How the human brain is inspiring energy-efficient AI

Large language models are expected to increase global energy consumption, but thanks to University of Sydney researchers, there may now be a way to create energy efficient computing that works a bit like the most complex computer of all - the human brain.

While industries are making inroads in driving down emissions and energy use, advanced large language models like ChatGPT could require as much electricity as up to 17,000 households. Future generations under development could consume even more.

According to the US Office of Energy Efficiency and Renewable Energy, data centres account for 2 percent of the United States' total energy use. In Australia, reports suggest data centres they account for 1 percent of total energy use, potentially reaching 8 percent by 2030. Large language models, like Open AI's, require large amounts of computational power to sift through vast troves of data.

How the human brain could inspire sustainable computing

Associate Professor Chang Xu in the University's Net Zero Institute is working to improve the efficiency of algorithms so that the hardware they run on requires less energy to work.

"We're meant to be scaling back our energy use, but the advent of large language models has been a shot in the arm and we're seeing energy usage of computing soar," said Associate Professor Xu. "This is totally the wrong direction."

"Most of the time when people use large language models like ChatGPT, they are making small queries or asking for help on pretty simple tasks. Yet these models still fire on all cylinders to develop a response, using increasing amounts of energy," he said.

Associate Professor Xu says we need only need to think about the human brain to understand how his technique works.

"When you think about a healthy human brain - it doesn't fire all neurons or use all of its brain power at once. It operates with incredibly energy efficiency, just 20 Watts of power despite having around 100 billion neurons, which it selectively uses from different hemispheres of the brain to perform different tasks or thinking.

"In contrast, advanced AI programs like ChatGPT, which contains 175 billion parameters, requires a staggering 9 megawatts, equivalent to a medium-sized power station. This reminds us of the need to push the limits of machine intelligence, focusing not only on its accuracy but also on its efficiency.

"We are developing algorithms that do just that, that bypass the redundant computations they don't need, so they don't automatically go into high gear, meaning far less energy is required."

He hopes his technique will change how hardware is developed, to support more energy efficient applications.

Director of the Net Zero Institute, Professor Deanna D'Alessandro said climate change is a pernicious, whole-of-society issue that is deeply embedded in every facet of society and must be tackled on every front.

"When people think of climate change and emissions sources, they don't generally think of AI or computing," said Professor D'Alessandro. "While AI is helping many researchers understand climate change and develop solutions which weren't previously possible, we need to make sure new technologies aren't creating an even bigger problem by becoming a significant emissions source."

About the Net Zero Institute

The Net Zero Institute is one of the University's flagship centres that is accelerating solution-based research and assist the world in meeting its climate change goal of net zero carbon emissions by 2050.

It brings together more than 150 researchers from across the University to develop solutions across various disciplines, from extracting critical minerals from waste and greenhouse gas removals to net zero health and green computing.