Home Technology Carbon footprint will plague the AI ​​industry

Carbon footprint will plague the AI ​​industry

Carbon footprint will plague the AI ​​industry

The level of CO2 emissions generated when creating specific AI models is starting to worry environmentalists (photo: CC0 Public Domain)

How much electricity do AI algorithms consume? This question is slowly but surely looming like a storm cloud over the fast-growing industry of artificial intelligence and generative AI in particular. It, along with the issue of water consumption, promises to plague the tech world going forward.

If we ask ChatGPT “How much energy do you consume”, what will it tell us? “As a language model with artificial intelligence, I have no physical presence or consume energy directly,” it will reply. Or perhaps: “the energy consumption associated with my calculations is mainly related to the servers and infrastructure used to host and manage the model”.

Google’s Bard is even bolder. “My carbon footprint is zero,” the system claims. Asked about the energy that goes into its creation and training, he says, “it’s not publicly known.”

Exactly how much energy does AI consume?

Yes, AI programs are undeniably disembodied. But they are powered by networks of servers in data centers around the world that require large amounts of energy for their computing activity to cool. And because AI programs are quite complex, they require a lot of energy – more than other types of computing.

But exactly how much energy do they “swallow”? No one can say that for sure. On the other hand, the eco-pedant community is becoming more and more persistent in its demands to reduce the carbon footprint of the IT industry.

As they race to build increasingly sophisticated AI models, companies like OpenAI (creator of ChatGPT), Google, Meta and Microsoft are not disclosing exactly how much electricity is needed to train and run their AI models, what power sources power their data centers, nor even where those data centers are located.

Danger of environmental damage

Now, as the tech industry rushes to incorporate generative AI into almost everything—from email to search engines to food delivery apps and mental health services—industry experts and researchers are warning that unchecked development of the technology could lead to significant environmental damage .

“This exponential increase in the use of AI brings with it the need for more and more energy,” said Sascha Luccioni, eco manager at AI company Hugging Face. Luccioni is among a handful of researchers trying to estimate the CO2 emissions generated by creating specific AI models.

In a yet-to-be-peer-reviewed research paper, she and her co-authors calculate the amount of energy used to train the large Hugging Face–Bloom language model; the energy used to produce the supercomputer hardware it runs on and maintain its infrastructure; the electricity used for the program’s computing activity since its launch. Scientists have found that it generates about 50 tons of carbon dioxide emissions.

Meanwhile, newer AI models are getting bigger – and more energy-hungry. They require the use of more and more powerful graphics processing units (GPUs) and take longer to train — consuming more resources and energy, Lucchioni said.

In search of cheap energy

In general, companies developing AI models tend to build data centers where energy is cheapest. As big tech firms like Google and Microsoft strive to achieve net zero emissions, they are particularly motivated to build their computing facilities in areas where solar or wind energy is abundant.

However, such areas are traditionally poor in water resources. This means limited possibilities for water cooling of data centers, which in turn affects the consumption of electricity for cooling.

When leading experts in the field of artificial intelligence recently called for regulating AI development and preventing the “existential risk” posed by artificial intelligence, it sparked a cascade of speculation about the threats superintelligence poses to society. But the researchers warned that one of the most immediate risks is actually environmental.

Is it worth the price

If companies were more transparent about the natural resources used and the carbon emissions emitted when creating and using AI models, they could help solve the problem of the carbon footprint of generative models. According to Luconi, this question should be considered through the lens of usefulness for human society.

Using AI may be worth the environmental price when algorithms are used to treat cancer, but in many cases algorithms are harnessed to work that has little social value – and then it’s not worth it the price paid in terms of consumption of natural resources.


Please enter your comment!
Please enter your name here