Will AI deliver a net benefit for the planet or is this claim just misdirection? The answer is explored in a new report with a hard-hitting title: “The AI Climate Hoax: Behind the Curtain of How Big Tech Greenwashes Impacts”

It is written by Ketan Joshi, in partnership with Friends of the Earth, Beyond Fossil Fuels, Stand.Earth, Climate Action Against Disinformation, the Green Screen Coalition, and the Green Web Foundation.

This blog summarises its three key findings.

1. AI is an environmentally destructive heavy industry

Perhaps not everyone would immediately think of AI as heavy industry, but this report explains that AI data centres require concrete, steel, machinery, and extraction of raw materials for manufacturing computer chips – just at the construction stage. Data centres themselves create a new demand for fossil fuels and vast volumes of water (many imposed on communities in already water-stressed regions).

This massive industry appears to have emerged “in the blink of an eye”. The report claims that it is the novelty of generative AI and the intensity of data centre expansion that have lent themselves to greenwashing tactics, that many people haven’t yet been able to decipher.

2. Not all AI is the same

The report highlights a quote by Karen Hao (the author of “Empire of AI”):

“People think that to get any benefit from AI in general, which is a huge category of technologies, they somehow have to accept the most environmentally egregious form of AI. I really reject that premise.

The term AI is so vague that it’s like the term transportation. You could be talking about a bicycle or a rocket, and those are fundamentally different forms of transportation. And in AI, we mishmash a bicycle in the rocket into the same, broad category.

But that doesn’t mean that the benefit from using the bicycle also justifies us building the rocket”

Traditional AI is a type of AI that can be used to help climate goals (and 97% of climate claims are related to traditional AI), while this report found no examples where consumer generative systems such as ChatGPT, Gemini or Copilot led to a verifiable and substantial level of emissions reductions.

By 2030, projected energy consumption of generative AI will be 13 times higher than traditional AI.

Using the earlier analogy, traditional AI would be a bicycle, while generative AI would a rocket, probably flying to Mars, the benefits of which, if any, are very remote.

Although this analogy would only work if traditional AI was used purely ‘for good’, to find ways to reduce emissions, whereas in reality it is also used by fossil fuel companies to streamline exploration and increase extraction.

3. Climate claims should be challenged

The environmental damage of AI is often justified on the grounds that AI will ultimately help deliver net benefit to climate action, but is it just misdirection?

The report examined this premise in detail and found that:

• Only 26% of the climate claims in relation to AI (made by big tech companies) cited published academic research, with 36% citing no evidence at all.

• Many technology companies share relatively meaningless “disclosures” of AI-related sustainability information. One recent example is Google sharing the carbon emissions of a single chatbot query, but not disclosing the entire system’s carbon footprint, nor the impacts of images, videos or longer queries.

• Many companies artificially lower their reported emissions using renewable energy certificates while their actual operations are powered by fossil fuels.

• The evidence of substantial harm is strong, clear, and proven. The evidence of any massive climate benefits of generative AI is weak and often “based on wishful thinking.”

This report used “the practice of making unclear or not well- substantiated environmental claims” as their definition of greenwashing (from a proposal for the EU Green Claims Directive). With that in mind, it concluded that climate AI claims are indeed a new form of greenwashing used by the tech industry. The report challenges the reader not to accept them on face value.

You can read the full report here.

I can also recommend Sasha Luccioni’s recent TED talk: “We’re doing AI all wrong. Here is how to get it right.”