By Fieke Jansen
The relationship between climate change and Artificial Intelligence is one of the biggest blindspots in the European Commission ‘leaked’ white paper on AI, writes Fieke Jansen.
Fieke Jansen is a PhD candidate at the Data Justice Lab and Mozilla Foundation Fellow 2019-2020.
Only weeks after the initial ‘leaked’ white paper on AI and the EU’s consideration to temporarily ban facial recognition got media coverage, this option has already been dropped.
While the white paper still covered a range of other AI challenges and opportunities, one topic still remains conspicuously absent: The environmental impact of AI.
The white paper notes that the volume of data stored across the world — data that is used to train AI — will most likely quadruple from the current 40 zetabytes to 175 zetabytes by 2025.
And where 80% of the 40 zetabytes is currently stored in the cloud, the EU assumes that the rise of IoT products and edge computing will be a catalyst for even more decentralized storage. The paper stops there, failing to acknowledge how this massive increase in data and shift to device storage will produce a carbon footprint.
This oversight is both disheartening and dangerous, and contradictory to the European Green Deal Communication.
This EU strategic orientation aims to address the ‘twin challenges of the green and digitals transformation.’ Stressing that while technology can enable the EU to reach the sustainability goals the digital sector needs to put sustainability at its heart.
Then why has this commitment to technological environmental has not been integrated into the AI white paper?
The impact of technology on the climate crisis is not new. The shift project released their report on ‘lean ICT’ in 2019, revealing that technological energy consumption is increasing by 9% every year. Further, compared to 2010, ‘the direct energy consumption generated by 1 Euro invested in digital technologies has increased by 37%’.
To buy a book on Amazon, this project shows that the website forces you to navigate 12 ‘heavy’ interfaces, which is worth 8,724 A4 pages of printed code or 87.33MB of information. It costs 30 Watt-hours to load these pages.
In 2014 google.com received approximately 47000 global requests per second, which represents an estimated amount of 500 kg of CO2 emissions per second.
Advanced AI’s carbon footprint can be even more overwhelming. Researchers at the University of Massachusetts recently studied the environmental impact of the development and testing of natural language processing (NLP) model, a type of AI.
This research found that the training, building and testing of an academic paper-worthy NLP model can emit more than 626,000 pounds of carbon dioxide. MIT Technology review compared these carbon emissions with ‘nearly five times the lifetime emissions of the average American car’. Meanwhile, NLP is just a small part of the AI supply chain.
Despite this frightening evidence, the ‘Green’ European Commission presents a pollyannaish view on the link between AI and the environment throughout their white paper.
One section of the paper notes that ‘At home, a smart thermostat can reduce energy bills by up to 25%’ — but what about the energy it takes to train and run those thermostats?
The paper also notes that AI will allow for us more accurately predict environmental issues and hone our crisis management capacity — but doesn’t note that AI can also contribute to these very issues and crises.
The Commission’s failure to acknowledge the environmental impact of AI is especially problematic because this decade, the EU will significantly increase its funding of AI. In 2016, the EU spent €3.2 billion on AI projects and now aims to increase this investment to €20 billion per year.
This means more and more AI will be deployed without necessary environmental safeguards.
Going forward, the EU must weigh up AI’s effect on the climate crisis. This requires the EU to have a critical analysis and clearly articulated position on the relationship between AI and climate change, both in legislative options and funding mechanisms.
In particular, the EU needs to integrate environmental impact assessment on AI projects and build an explicit research agenda that looks at the relationship between AI and the climate crisis.
This isn’t a new idea. In the past, the High-Level Expert Group on AI has specifically stated that ‘AI systems promise to help to tackle some of the most pressing societal concerns, yet it must be ensured that this occurs in the most environmentally friendly way possible’.
It’s time to live up to that language.