By Oscar Braun
Every day, the use of artificial intelligence – particularly large language models (LLMs) – plays an increasingly significant role in our lives, aiding us with the most mundane tasks, speeding up the analysis of data queries from what used to take days into just minutes. Naturally, the profound analytical power of this technology has led many to question how it can be harnessed for good, and particularly, how it might be applied to generate solutions to the ongoing climate crisis. Many industry leaders, like OpenAI CEO Sam Altman, are prophesying the growth of this technology as a catalyst to finding solutions to the issue of climate change by accelerating the coming of the “intelligence age”. But AI’s environmental impact is significant and complex, and it is not yet clear whether its costs outweigh the societal benefits that are promised.
Revolutionising green technologies
AI is evolving at an exponential speed; its capacity to process and analyse data increases each day as different models are trained with more real-world data. Climate scientists are working on ways to harness this power to transform how we undertake climate science, through improving predictability, optimisation, and recording of data. Utilising this technology opens many opportunities for learning more about the planet and pushing the boundaries of climate science. With processing abilities 10,000 times faster than a human, with the right algorithms, AI can find out within seconds where and how fast the world’s icebergs are melting, precisely predict deforestation using patterns from satellite data, and calculate weather patterns before extreme events, allowing increased levels of disaster preparedness. All of these schemes have been successfully trialled; the British Antarctic Survey, for example, has developed a predictive tool called IceNet that forecasts changes in sea ice levels, informing governments of risks in sea level rise in a quantifiable and objective way.
The optimisation function of AI within climate science can also allow for energy in national power grids to be distributed in the most efficient way, enabling maximum energy savings, using techniques like optimising storage and load balancing. In industry, AI can help optimise supply chains, cutting emissions and waste in transportation, as well as forecasting demand to avoid overproduction. In financial markets, AI can also optimise algorithms to improve the efficiency of carbon markets, as well as climate-related financial models for organisations.
Finally, AI can help in the recording of data, detecting changes in images, as well as helping to make data more understandable and processable for scientists, for example, by removing cloud cover from satellite images. Practically, AI can also help remove errors in data measured from the environment, particularly in fields like air and water monitoring where anomalies can make data harder to interpret and trends harder to discern.
The drawbacks and dangers
Despite these opportunities for progress, the drawbacks and dangers of AI are wide-ranging. From reducing our cognitive capacity through overreliance, to ethical concerns surrounding automation and the job market, policymakers are faced with a variety of urgent legislative questions. One aspect that remains relatively unchallenged by policymakers, however, is the environmental degradation that comes with this explosion of AI models. The damage to the environment comes through three main areas of development: training, maintenance, and use. Additional concern should be paid to the environmental inequality that AI can perpetuate.
To develop functioning LLMs, models must be trained on real-world data, with each one requiring the same amount of electricity that could power 130 US family homes for one year. To put this in perspective, there are at least 10,000 different AI tools now, and that number is increasing daily. AI models are also constantly improving; they are like software updates in that they have a short shelf life and will soon need to be replaced by updated and retrained models, further increasing energy demand.
When maintaining these centres, it is not just energy that is required, but water too. Each kilowatt-hour (kWh) of energy consumed requires approximately 2 litres of water for cooling. For visual reference, by 2030, data centres are expected to use 945,000,000,000 kWh of energy, which amounts to 1,890,000,000,000 litres of water per year. To exacerbate this problem, around two-thirds of data centres within the US are in areas already facing water scarcity, such as Arizona and California. With the effects of climate change bringing longer periods of prolonged drought, data centres are drawing water from areas that are already suffering.
Data centres have existed since the 1940s, albeit on an experimental scale, and have since expanded to take on half of the world’s data storage. Once built and in use, they require an incredible amount of electricity to operate. One prompt can demand up to ten times the power requirements of a single Google search. This may not seem large, but the problem arises when the use of chatbots such as ChatGPT is framed as an iterative process, where 20 or even 30 prompts are required to get to your desired output, each input putting extra strain on a data centre being pumped with water and energy. With the benefits these models bring to the consumer, there is virtually to no incentive to cut back on use. Instead of reducing AI use, which would put the profits of invested stakeholders at risk, some companies are exploring the potential of nuclear technologies to generate the additional power needed. This will take many years, and in the meantime, most new mega data centres will be powered by fossil fuels, especially in developing countries.
The final issue to consider in AI’s battle to save our warming planet is the gap between developed and developing countries. Developing countries will undoubtedly face the effects of climate change most severely, with limited infrastructure, funding, and policy measures in place. It would therefore be prudent to focus the use of AI on aiding developing countries in coping with their forthcoming climatic changes; however, the ‘positive’ impact of AI is not uniform across the world. The sustainability of data centres worldwide is unequal. Google, for example, has centres all around the world. In Finland, 97% of the energy used comes from renewable sources, but in Asia, that figure falls to between 4-18%. The continued growth of centres in less developed countries, where startup costs are lower and renewable energy infrastructure is less common, will only exacerbate current energy and water scarcity challenges. Furthermore, advancement in AI is seen to be perpetuating the “digital divide”, representing differences in access to technology. Less developed countries already have less access to the internet and hence, less access to AI tools. For scientists in these countries, reduced access may slow or even stop the deployment of AI tools in the face of climate change, forcing them to rely on foreign technological assistance.
The future of AI in the climate crisis
AI is becoming more powerful every day; its capability and reach are expanding exponentially. There is a choice: either to use this immense computational power to speed up processes that would have taken scientists years, allowing huge progress in the study of our climate and related data, or to use it in large organisations to grow profits more efficiently. The problem is, there is no control over how these AI tools are used once they are published, often as open-source programs. Paradoxically, something so useful can also be so damaging to our climate.
Many have argued that we already have the tools to fight climate change, and there is already the technology available to us. What is missing is drastic policy action; the current attitude of inertia is holding us back. No AI tool can solve the problem of human disagreement or resolve a debate in parliament. AI is brilliant at optimising, analysing, and calculating, but the problem of climate change is ultimately a human problem that can only be solved by humans. New technologies should always be utilised and embraced sustainably to help the fight against a planet that is becoming increasingly unlivable- but early assessments indicate that AI brings sustainability challenges of its own.
The views expressed in this article are the author’s own and may not reflect the opinions of The St Andrews Economist.
Image Source: Business Today Malaysia