AI usage puts huge pressure on energy grids says report
)
A recent surge in the popularity of commercial AI products based on generative, multi-purpose AI systems has seen large language models (LLMs) sit at the heart of many generative AI systems, trained on stores of written information which helps these models to generate text in response to queries.
The report says that this content is generated from scratch, placing a large workload among computers which, according to the report, might use around 33 times more energy than machines running task-specific software.
This pressure is felt most on data centres, causing a large build-up of heat and energy usage. In 2022, data centres around the world used up 460 terawatt hours of electricity, a rate that the International Energy Agency expects to double in four years’ time.
Data centres could be using a total of 1,000 terawatts hours annually by 2026, a demand roughly equivalent to the electricity consumption of Japan and its population of 125 million people.
The researchers found that multi-purpose, generative architectures are massively more expensive than task-specific AI systems for a variety of tasks, even when controlling for the number of model parameters.
Speaking in the paper, the researchers said: “We recognise that our work can be perceived as a critique of ML deployment in general, given the analysis that we provide of its environmental impacts. This could be used as an argument to stop pursuing ML research and development, or as a way of targeting specific companies or organisations.
“Our intention, however, is to shed additional light on the environmental impacts of ML, in order to help model developers and researchers make more informed choices as a function of their environmental footprint or energy usage.”
Source: Inavate