Artificial Intelligence (AI) is regarded as one of the most transformative technologies of our time, promising to revolutionise industries and drive unprecedented innovation. But it brings with its significant environmental implications that can threaten the Net Zero goals of businesses, says Andrew Grigg, Head of Sustainability Consulting, Sopra Steria Next UK.
The use of AI is growing fast. Analysis by Sopra Steria Next predicts that global AI spending will reach $1.27 trillion in 2028. And according to a McKinsey report, 65% of respondents say their organisations are regularly using Generative AI (AI capable of generating text, images, videos, or other data using generative models, often in response to prompts), nearly double the percentage from its previous survey a year earlier.
However, as AI adoption accelerates, so do the environmental implications – posing challenges for businesses that need to be addressed if Net Zero goals are to be met in the future.
The environmental challenges of AI
There are four main environmental impacts of AI:
Energy: AI models, particularly those requiring extensive training, consume vast amounts of energy. According to the International Energy Agency, the energy consumption of AI data centres is projected to double in the next two years, equating to the entire energy use of a country the size of Japan.
Water: Data centres require significant water for cooling purposes. As AI adoption grows, so does the strain on water resources, with estimates that a single session on GPT-3 can consumer half a litre of water.
Carbon emissions: The carbon footprint of AI is considerable. For instance, training a single large AI model can emit as much CO2 as five cars over their lifetimes. Reducing the carbon emissions associated with AI is crucial for meeting global climate targets.
Abiotic resources: The hardware used for AI consumes vast amounts of rare elements and metals. The manufacturers of AI hardware are competing to produce computer chips, which creates tension in supply chains.
Achieving Net Zero targets are a priority for businesses worldwide, however they’ll remain out of reach without addressing the sustainability of AI. In other words, Net Zero is at serious risk without action on AI.
The business challenges
Businesses face several challenges in mitigating these environmental problems:
Unnecessary usage
Businesses face the challenge of being able to recognise when it is beneficial to use AI, and critically, when it is not. A lot of problems don’t need AI to solve them, but misunderstanding of the real potential of AI is common. This leads to money being wasted on building and running AI solutions that aren’t needed.
Mismatch between efficiency and demand
Major players in AI, such as Microsoft and Google, argue that energy consumption will be reduced through specialised AI hardware. However, as the May 2024 International Scientific Report on the Safety of Advanced AI reports, the energy efficiency of computation generally improves by an estimated 26% annually. The same report compares this efficiency against a 400% annual growth in demand for computing power used for AI training.
Lack of reliable measurement of AI’s impact
Businesses face a lack of clarity regarding the true environmental impact of AI systems. As noted in this OECD paper, there are no globally recognised measurement standards and hyperscalers aren’t incentivised to publish robust data on the environmental impact of their services. This makes it difficult for organisations to make informed decisions and implement effective AI strategies.
No Life Cycle Assessment (LCA) standard for AIs
There is no generally recognised standard for LCA of AI as this research paper by MRS Communications observes. This lack of standardisation makes comparisons and reporting more complex for businesses when facing decisions around which AIs to use.
Why sustainable AI is essential
Integrating sustainability into AI initiatives isn’t just ethically important; it also offers tangible benefits for businesses. These include:
Meeting Net Zero goals: By adopting sustainable AI practices, businesses can significantly reduce their carbon footprint and align with global Net Zero targets. Meeting Net Zero goals can provide a host of benefits including improved brand reputation, compliance with regulation, and reduced costs.
Quicker time to market: Sustainable AI practices can reduce time to market due to reduced AI model training time, streamlined processes and improved efficiency. This leads to faster time-to-market for AI-driven solutions, which is what Sopra Steria is doing in partnership with Confiance.ai.
Cost savings: Energy-efficient AI solutions can lower operational costs by reducing energy consumption and optimising resource use.
Reduced environmental impact: Sustainable AI practices minimise the negative environmental impact of AI technologies, contributing to broader efforts to combat climate change and preserve natural resources.
Climate risk mitigation and adaptation: By reducing the use of energy, water, and abiotic resources in their use of AI - and working with their digital services supply chain to manage risk of infrastructure being in climate vulnerable geographies - organisations can mitigate the risks that climate change poses.
Essential steps for implementing AI responsibly
At Sopra Steria, we advise our clients to take the following actions to implement Sustainable AI practices:
Implement green AI practices
Educate AI and sustainability teams on the relationship between AI and environmental impact, and importantly on when and when not to use AI. Promote a culture of sustainability within the organisation by empowering all employees to take action and accountability for sustainability outcomes. Collaborate with the AI expert community to share best practices and innovations in sustainable AI.
Measure the environmental impact of AI
Before running AI models, firstly conduct a full lifecycle assessment to understand the environmental impact of data collection, storage, training, and deployment stages.
Secondly, following deployment, conduct a full analysis of the target platform for your solution and equip it with tools to track your real consumption for the run phase. Use tools and methodologies such as CodeCarbon, Carbon Tracker and EcoLogits to measure and monitor the carbon footprint of AI projects accurately.
Algorithm and model optimisation
Develop and use algorithms that require less computational power without compromising on accuracy. Techniques such as pruning, quantisation, and knowledge distillation, can help in optimising models. We estimate a 50% reduction in CO2 can be achieved by reducing training time by half, whilst only losing a few tenths of precision in the model accuracy.
Optimise data centres
Choose locations for data centres that have a lower environmental impact, considering factors like the local energy mix and availability of renewable energy sources. We estimate that, based on Ember and Energy Institute data, by moving data centres from Germany to France a business could save 86% of CO2.
Transparency and reporting
Advocate for the development and adoption of global standards for measuring the environmental impact of AI. Ask your AI providers for metrics and data on the environmental impact of their technology. Support initiatives that aim to create LCA standards for AI systems. Regularly publish reports on the environmental impact of your AI initiatives to maintain transparency and accountability.