to the centres increasingly being built in black, Hispanic/Latino, indigenous and poor white communities. xAI’s Colossus uses 35 methane gas turbines as supplemental and contingency power. But it is not just methane, oxides of nitrogen were found to have increased after the construction of Colossus; locally by 3%, in Boxtown by 9%, and immediately around the datacentre by a whopping 79%. Ever since Colossus’ construction, elderly people locally with no past respiratory complications have suddenly developed asthma and residents with Chronic Obstructive Pulmonary Disease have reported extreme health complications.
How to slay a beast:
As AI will progress further in the future, we must consider who must bear the forefront of AI’s physical repercussions. Systemic cycles such as environmental racism perpetuate the suffering of ethnic minorities so that the ethnic majority is able to use these tools. When it comes to AI, there is always a price to pay, its not some magical semi-sentient personal assistant who lives in your phone. AI is code. It cannot think or feel or have opinions. And when we as a collective society treat AI otherwise, we forget the real, very human cost associated with it.
Edward Hancox
