As illustrated by the success of Large Language Models (LLMs), particularly in chatbots such as ChatGPT, the use of Artificial Intelligence (AI) has become increasingly widespread in recent years, with the industry growing from a market size of $40 billion in 2022 to a projected size of $1.3 trillion in the next ten years. Much of the discussion surrounding AI technology focuses on its application in terms of improvement to productivity. Yet, the use and creation of AI in ChatGPT and other similar LLMs are highly water and energy intensive. The sustainability of the new technology is increasingly questioned. Amongst concerns are whether its potential merits come at too great an environmental price. This blog post outlines contemporary AI sustainability issues and assesses the prospects for their future mitigation.
What is unsustainable Artificial Intelligence (AI)?
When using a virtual service many consumers are unaware of the direct physical impact of their decisions. However, the physical impact of AI on the planet is significant. The three major environmental costs that accompany AI are energy consumption, water consumption and CO2 emissions.
Energy consumption:
Training a singular large language model like ChatGPT requires an equivalent electrical consumption to over 300 three-bedroom houses in England for a year. Asking ChatGPT a question can be 10 times more energy-intensive than doing a search on Google. With 100 million weekly users, this increase is certainly not trivial.
Water Consumption:
Data centres use water as a coolant to manage the high temperatures generated from AI components. With water being used every time the AI receives a query, water consumption is predicted to rise to 6.6 billion cubic metres by 2027. With two billion people globally not having access to safe drinking water, there is a strong ethical impetus to find an alternative measure.
CO2 Emissions:
The training of OpenAI’s GPT-3, the third generation of the Generative Pre-trained Transformer model, emitted 502 metric tons of carbon. ChatGPT gets over 10 million queries a day, with each query emitting 4.32g of CO2. An alternate energy source must be considered, especially when this use of AI will likely grow with an increase in chatbot proliferation in the future.
Big Tech's roadmap on AI sustainability
Big Tech companies such as Google, Meta, Amazon, and Microsoft have implemented AI into their businesses, creating their own models. Their corporate sustainability reports share how they plan to use AI to help the environment and make AI itself more sustainable.
Amazon's report stated that its AI model helped create a new, more efficient packaging strategy, reduce packaging materials, and identify when recyclable paper and cardboard packaging can be used. To combat AI's environmental impact, Amazon has opted to use carbon-free energy where viable.
Meta has highlighted the carbon emissions from concrete production and is using AI to refine the concrete formula, reducing its carbon footprint by up to 40 per cent, and they plan to deploy the AI model worldwide. They are also using AI to map all the forests on Earth, gathering data for more accurate calculations of how much carbon forests capture.
Big tech companies are using AI to boost sustainability and offset their high carbon emissions in other processes. Amazon, for instance, aims to reach net zero by 2040 by ‘leveraging AI’, as seen with their packaging scheme and Google have cut 2.4 million metric tons of CO2e emissions with AI-driven fuel-efficient routing.
However, corporate sustainability reports offer little data on AI specific emissions, making it hard to assess whether mitigations in some areas justify using AI when it is so energy intensive. With growing energy demands, transparency surrounding this data would help experts establish whether these companies can keep their emissions sufficiently low to maintain sustainability.
Are there any EU legislative acts in place to mitigate AI's effect on the environment?
The European Green Deal initiative Fit for 55 makes it a legal obligation for the EU to reduce emissions by 55 per cent by 2030. In 2024, data centres emitted 200 million tons of greenhouse gases (GHG), accounting for two per cent of global emissions. In response, companies such as Amazon and Meta have attempted to match their energy consumption with energy certificates. However, these net reductions within firms do not directly affect AI emissions.
Another issue is how the EU Emissions Trading System (ETS) governs AI emissions. The ETS is a cap-and-trade system that limits the amount of GHG emissions, as outlined in the EU ETS directive. AI is subject to the ETS directive when operating within regulated sectors. This indirectly limits AI, as exceeding the emissions cap results in direct financial sanctions for the offending firm. However, when AI is used outside of regulated sectors, it does not fall under the scope of the ETS. In its current state, the ETS does not comprehensively regulate AI's environmental impact.
Some areas of current legislation do not set sufficiently stringent restrictions on AI emissions. With AI’s increasing energy demands, more specific rules should be included to ensure that AI can have a net positive environmental impact.
Can we solve AI's love-hate relationship with sustainability?
AI is still in its infancy, and although there are valid sustainability concerns, I believe that it still has the potential to provide environmental benefits beyond the industry. However, there is still room for improvement in AI regulation, as the recently adopted EU Artificial Intelligence Act failed to outline a legislative AI sustainability framework. The Act makes minimal reference to sustainability, failing to outline standards that would effectively mandate logging of emission data and reductions of energy and resource use.
With a significant increase in AI’s use, more transparency is necessary. Data on emissions will be invaluable in aiding legislators to create more targeted regulations. Mitigation of AI’s adverse effects on the environment must be prioritised for us to effectively benefit from what AI has to offer for a more sustainable future.
