AI and the Environment: A Tumultuous Relationship

For nearly two years, there has been enthusiasm about generative artificial intelligence (AI) with the launch of ChatGPT on the web by OpenAI’s Sam Altman and its rapid adoption by internet users. Last month in Paris, under the presidency of France and India, the first international summit on AI took place.

The controversy is raging between those who are fervent believers in high-tech and the antagonists. For some, we are facing a new technological industrial revolution that will completely change our world in terms of society, work, education, industry, recreation, and security, for example. For others, it is synonymous with the annihilation of creativity, job loss, the dumbing down of youth, plagiarism, and a polluting sector.

Although the controversy is raging and will not cease for at least another 10 years, one thing is certain: AI will gradually become part of our daily lives. The numbers speak for themselves. According to estimates, nearly 70% of young people between the ages of 16 and 24 who have access to this new technology use it every day, whether for photo filters on Instagram or conversational chat queries. The number of users has increased by more than 60% in just one year.

The debate arising from this new technology is AI’s true purpose: a tool for the betterment of humanity, designed to make our lives easier, or just a mirage serving financial speculation while preying on the environment.

What Exactly Is AI?

One of the pioneers of AI, John McCarthy, admitted that he coined the term "artificial intelligence" in 1956 to secure funding for his projects during the Cold War, but that ultimately, there is nothing intelligent about it.

An acceptable definition is that generative artificial intelligence refers to a set of models and algorithms capable of creating new content (text, images, videos, music, code, among others) based on existing data through learning.

Generative AI relies primarily on deep learning models, particularly generative neural networks, such as Generative Adversarial Networks (GANs): two networks compete, one generating content and the other evaluating its quality; Variational Autoencoders (VAEs): they learn to compress and reconstruct data, thus generating credible variations; and transformer models ( GPT, DALL·E, Stable Diffusion, for example): based on transformer architecture, they enable the production of text or images with impressive coherence.

The main characteristics are the generation of original content based on training data, adaptability: it can be refined for specific tasks. Learning by imitation: it generates realistic content by drawing inspiration from training data. And dynamic interactions: enabling natural dialogue, the creation of realistic images, or even the generation of functional code.

But that is easier said than done. To achieve suitable products, the most powerful hardware is needed, which, unfortunately, means high energy consumption and a large ecological footprint.

It is no coincidence that one of the most significant measures taken by President Donald Trump at the beginning of his term was to announce the withdrawal of the United States from the Paris Climate Agreement and the grand announcement of an investment of over US$500 billion to finance the Stargate project, aimed at building the largest network of data centers to multiply the computing power necessary for advancing the sector.

This new technology has and will have a real impact on our lives and our environment. More and more data centers are needed, which are essential for AI development.

The equation is simple: data centers will become increasingly numerous and powerful, concentrating machines that produce a lot of heat that must be cooled. For AI needs, supercomputers are used, consuming five to 10 times more electricity than conventional data centers. In other words, they will need more and more energy and water. Google, in its 2024 annual report, revealed that it withdrew 28 billion liters of water to cool its data centers and that its water consumption increased by 82% compared to 2022.

This increase in water consumption can be explained by low-carbon consumption labels, where the focus is more on electricity consumption than on other resources. Indicators are used to measure the energy efficiency of data centers, such as Power Usage Effectiveness (PUE), and the trick to obtaining greener PUE scores is to use more water than electrical energy.

Estimates predict that soon, the electricity consumption of the sector will be equivalent to seven times the consumption of Denmark. That is why, during the summit, President Emmanuel Macron boasted that French electricity is the cheapest, most stable, and decarbonized. But the race for performance and innovation is in full swing. Data centers are popping up all over the world, with an annual growth rate of over 12%.

It is therefore imperative not to underestimate the problem. As mentioned earlier, AI relies on enormous volumes of data that are processed and stored. With innovations in the sector, computer learning must be continuously renewed (while simplifying the process). It must be understood that AI is not intelligent: it is a sequence of probabilistic and mathematical calculations based on algorithms, computing power, and datasets.

To give a few more illustrative examples, the rush for AI has driven big tech companies to explode their consumption, such as Microsoft, with a 30% increase in its carbon emissions between 2020 and 2023, and Google, with a 48% increase between 2019 and 2023. A single GPT-3 query consumes 4 watt-hours (Wh), the equivalent of one-third of a conventional smartphone battery charge (for an image, it is 60 times more). Twenty queries result in half a liter of water consumption. Finally, in the United States, it is projected that by 2028, 15% of the national electricity consumption will be attributed to the AI sector.

There are five main groups of environmental impacts:

• Electricity consumption.

• Consumption of mining resources for semiconductors.

• Water withdrawals for cooling.

• Electronic waste management.

• The misuse of AI.

This last issue is about whether AI is used properly, to improve well-being and the environment, or simply to serve the pursuit of profit for a handful of billionaires. Everything will depend on how it is used, although it is true that AI has the potential to facilitate environmentally friendly initiatives.

AI can play a crucial role in monitoring and protecting biodiversity, as seen with Wildbook, which uses image recognition to identify and track endangered species. It analyzes photos taken by tourists and researchers to survey animal populations, such as whale sharks and zebras. It can also be used to combat climate change, as demonstrated by DeepMind and IBM Green Horizon, which develop applications for optimizing renewable energy. DeepMind's AI optimizes wind turbine and solar panel management to maximize their efficiency. IBM Green Horizon predicts pollution levels and optimizes the use of clean energy in cities.

Other solutions help monitor the oceans and fight plastic pollution. The Ocean Cleanup, for instance, analyzes ocean currents to optimize the routes of ships collecting plastic waste. Microsoft’s PISCES uses AI and satellites to detect marine pollution zones and track oil spills. Ultimately, AI is a powerful ally in protecting the planet.

To conclude, let’s look at projections in Mexico. Although today the country ranks 20th worldwide in terms of the number of data centers, with 166 exactly, its strategic geographical location is attracting investments from Silicon Valley giants. In 2024, Microsoft inaugurated its first regional data center, "Mexico Central," in Queretaro, and Amazon is planning a multimillion-dollar investment in the coming years to create a data city for its Latin American cloud. However, environmental authorities must remain vigilant because Queretaro already concentrates more than half of the country’s data centers. With these projections, the environmental impact will be felt much more. Let’s not forget that this is a semi-arid region already experiencing water stress.

AI will either exacerbate the ecological crisis or help find solutions. Everything will depend on its good or bad use. Technology and innovation are only valuable if they are used and shared by all.