This post is also available in: heעברית (Hebrew)

In this time when people, companies, and governments are obsessed with generative artificial intelligence, the intense interest has brought with it fears of the tech’s impact on employment, misinformation, etc, as well as anxious predictions about an existential risk to humanity.

Expert John Naughton of the Guardian claims that all this discussion is diverting attention from the downsides of the technology that are already impacting the world: bias, inscrutability, unaccountability and so much more. What is barely discussed at all, unfortunately, is the technology’s environmental impact on the planet, which will be significant at best and extremely problematic at worst.

The reason for this, according to the Guardian, is that the technology consumes electricity at a colossal rate. AI needs incredible amounts of computing power, which in turn needs electricity, which runs hot and needs cooling using water and other electricity. All this means large-scale CO2 emissions.

The implications of this are dire: if the industry’s dream of “AI everywhere” is realized, it would cause a world that is dependent on a technology that has a major growing environmental footprint.

If you’re asking yourself “Why isn’t anybody talking about this?!”, then fortunately some people are- a study performed in 2019 estimated the carbon footprint of training a single large language model at the equivalent of 125 round-trip flights between New York and Beijing. It is important to note that models have become much, much bigger since then, and their training footprints have proportionately grown.

So, what happens when the AI is done training and goes into service? Well, the researchers trying to estimate have found that generative tasks (meaning text generation, summarizing, image generation, and captioning) require more energy than other tasks, and that tasks involving images emit more carbon than ones involving only text.

There currently is no solution, but Naughton hopes that the AI craze might “die down” and won’t pose such an environmental danger. Regardless, this is definitely a topic that should be discussed and brought to the attention of the consumers of AI tools.