Assertions that AI can address climate issues labeled as greenwashing

Tech companies are merging the concepts of traditional artificial intelligence (AI) with generative AI in their claims that this energy-intensive technology may help mitigate climate change, according to a recent report.

Most corporate assertions that AI can assist in combating climate change center around machine learning technologies, rather than the energy-intensive chatbots and image-generating tools responsible for the sector’s rapid expansion of energy-consuming data centers. This conclusion emerged from an analysis of 154 statements.

The study, commissioned by various nonprofit organizations, including Beyond Fossil Fuels and Climate Action Against Disinformation, revealed no instances where popular tools like Google’s Gemini or Microsoft’s Copilot led to a “material, verifiable, and substantial” drop in environmentally harmful emissions.

Ketan Joshi, an energy analyst and the report’s author, criticized the industry’s narrative as “diversionary,” employing methods that can be categorized as “greenwashing.” He drew a parallel to fossil fuel companies that highlight minor investments in solar technologies while exaggerating the potential efficacy of carbon capture.

“These technologies produce negligible reductions in emissions compared to the significant emissions from their core activities,” Joshi remarked. “The tech industry has adopted this strategy and scaled it up.”

Most analyzed claims originated from a report by the International Energy Agency (IEA), which had been scrutinized by top technology firms, alongside corporate documents from Google and Microsoft.

The IEA report included two chapters dedicated to the climate advantages of traditional AI but displayed a roughly equal distribution between claims grounded in academic studies, corporate websites, and those lacking evidence altogether. For both Google and Microsoft, the majority of claims were unsupported by evidence.

Released during the AI Impact Summit in Delhi, the analysis contends that the tech sector has misleadingly framed climate solutions and carbon emissions as interlinked by “muddling” different types of AI technologies.

Sasha Luccioni, the AI and climate lead at Hugging Face, an open-source AI platform and community, who was not directly involved in the report, mentioned that it brings nuance to a discussion that has frequently categorized very different applications as a single entity.

“When we refer to AI that is relatively detrimental to the planet, we mean generative AI and large language models,” Luccioni explained. She has advocated for the industry to be more forthcoming about its carbon emissions. “Conversely, when we consider AI that benefits the planet, we’re typically discussing predictive models, extractive models, or traditional AI technologies.”

Even claims about traditional AI’s environmental benefits tended to rely on weak evidence, with only 26% of the green claims analyzed having academic research to back them, while 36% were unsubstantiated.

One of the earliest claims addressed was a general assertion that AI could help alleviate 5-10% of global greenhouse gas emissions by 2030. This figure, which Google repeated as recently as April, originated from a report commissioned by the company from BCG, a consulting firm, referencing a blog post from 2021 that suggested the figure based on their client interactions.

Data centers currently use around 1% of global electricity, but this share is expected to soar to over 8.6% in the U.S. by 2035, according to analyses by BloombergNEF. The IEA foresees that these data centers will account for at least 20% of electricity demand spike in developed nations by the end of the decade.

A simple text query to a large language model like ChatGPT may consume energy equivalent to running a light bulb for a minute, but preliminary industry disclosures hint that the energy demand significantly increases for more intricate tasks like video generation and deep research. This has created concerns among some energy experts about the rapid and extensive growth of this energy consumption.

In response, a Google spokesperson commented, “Our estimated emissions reductions are grounded in a robust substantiation process based on the best available scientific evidence, and we have openly shared the principles and methodologies that inform this.”

Microsoft chose not to comment, and the IEA did not respond to inquiries for feedback.

Joshi emphasized the importance of reevaluating the conversation surrounding AI’s environmental benefits. “The misleading connection between a significant issue and a minor solution serves only to divert attention from the avoidable harms caused by unchecked data center proliferation,” he noted.