Researchers at Western Sydney University in Australia are gearing up to launch DeepSouth, a new supercomputer designed to emulate the human brain.
DeepSouth aims to overcome limitations faced by traditional computers, particularly their colossal energy requirements. It’s set to become operational in April 2024.
Here’s how DeepSouth works in brief terms:
- Performance: DeepSouth can perform 228 trillion synaptic operations per second, a rate that matches the estimated operational capacity of the human brain.
- Neuromorphic architecture: Unlike traditional supercomputers, DeepSouth employs a neuromorphic system that closely mimics biological processes. It uses hardware specifically designed to efficiently emulate large networks of spiking neurons. There are few details on precisely what hardware it uses, though there are a few neuromorphic chip projects worldwide. The team said DeepSouth uses “off-the-shelf configurable hardware.”
Professor André van Schaik, leading the project as the director of the International Centre for Neuromorphic Systems at the university, explains, “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.”
AI is hugely intelligent and powerful, but it requires colossal computing power. Leading AI developers like OpenAI, Microsoft, Google, Inflection, Anthropic, etc, have been stocking up on tens of thousands of high-end AI-specific hardware. This approach is immobile and extremely power-hungry.
Van Schaik also highlights the current limitations of AI models like ChatGPT, which require immense energy for training and still make errors. A recent study estimated AI training operations to be using similar energy to the entire of the Netherlands.
“We have really good AI at the moment with the large language models that are really capturing people’s imaginations and fears. But they don’t compute at all like a brain,” he states. He anticipates that understanding the brain’s efficient computing methods will lead to smarter AI systems.
When we imagine super-intelligent AI, we often imagine biological-esque robots that move around and interact with the environment wirelessly. They replicate nature’s machinery but with electronic components, walking among us.
Thus far, AI development simply doesn’t lend itself to these forms of bio-inspired technologies, but with technologies like DeepSouth and neuromorphic chips, this is changing.
Neuromorphic computing seeks to replicate the integrated processing and memory functions of neurons and synapses to address inefficiency. Recent experiments have gone further by running machine learning tasks in biological brain-like organoids.
The potential applications of neuromorphic technology are vast. It could lead to more powerful mobile devices operating independently of cloud networks, enhancing capabilities in remote search-and-rescue operations and deep space exploration.
We could use these advanced robots to build colonies on other planets, for example.
The launch of DeepSouth promises a significant stride in understanding the brain’s workings and advancing AI technology.
Similar projects have been launched in the past and met challenges, such as the Human Brain Project (HBP), but DeepSouth will undoubtedly conjure up some unique insights.