The world’s leading publication for data science, AI, and ML professionals.

AI and the Environment III: Beyond Energy

AI sustainability is not just about energy. It about rare-Earth element extraction, human labour, pollution, and water preservation as…

Image by author
Image by author

Introduction

This is the third post in my series on AI and the Environment. In this post, I will give a brief overview of all the other environmental impacts of AI – from rare-Earth element extraction and human labour to water conservation and pollution.

If you have not read the first two posts, that is fine! However, I do suggest scanning through their introductions ([here](https://medium.com/@jeffrey.sardina/ai-and-the-environment-ii-engaging-the-dialogue-e0460f3ad6d9) and here) to get an overview what AI is and how it affects global energy use and carbon emissions.

This article takes a new perspective: going beyond energy, what does AI (and computation in general) imply for the world we live in? There are a huge number of factors here, so I will give an overview of the most important ones and leave some references for further reading. The key take-away points are:

  • Creating the computers AI runs on requires rare-Earth elements, which have huge environmental and human costs to extract and refine
  • AI data is sourced from the public and from public data sets, often without people’s knowledge.
  • AIs are often private, not public, which limits re-use and re-application of Green AI
  • The data centres that create AIs require huge amounts of water and energy for cooling

Rare-Earth Elements

There are 17 elements called the "rare-Earth elements" [1]. This term is a bit of a misnomer: it refers not to how rare the elements are, but rather to how hard they are to obtain [1]. The important thing, however, is that these elements all have three things in common

  • They are in very high demand for building computers and other technology, including AI [1]
  • They have huge environmental costs to extract and refine [1]

99.8% of the earth extracted while mining rare-Earth elements is discarded and dumped onto land in into streams, with only 0.2% actually containing the desired elements [1]. On top of that, the Chinese Society of Rare Earths found that refining a single ton of rare-Earthelements produced one ton of radioactive waste and 75,000 litres of acidic water [1].

Interwoven with the concerns of rare-Earth and other elements are all the humans involved in their extraction – a process that is often dangerous. The workers who extract Cobalt in The Conga, for example, face terrible wages – equating to around $1.00, or around €0.90, per day [4]. All the while, they have to work in conditions hazardous to their lives and health [4]. Studies in many other areas of extraction have largely painted the same picture of exploitation [1].

Looking beyond rare-Earth elements at all the other parts needed – such as plastics and glass – the total process of building a computer from extraction of raw materials to assembly in a factory involves transporting elements millions of kilometres [1]. And of course, shipping is not free – shipping is estimated to create 3.1% of global carbon emissions [1]. This shipping, which often uses low-grade fuel, emits many other toxins as well such as sulfurs [3]. It is estimated that these emissions lead to around 30,000 human deaths per year [3].

Hidden Human Labour: the Data Economy

Going beyond the human labour directly involved in building computers [1,4], there is a huge hidden layer of human labour: each and every one of us reading this blog are labourers for Google, Amazon, Microsoft, Apple, and likely many others [1].

Every interaction with Amazon’s Alexa (or any other similar device) is recorded as data. Did you ask the same question again? If so, then Alexa can infer you did not like its answer. Was any action taken after a question / command was asked or given? If so, Alexa assumes it did its job well. And so on [1].

But more than that, Alexa and related devices always listen. They use voice recordings to improve speech recognition behind the scenes. This data comes from every one of us who is around these devices.

Similarly, remember all those "CAPTCHA" you have to complete to sign on to a website? The ones that say "I’m not a robot", or that ask you to select all images that contain bikes? Each one of those is collecting your data and using it to train AIs to be better at image recognition – or any number of other tasks [1].

And here’s the thing: that data comes from the public. But the tools developed are all privatized, the public having no right to see how they are made, or what they are used for.

As a case study, consider DeepMind Technologies and the Royal Free hospital. DeepMind gained access to public data from the hospital, and used it to create a private tool [1]. However, that tool was kept private at the time – not made available to other public hospitals [1].

AI serves as a way to privatize the application and use of public data. And this can have an impact on the environment as well.

When results are held private, that means that tools made on public data – potentially valuable health data – are reserved for only those who are given access. But if these tools are some of the many AIs used to help reduce emissions and perform smarter power gird planning (as in [5]) this means that tools to help the environment are similarly hidden (though to be fair, much of that data was likely private to begin with, not public).

But it does not end there. Hidden behind all of this is another realm of data extraction: the co-called "clickworker" factories [1]. These ‘digital factories’ involve large amounts of people paid to label data for an AI – for example, to tell if a picture includes a bike. This is often done with Amazon Mechanical Turk, a program that allows people to log on to do this work [1].

This allows producing huge datasets to train AIs. Huge data feeds into huge AIs, requiring large amounts of energy. But if the results are private, nobody else can benefit from it without redoing all this work themselves. From data collection to AI creation, with all the energy involved in between, it must be done again.

Water Conservation

Taking a look back at datacentres (warehouses of computers that, among other things, can create AIs), we see several things.

From one, every single computer, processor, and cable in every data centre draws upon rare-Earth element extraction, huge networks of shipping raw materials, assembly, human labour, and the data economy [1]. So to does every single phone, laptop, computer, gaming consoles – you name it.

But once that is done and the data centre is built, there are two major problems left: energy and water. I’ve already covered energy in detail here, so in this section I intend to focus on water.

As AIs are created in data centres, the computers generate heat – just the same as you your laptop of phone will get hot after being left on for a long time. In order to cool them down, large supplies of water must be used [5]. In fact, in the USA alone, data centres were estimated to have used 100 billion litres of water in 2014 [5].

If this cooling is done via misting or using cool water, huge amounts of energy can be involved beyond the energy needed to run the data centre computers [5]. But even with recent warm-water cooling, there is still a large water cost – even though the energy needed is far decreased [5].

While this may not be an issue in some areas where water is abundant, in many regions facing drought – or in deserts – this needs consideration for proper conservation.

Conclusion

The issues arising from AI and the environment go far beyond energy use: it is labour, it is mining and extraction, hidden data economies, pollution, and water conservation.

This is not to say that AI is "evil" – not even Red AI is necessarily bad. AI can do many great things – improving energy grid efficiency, detecting cancer, and creating new drugs to treat disease among many other applications. But those benefits should always enter the dialogue with a serious discussion of the other side – where AI comes from, whose labour creates it, and who benefits from it. In my next post, I will go the complexities of Red (and Green) AI in more detail.

If you found this post interesting and want to know more, I would highly recommend reading Anatomy of an AI System [1] and How to stop data centres from gobbling up the world’s electricity [5] – both are easy reads and are very well sourced.

For now, the key take-away points of this post are:

  • Creating the computers AI runs on requires rare-Earth elements, which have huge environmental and human costs to extract and refine
  • AI data is sourced from the public and rom public data sets, often without people’s knowledge.
  • AIs are often private, not public, which limits re-use and re-application of Green AI
  • The data centres that create AIs require huge amounts of water and energy for cooling

References

  1. Kate Crawford and Vladan Joler, "Anatomy of an AI System: The Amazon Echo As An Anatomical Map of Human Labor, Data and Planetary Resources," AI Now Institute and Share Lab, (September 7, 2018) https://anatomyof.ai
  2. Zoë Schlanger, "If Shipping Were a Country, It Would Be the Sixth-Biggest Greenhouse Gas Emitter," Quartz, April 17, 2018.
  3. John Vidal, "Health Risks of Shipping Pollution Have Been ‘Underestimated,’" The Guardian, April 9, 2009, sec. Environment, http://www.theguardian.com/environment/2009/apr/09/shipping-pollution.
  4. "This Is What We Die For: Human Rights Abuses in the Democratic Republic of the Congo Power the Global Trade in Cobalt" (Amnesty International, 2016), https://www.amnesty.org/en/documents/afr62/3183/2016/en/
  5. Nicola Jones, "How to stop data centres from gobbling up the world’s electricity," Nature, September 12, 2018.

Related Articles