WIll AI deliver climate problems or solutions?

Anybody who has accessed the internet in the past year has interacted with Artificial Intelligence (AI). Research centres are implementing AI into their workflows to automate operations and inform on decision making processes. And in recognition of the speed at which AI is developing, the EU rolled out its AI Act to all 27 member states in February 2024.

At present, the prospect of what AI could achieve has propelled it to the peak of phase two in what is known as the Gartner Hype Cycle, a chart analysing the five phases of new technological innovations as they progress from the first stages of development to the final plateau of productivity and application. 

The full scale of AI’s application is yet to be determined and we are sitting in something of a honeymoon phase of high expectation. 

As a nascent technology, its potential is evolving, but so too are the implications of its use. The data centres that power AI and store the information it relies upon are energy-intensive, contributing to carbon emissions with each process AI delivers. What’s more, the water consumption associated with these data centres is sharply increasing.

While demand for AI and research into its use is on a rapid incline, so too are the environmental repercussions associated with this technology. 

Building AI into the climate research space 

"AI can be seen as a new microscope," explained Samuel Kaski, professor of computer science at Aalto University. "It’s a tool that can help in scientific and climate research. It is essentially a better assistant, generating data predictions and recommendations which will be essential for developing new science and research." 

According to Kaski, AI is an improved decision-making tool which can be applied to some of the more tedious roles in research, such as data management, collection and labelling. In terms of environmental modelling, AI is a design tool.  

"Any design problem is a decision-making process," he added. "For instance, how do you design aircraft in such a way that they are more fuel efficient? How do you structure traffic and public transportation systems to improve efficiency and services that will benefit urban areas?"

Presently, AI could produce numerous benefits in terms of better structuring of these systems, which would in turn generate climate wins for clean air and emissions.

However, the issue we face when applying AI to research is the time and resources that go into training larger models. Billion-dollar corporations have the finances and research teams needed to train large, complex AI models, which could take as long as six months. 

Smaller research institutions, however, may not have the resources to invest in this process. The outcome is that climate research centres working to generate environmental solutions may not yet have the resources to use AI as an effective decision-making tool, especially if they are not backed by significant investment.

The result is corporations leading in AI being able to choose how they use it, and if climate solutions and human-centric results aren’t at the top of their agenda, it will be a long time before we see AI used to generate large-scale, complex climate resolutions with genuine benefits for humanity. 

AI’s potential for climate solutions

While AI has had some effective climate applications, it is being applied to smaller developments as opposed to expansive and definitive answers to environmental issues. 

Recently, airlines have been using AI to track flight paths and journey times, reducing airtime and delays, which help improve fuel efficiency. A similar application can be seen as smart sensor AI technology is used in urban areas to inform public policy on pedestrianisation, thus reducing traffic and pollution in city centres by taking cars off the road. 

A project run by Microsoft and the Pacific Northwest National Laboratory used AI to detect a new substance which scientists believe could reduce lithium use by up to 70 per cent, a major win for the manufacture of green technologies which are reliant on rare earth materials, the supply of which has often hindered the rollout of these technologies. 

On a more localised scale, farmers are using AI to scan their fields and distinguish their harvestable crops from weeds, targeting the spray of pesticides on weeds alone and thereby reducing the use of pesticides in the farming process. Farmers are even able to scan their fields and have AI generate accurate recommendations as to when and where they should be planting their seeds to generate the most promising crop yields, a particularly crucial tool for farmers operating in challenging climates. 

These are but a handful of examples as to how AI can be tailored to deliver environmental solutions on a more localised scale. But while the potential benefits AI could unearth for humanity is driving its development, the risks and moral questions it raises are stirring significant debate. 

One of these risks is the associated carbon footprint and environmental impact AI is having. 

Water use

Understanding the full scale of environmental degradation and resource demand AI yields is trickier to grasp than the tangible, positive results of its application. The average user of AI can revel in its ability to generate a 1,000 word essay on the French Revolution that gets a passing grade, but disregard the amount of water it takes to cool an AI server. 

On average, a data centre with cooling towers uses up to 9 litres of water per kWh of energy used. This quickly stacks up over time, and the average Google data centre is reported to use as much as 1.7 million litres of water  per day.

Furthermore, a conversation with ChatGPT is estimated to devour around 500 millilitres of water when using between 5 to 50 prompts. 

The link between water and AI can also be seen in Microsoft’s 2022 Environmental Sustainability Report, as their water consumption increased by a third to 1.7 billion gallons, enough to fill over 2,500 Olympic swimming pools, between 2021 and 2022. 

This increase has been attributed to growth in generative AI. In a similar report, Google highlighted 20 per cent growth in their water use over the same period, a pattern that is similarly attributed to investment in AI.        

An energy intensive asset 

How, then, do we balance the informative and solution-driven qualities of AI with the present problems it is causing?

While the technology has demonstrated positive value in terms of carbon and resource savings, it is not yet clear whether the scale of these savings will exceed the sheer quantities of water and energy it consumes to power these decisions. 

"Most of [AI] is based on GPUs, which are processing units," Mauricio Alvarez, senior lecturer in machine learning at the University of Manchester, told FairPlanet. 

"To build these GPUs you will need materials and energy. It can take months to train these models, all the while energy is being used up, particularly for larger models. Once these models have been trained and are ready for use, you will have a rush of users wanting to access the model, which will generate energy demand to service these requests."

According to an electricity report from the International Energy Agency, electricity consumption from data centres, AI and cryptocurrency sectors could double by 2026, with data centres being the primary drivers of this growth. The electricity consumption of data centres alone could exceed 1,000 TWh in 2026, an energy demand roughly equivalent to that of Japan. 

The development of sustainable data centres, experts point out, is essential for mitigating the climate impacts of AI. This means powering data centres with renewable energy sources, such as solar and wind, and managing their efficiency to reduce energy consumption and waste. Organisations could further mitigate their climate impact by curating concise strategies for AI usage. 

"AI developments have this underlying philosophy of us needing to scale-up these models," Alvarez from the University of Manchester said. "Within a short period of time we’ve gone from models that use millions of parameters, to newer ones that are using billions.

"People believe scaling-up these models means gaining new possibilities, but there is a contentious claim about diminishing returns and how much extra resources we are using compared to what we get out of these models. Perhaps, in some cases, the resources we are getting out of AI do not justify the resources being put in."

AI has the potential to deliver real solutions and assist in research, policy and science surrounding climate change. But the full extent of its integration into this field remains uncertain, leaving unanswered questions about how its outcomes will be managed.

Image by Markus Spiske.