How artificial intelligence harms the environment – 09/08/2023 – Science

There is no doubt that artificial intelligence (AI) is a booming industry. Thanks, in part, to the enthusiasm for new tools like ChatGPT, conceived by the American company OpenAI, which is financed by Microsoft and co-founded by billionaire Elon Musk.

ChatGPT has the ability to chat, text, and compose poems and essays in a surprisingly human way. And that has set off a race among tech giants to bring similar and possibly more sophisticated products to market.

With that, investments in AI are also growing rapidly. Today, it is estimated that this market is valued at 142.3 billion dollars (129.6 billion euros). It is expected to grow to nearly 2 trillion by 2030.

AI systems are already present in everyday life in many ways. They help, for example, people, governments or companies to work more efficiently and make decisions based on data. But they also have their disadvantages.

Environmental impact

For AI models to do their jobs, they need to process mountains of data. Or rather: be “trained” for it. To learn to recognize an image of a car, an algorithm needs to examine millions of images of cars. In the case of ChatGPT, it is stocked with vast databases of text from the internet to learn human language.

The development of this occurs in datacenters (server rooms), which demand enormous computational capacity and consume a lot of energy.

“The joint infrastructure of computer centers and data transfer networks is responsible for 2% to 4% of global CO2 emissions worldwide [número próximo ao do setor da aviação]. It’s not just AI, but it’s a big part of the problem,” says Anne Mollen, a researcher at the Berlin-based non-governmental organization Algorithmwatch.

In a 2019 study, researchers at the University of Massachusetts found that training a common large AI model can emit up to 284 tonnes of CO2 equivalent – ​​nearly five times the emissions of a car over its entire lifetime. useful, including manufacturing.

“When I first read this data, I was really shocked,” says Benedetta Brevini, associate professor of political economy of communication at the University of Sydney, Australia, and author of “Is AI good for the planet?” (Is AI good for the planet?).

“If you take a plane from London to New York, the carbon emissions will be 986 kilos. To train an algorithm, however, we emit 284 tons. Why don’t we talk about how we can reduce that carbon footprint?”, asks Brevini .

Still, the estimate based on the study done in Massachusetts referred to a particularly energy-intensive AI model. Smaller models consume less and can run on a laptop. But models that work with so-called deep learning, such as algorithms that curate social media content or ChatGPT, require a considerable amount of processing power.

In addition to the “training phase” of the technology, other emissions occur when the model is applied in the real world, which can happen billions of times a day – for example, every time an online translator translates a word or a chatbot responds to a query. question. According to Mollen, this phase can be responsible for up to 90% of emissions in the lifecycle of an AI.

How can AI’s carbon footprint be reduced?

All environmental issues must be taken into account from the beginning, including in the algorithm design and training phase.

“We have to take into account the entire production chain and all the associated environmental problems, especially energy consumption and emissions, but also the toxicity of materials and electronic waste”, argues Brevini.

Rather than developing ever larger AI models as the current trend points out, Mollen suggests that companies use smaller models with reduced datasets and ensure the AI ​​is trained on the most efficient hardware available.

Using data centers in regions that rely on renewable energy and don’t require a lot of water for cooling can also reduce AI’s carbon footprint.

Huge installations in the United States or Australia, for example, where fossil fuels make up a large part of the energy mix, produce more emissions than those in Iceland, which uses a lot of geothermal energy, and cooler outdoor temperatures make it easier for servers to cool .

Mollen points out that tech giants have a very good track record when it comes to using renewable energy in their operations. Google, for example, claims to have a zero carbon footprint thanks to investments in offset measures and aims to operate on CO2-free energy by 2030.

Microsoft has committed to being CO2 neutral by 2030 through the use of technologies such as carbon capture and storage. And Grupo Meta plans to achieve a zero net footprint across its entire value chain by 2030.

Energy, however, is not the only aspect regarding the environmental impact of AI. The huge amounts of water that datacenters need to keep servers from overheating cause major problems in regions with water shortages, such as Santiago, Chile, where the presence of Google’s datacenter “aggravates the drought, and local communities are protesting against the center and building new datacenters”, says Mollen.

Climate and environmental damage

Even if the big tech companies reduce the power consumption of their AI, there is another problem. And that can be even more damaging to the environment, points out David Rolnick, an assistant professor in the Department of Computer Science at McGill University in Canada and co-founder of the organization Climate Change AI. As an example, he cites the use of algorithms in advertising: “They are intentionally designed to increase consumption, which certainly comes at a significant cost to the climate.”

Rolnick also highlights a report by technology consultancy Accenture and the World Economic Forum, which predicts that AI and advanced analytics will help the oil and gas industry earn an estimated $425 billion by 2025.

Greenpeace, on the other hand, has been sharply critical of AI contracts entered into between fossil fuel companies and Amazon, Microsoft and Google. In a report, the environmental organization said Shell, BP and ExxonMobil would use AI tools to expand their operations, cut costs and, in some cases, increase production. Such contracts would “significantly undermine the climate commitments made” by the tech giants.

Google has since declared that it will no longer develop custom AI tools to help companies extract fossil fuels.

The rapid development of AI

The role of AI is likely to become even more significant in the future. And that’s another reason why Rolnick believes more regulation is needed to ensure that development is sustainable and doesn’t make it harder to meet greenhouse gas emissions targets.

“It’s a matter of prioritizing, intervening early and shaping the decisions that will be made,” he says.

In the European Union (EU), legislators have been working for two years on a law that tends to be a landmark in the area. AI must be regulated, and the respective risks of the various applications must be classified. So far, however, it remains unclear whether environmental concerns will also be factored into the bill.

Meanwhile, other governments are also working to regulate the use of AI. On the one hand, to leverage innovation in the industry and reap the benefits of technology. On the other hand, to avoid possible threats and protect its citizens.