By Angus Dalton
Scientists from Western Sydney University are preparing to boot up the world’s first-ever supercomputer that simulates the design and phenomenal power of a human brain.
The technology could cast off bottlenecks that restrain traditional computers – including their enormous energy demands – and supercharge artificial intelligence, potentially giving rise to new forms of AI that aren’t algorithms but rather physical, genuinely intelligent decision-making devices.
“This is not the biggest number-crunching supercomputer. But what’s special about this one is that it’s really geared towards simulating how our neural system and brains compute,” Professor Andre van Schaik, director of the university’s International Centre for Neuromorphic Systems, said.
The firing up of the computer, called DeepSouth, in April 2024 near Penrith, will represent a major global milestone in neuromorphic computing, which refers to technology based on the architecture of the brain. It can be thought of as biology’s answer to quantum computing.
Our brains can perform an exaflop (a billion squared) calculations in a second with only 20 watts – less power than a lightbulb. That’s an estimated million times more efficient than the computers we build, which are kneecapped because data must be shuttled between devices’ memory and processors, slamming the brakes on power and speed.
Neuromorphic computing skirts this problem by replicating the ability of neurons and synapses to bring processing and memory together.
Van Schaik said it took millions of dollars worth of energy to train AI models such as ChatGPT, which still make “stupid mistakes”.
“We have really good AI at the moment with the large language models that’re really capturing people’s imaginations and fears. But they don’t compute at all like a brain,” van Schaik said. “They’re more number crunching machines and they are very energy hungry. Once we understand how brains do what they do so efficiently, that will lead to different forms of AI and smarter systems.”
The rabid power demands of data centres that underpin cloud computing are another fast approaching blockade to better computing, said Zdenka Kuncic, Professor of Physics at the University of Sydney, who’s creating her own neuromorphic device based on synthetic synapses.
A fifth of Ireland’s power, for example, is devoured by data centres. “Ultimately, that’s unsustainable,” Kuncic said. “How many more data centres can we seriously continue to build going into the future?”
By slashing energy needs, neuromorphic tech could create more powerful mobile devices that operate independently of the cloud in remote areas, spurring better search-and-rescue robots and smarter self-navigating rovers in the realms of deep space.
ICNS have already deployed neuromorphic sensors to the International Space Station, which use their brain-like capability to capture images of rare “sprite” lightning events, red eruptions that strike upwards towards space rather than down towards earth.
Like brains, neuromorphic computers can immediately process streams of incoming sensory information and filter out redundant data, acting more like an eye and optic nerve as opposed to a video camera.
DeepSouth will initially help scientists study the brain’s core processing and neurological disease without having to deal with all the “messy biology”, Kuncic said.
“It’s definitely worth shouting about,” Kuncic, who was an investigator on the Australian Research Council grant that partially funded DeepSouth but isn’t involved in the supercomputer’s construction.
“It’s really going to put Australia not just on the map, but at the front in terms of leading the way forward with this research.”
The Examine newsletter explains and analyses science with a rigorous focus on the evidence. Sign up to get it each week.