At a time when artificial intelligence is becoming increasingly popular and widespread, it is worth considering its impact on the environment. Although we associate AI mainly with the digital dimension, the development of machines is inevitably associated with a huge demand for energy, requires increasingly efficient cooling systems and therefore large amounts of water, and is responsible for significant carbon dioxide emissions. The climate cost of artificial intelligence is a fact that fortunately more and more AI leaders are becoming aware of. As a result, they are planning and taking steps to reduce the negative impact of AI on the environment.
AI on the rise
We are undoubtedly living in a time of rapid development in artificial intelligence. Our daily lives are increasingly being accompanied by new and innovative solutions to make our lives easier. It has never been easier or more accessible to create ‘tailor-made’ content. In just a few years, AI has become an integral part of our lives and the number of users of solutions such as ChatGPT or Microsoft Bing’s Image Wizard is growing exponentially.
As a result, the industry is booming and making record profits. The scale and speed of this phenomenon can be seen in the huge success of ChatGPT, which celebrated its first official anniversary a few days ago.
Not so long ago, few people were interested in the potential of large language models. Today, it could be said that the world has gone mad for the omniscient e-companion. According to the latest figures available, ChatGPT now has around 180.5 million users; OpenAI’s chief executive, Sam Altman, estimates that it receives 100 million visitors a week. In October alone, the site had a whopping 1.7 billion page views.
Development at the expense of the environment
However, the range of opportunities offered by artificial intelligence should not overshadow the cost of implementing the innovation. AI is being developed in huge research and development complexes that use an equally huge amount of electricity.
A study by the University of California, Riverside, found that training GPT-3, the predecessor of ChatGPT, consumed 1287 MWh and resulted in the emission of the equivalent of more than 550 tonnes of carbon dioxide.
This is roughly equal to 550 round trips between New York and San Francisco.
By comparison, GPT-4, the successor to the current version of Chat, is reported to be trained on about 570 times more parameters than GPT-3. This does not indicate a proportional increase in energy consumption, but it does suggest that Chat’s learning process is becoming more energy intensive.
The figures for water consumption for cooling the server rooms are also interesting. GPT-3 training at Microsoft’s state-of-the-art data centres in the US can consume a total of 5.4 million litres of water, of which 700,000 litres is used to cool the servers on site.
In addition, the GPT-3 needs to ‘drink’ about a 500ml bottle of water for about 10-50 generated responses, depending on when and where it is used. These numbers may increase with the newly launched GPT-4 due to the aforementioned greater number of parameters used to train the language model.
Staying on the subject of Big Tech’s water consumption, this issue is addressed in Google’s latest Environmental Report, which shows that the tech giant consumed 21.2 billion litres of water in 2022.
This represents an increase of almost 5 billion litres compared to 2021 and 8.3 billion litres compared to 2020. The 2022 figure can be compared to the water consumption of London, a city of nine million people, for 10 full days.
Carbon-free and Big Tech’s green plans
While the figures above are essentially bleak, they should not obscure the changes being made by major players in the technology industry.
To reduce their environmental impact, scientists and engineers are looking at ways to make AI models more energy efficient, develop algorithms that require less computing power, or reduce CO2 emissions by powering data centres with green energy.
In September, Apple unveiled a new range of watches that it claims will be the first carbon-neutral products. To reduce the carbon footprint of these products, Apple has relied on the partial use of recycled materials and reduced air transport. The giant also announced an update to its environmental commitment to reduce greenhouse gas emissions by 90 % by 2050.
Similar climate action is being taken by Microsoft, which has pledged to become carbon-free by 2030  and to produce more water than it consumes.
Finally, there is Googles promising investment in geothermal energy. At the end of November, Google announced the launch of a geothermal project in partnership with Fervo, which has resulted in carbon-free electricity (CFE) being supplied to the local grid for the company’s data centres in Nevada.
Google, like the rest of the Big Five, has big plans for decarbonisation. By 2030, it aims to operate all of its data centres and office campuses on 24/7 carbon-free energy.
The regulations currently under consideration to reduce the negative impact of AI on the environment should also not be overlooked.
For data-driven companies, concern for the environment is already part of the activities regulated under ESG requirements. This includes, for example, energy efficiency or the use of renewable energy sources in data centres.
With the number of AI models growing every day at a rate unmatched by any other technology developed to date, there is no denying that the climate costs associated with AI will become increasingly important.
When weighing the opportunities and benefits of AI against the potential risks that are so often discussed, for example in relation to jobs, the environmental costs cannot be overlooked.
As artificial intelligence becomes an integral part of our lives, it is important that its development is accompanied by a commitment to sustainability and minimising any associated impact on the environment. The leading companies responsible for AI-based solutions are showing awareness and initiative in this area. This gives us hope for a future where we can enjoy the benefits of technological progress without further damaging our planet.
Questions? Contact us
 P. Li et al., Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models, p. 7.
 P. Li et al., Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models, s. 3.