The report says tech companies are confusing traditional artificial intelligence with generative AI by claiming their energy-intensive technologies could help avert climate change.
Most claims that AI can help avert climate change refer to machine learning, rather than the energy-guzzling chatbots and image-generating tools that fuel the explosive growth of gas-guzzling data centers, an analysis of 154 statements found.
The study, commissioned by nonprofit organizations including Beyond Fossil Fuels and Climate Action Against Disinformation, found no examples of popular tools such as Google’s Gemini or Microsoft’s Copilot leading to “material, verifiable, and significant” reductions in global warming emissions.
Ketan Joshi, an energy analyst and author of the report, said the industry’s tactics were “diversionary” and relied on proven techniques that amounted to “greenwashing.”
He likened this to fossil fuel companies promoting modest investments in solar panels and exaggerating their carbon capture potential.
“These technologies avoid only a small portion of emissions compared to the large amount of emissions from our core operations,” Joshi said. “Big tech companies have taken that approach, upgraded it, and expanded it.”
Most of the scrutinized claims were based on International Energy Agency (IEA) reports reviewed by big tech companies, as well as company reports from Google and Microsoft.
The analysis found that the IEA report devoted two chapters to the potential climate benefits of traditional AI, and was split almost evenly between claims based on academic publications and company websites, and claims without evidence. In the case of Google and Microsoft, most of the claims had no evidence.
The analysis, released during the AI Impact Summit in Delhi this week, argues that by “confusing” types of AI, the tech industry has misleadingly presented climate action and carbon pollution as a package deal.
Sasha Luccioni, Head of AI and Climate Change at Hugging Face, an open source AI platform and community. He, who was not involved in the report, said it added nuance to a discussion that has often lumped together very different applications.
“When we talk about AI that is relatively bad for the planet, it’s mainly generative AI and large-scale language models,” Luccioni said, calling for industry to be more transparent about its carbon footprint.
“When we talk about AI that is ‘good’ for the planet, it is often predictive models, extractive models, or old-school AI models.”
Even for traditional AI, the analysis found that green claims tend to rely on weak forms of evidence that have not been independently verified. Of the green claims studied, only 26% cited published academic research, and 36% cited no evidence at all.
One of the earliest examples identified in the report was the widespread claim that AI could help reduce global greenhouse gas emissions by 5-10% by 2030.
Google reiterated this number last April, according to a report it commissioned from consulting firm BCG, citing a blog post the company wrote in 2021 that attributed the numbers to its “experience with customers.”
Data centers consume only 1% of the world’s electricity, but their share of U.S. electricity is expected to more than double to 8.6% by 2035, according to BloombergNEF. The IEA predicts that electricity will account for at least 20% of the growth in electricity demand among the wealthy by the end of the 20th century.
While the energy consumption of a simple text query against a large-scale language model such as ChatGPT may be as low as turning on a light bulb for a minute, complex functions such as video generation and in-depth research can significantly increase energy consumption, as partial industry disclosures suggest, and the speed and scale of its growth has troubled some energy researchers.
“Our estimated emissions reductions are based on a robust empirical process based on the best available science, and we transparently share the principles and methodology that guides us,” a Google spokesperson said.
Microsoft declined to comment, and the IEA did not respond to requests for comment.
Joshi said the debate over the climate benefits of AI needs to be “bringed back to reality”.
“Incorrectly linking big problems to small solutions distracts from the highly preventable harm caused by unrestricted data center expansion,” he said.
