Image by Alan Warburton / © BBC / Better Images of AI / Nature / CC-BY 4.0
The constant progress and development of AI systems are raising important ethical, human rights and environmental questions. While the use of AI systems are on the rise especially with the arrival of generative AI systems openly accessible such as ChatGPT and DALL-E, the question of their environmental impact remains unaddressed.
This week, our MediaFutures partner KUL attended the CPDP conference. CPDP stands for Computer, Privacy and Data protection conference. It is a world-leading multidisciplinary conference which offers a cutting edge in legal, regulatory, academic and technological development in privacy and data protection. CPDP gathers academics, lawyers, practitioners, policy-makers, industry and civil society from all over the world. This year, one panel really stood out in our opinion and triggered this blogpost. Here are the takeaways from the discussions and panelists speeches.
The panel organised by AlgorithmWatch was entitled “The global harms of powering artificial intelligence towards a sustainable future of data use and governance”. It was composed of academics, policymakers and civil society (full list of speakers here). While the three days conference programme is quite comprehensive ranging from privacy, cybersecurity, Artificial intelligence, advertising, online economy and so forth… We could observe how rare were the sessions focusing on the environmental impact of the development, deployment, application and end of life of technology. How come that policymaking, civil society and activism are often addressed in silos? For instance, out of the political priorities of this European Commission mandate we can indeed find two prominent and ambitious ones but not always compatible : A European Green Deal and a Europe fit for the Digital Age. AI has a twofold impact, it is often portrayed as a tool to fight the climate crisis (smart design or resources use, modeling predictions) but its carbon footprint and natural resources needs are actually raising serious concerns. One can wonder about the (un)balance between the two impacts and the necessity of the current scale of AI development. The environmental and tech agendas need to be looked at together to really achieve a sustainable use of technology.
The AlgorithmWatch panel shed light on how little the environmental impact of technology is addressed by legislation and policy initiatives, how the subject is not on the tech conversation tables right now and how it is hard to have exact numbers, statistics, quantifiable harm about the impact of AI systems on ecosystems, climate change and the environment. Little research has been done due to a lack of data and access to information from manufacturers, developers and deployers. How energy-intensive are such AI systems per months/years for training, for use? How environmentally costly has it been to produce and deploy them and how costly will it be to destroy/recycle them at the end of their lives? While elements of responses are there and harms to ecosystems and the environment are already visible, much is yet to be researched, uncovered and documented. This transparency need to assess concretely the risks comes at a time where the use of the systems as components of products or services are literally exploding. See the massive use of generative AI systems by the general public. A group of experts even called for a 6 month pause on generative AI development warning potential risks to society and humanity.
In this context, how come that the environmental impact was not part of the EC proposal for the AI Act in order to quantify the category of risks attached to the AI system while we are facing inexorable global warming and over exploitation of natural resources? The EU proposal for legislation is currently being negotiated and is set to be the first EU internal market instrument focusing on AI systems following a risk-based approach. Thankfully, the European Parliament draft negotiating mandate has introduced the environment and climate change impact and resources used in their amendments for risks assessment and transparency provisions. It remains to be seen what will survive the negotiation process.
A closer look is needed to the following fundamental questions: how and why do we produce, develop, deploy AI systems technology? Who benefits from AI and who suffers from it are crucial questions to reach an informed balancing assessment of the risks and opportunities of AI. Environmental impact of AI systems comes along with its societal, economical, human rights impacts. Sustainable by design should be further developed and co-exist with privacy by design.
For further reading, the below publications and initiatives are putting this environmental impact of AI under the spotlight:
- SustAIn: The Sustainability Index for Artificial Intelligence (AlgorithmWatch)
- Anatomy of an AI system (K. Crawford & V. Joler)
- The carbon impact of artificial intelligence (P. Dhar)
- Underscored by the algorithm: AI’s impacts on labour and environment (J. Renato Laranjeira de Pereira & T. Guimarães Moraes)
- AI can help us fight climate change. But it has an energy problem, too (A. Ekin)
- Measuring the environmental impacts of artificial intelligence computing and applications: the AI footprint (OECD.AI)
- Behind the Figures: understanding the environmental impacts of ICT and taking action (Study commissioned by the Greens/EFA of the European Parliament)
- The role of Artificial Intelligence in the European Green Deal (Öko-Institut. e.V. for the European Parliament’s special committee on Artificial Intelligence in a Digital Age).
By KU LEUVEN – CENTRE FOR IT & IP LAW