Claims from Big Tech that AI’s climate benefits will be able to offset its environmental strain have been dismissed as greenwashing in a new report.
The intense demands of water, land and energy required to run high-powered modern AI models have been a constant source of criticism for the technology. However, many industry figures in support of the expanded use of AI have claimed its ability to support climate change efforts makes up for its strain on the environment.
According to a report authored by climate and energy analyst Ketan Joshi, as much as 74% of claims about AI’s climate benefits are unproven and serve to downplay the climate harms of the technology to the benefit of tech and fossil fuel companies.
The research was commissioned by a consortium of environmental organisations including including Beyond Fossil Fuels, Climate Action Against Disinformation (CAAD),
Friends of the Earth U.S., Green Screen Coalition, Green Web Foundation, and Stand.earth.
It claimed that only 26% of claims cited published academic papers, while more than a third (36%) cited no evidence at all. In general, it stated, these claims rely on “weak forms of evidence rather than robust, peer-reviewed academic papers”.
In Joshi’s study, he examined the types of AI underpinning common environmental claims, looking at the strength of the evidence behind them. His analysis found no example where consumer generative systems such as ChatGPT and Gemini, were leading to verifiable and substantial levels of emissions reductions.
“It appears tech companies are using vagueness about what happens within energy-hogging data centres to greenwash a planet-wrecking expansion,” Joshi said.
“This has bled through into organisations like the International Energy Agency. The promises of planet-saving tech remain hollow, while AI data centres breathe life into coal and gas every day.
“These claims of climate benefit are unjustified and overhyped, and could cover up irreversible damage being done to communities and society.”
The full report can be read here.
