While it might be challenging to calculate the precise energy cost of a single AI model, the carbon footprint of these models is significantly expanding.
Generative artificial intelligence (AI) is the capacity of an algorithm to synthesize complicated data, such as a sentence, a paragraph, an image, or even a brief movie.
For years, generative AI has been utilized in products like smart speakers to provide audio answers or in autocomplete to propose a search phrase. But it has only lately acquired the capacity to produce human-like language and accurate images.
But training these models is not inexpensive.
A study by the MIT Technology Review found that training just one AI model may result in emissions of more than 626,00 pounds of carbon dioxide equivalent, or about five times the lifetime emissions of a typical American automobile.
What does a model’s generative AI’s carbon footprint mean?
Larger AI models need a lot more energy, which is not surprising.
For instance, according to Kate Saenko, an AI research scientist at FAIR Labs, building the generative AI model BERT in 2019 required the energy of one person’s round-trip intercontinental journey. BERT has 110 million parameters.
With 175 billion parameters, the far bigger GPT-3 required 1,287 megawatt hours of power and 552 tons of carbon dioxide equivalent simply to make the model ready for flight.
While smaller than GPT-3 in terms of scale, the BLOOM model was created by the BigScience project in France.
According to a Google research, utilizing a more energy-efficient model architecture, processor, and data center can cut carbon emissions by 100 to 1,000 times.
More Carbon Footprint Is Created by Generative AI Queries
A single generative AI query is thought to have a carbon footprint that is four to five times greater than a search engine query.
The volume of inquiries that search engines get each day may increase exponentially as chatbots, picture generators, and AI language models gain popularity and are included into search engines.
ChatGPT, which received over 1.5 billion visits in March 2023, is one example of a well-known chatbot created using generative AI.
In addition, ChatGPT is now publicly accessible thanks to Microsoft’s integration into its Bing search engine.
An advantage of utilizing chatbots instead of search engines is that they can be more direct ways to access information, even if employing AI assistants for search and other purposes could result in significant energy expenses.
More research is required to increase the effectiveness of generative AI.
Saenko stated that « the good news is that AI can run on renewable energy, » adding:
“By bringing the computation to where green energy is more abundant, or scheduling computation for times of day when renewable energy is more available, emissions can be reduced by a factor of 30 to 40, compared to using a grid dominated by fossil fuels.”
