Study criticizes the AI industry's sustainability promises

Environmental associations and NGOs have accused large AI companies of whitewashing the climate impact of their applications. Technology companies often justify the enormous energy consumption of their new data centers with the argument that artificial intelligence is a key tool for tackling the climate crisis. However, an investigation by several non-governmental organizations, including AlgorithmWatch and Beyond Fossil Fuels, has now come to the conclusion that these claims are based on weak data. The authors of the study accuse the industry of concealing environmental damage through misleading communication.

Distinction between conventional and generative AI

A central point of criticism in the study is the lack of differentiation in the concept of artificial intelligence. The study shows that the positive climate effects propagated by companies such as Google and Microsoft relate almost exclusively to "conventional" AI applications. These include weather forecasting models, for example.

However, the current boom and the associated massive expansion of data centers are primarily driven by so-called "generative" AI for end users - i.e. systems such as ChatGPT, Copilot or Gemini, which generate texts, images and videos. The authors of the study were unable to find any examples of these resource-intensive applications that demonstrated a verifiable and substantial reduction in greenhouse gas emissions.

The authors describe the linking of the climate benefits of conventional AI with the expansion of generative models as a new form of "greenwashing". This refers to the strategy of creating a more climate-friendly image by making misleading, unclear or unsubstantiated claims about alleged environmental benefits in order to distract from the environmental damage actually caused.

Thin evidence for climate promises

For the analysis, 154 high-profile claims by tech companies and institutions about the positive climate effects of AI were evaluated. The result shows a clear discrepancy between promises and scientific evidence: only 26% of the statements examined were based on published scientific studies. For 36 percent of the claims, no evidence was cited at all, while the majority of the remainder merely referred to the companies' own websites or reports.

The authors conclude that even in the case of conventional AI, the alleged benefits for the climate are often greatly exaggerated, while the negative effects of AI growth are clearly measurable. Julian Bothe, Senior Policy Manager at AlgorithmWatch, said that if there are sustainability benefits from AI, it is through applications of traditional AI with low resource consumption. "The big language and image-generating models such as ChatGPT, which are the main focus of the current AI hype, consume vast amounts of electricity and water, cause CO2 emissions equivalent to entire countries, but bring no positive benefits for the environment."