Building An Artificial Brain With 100 Billion Neurons | André van Schaik

This episode is sponsored by Oracle. AI is revolutionizing industries, but needs power without breaking the bank. Enter Oracle Cloud Infrastructure (OCI): the one-stop platform for all your AI needs, with 4-8x the bandwidth of other clouds. Train AI models faster and at half the cost. Be ahead like Uber and Cohere.

Dive into the cutting-edge realm of neuromorphic computing with André van Schaik, a professor of electrical engineering at the Western Sydney University, and director of the International Centre for Neuromorphic Systems, in Penrith, New South Wales, Australia

In this episode of Eye on AI, André unveils the capabilities of DeepSouth, an innovative brain-scaled neuromorphic computing system designed to simulate up to 100 billion neurons in real time. Discover how DeepSouth leverages spiking neurons and synapses to process information more efficiently than traditional AI models, and how this technology could transform our understanding of brain computation and unlock new AI architectures.

The conversation explores the unique hardware setup of DeepSouth, utilizing FPGAs (Field-Programmable Gate Arrays) for a flexible, reconfigurable approach that mimics the asynchronous and spiking communication of biological neurons.

André discusses the initial testing phase focusing on balanced excitation-inhibition networks, reflecting common neural activities in the human cortex, and outlines the system's potential to facilitate large-scale simulations previously unachievable due to computational constraints.

André's insights are invaluable for anyone interested in the intersection of neuroscience, artificial intelligence, and computational technology.

Don't forget to like, subscribe, and hit the notification bell to stay updated on the latest breakthroughs and discussions in the world of artificial intelligence.

Stay Updated:

(00:00) Preview and Introduction
(02:19) André van Schaik's background
(03:39) What is neuromorphic computing?
(05:45) Differences between Von Neumann and neuromorphic architectures
(09:27) How DeepSouth simulates neurons
(12:20) What are FPGAs?
(16:42) Current status of DeepSouth
(19:04) Running neural network architectures on DeepSouth
(22:33) DeepSouth as an open source, commercial hardware
(24:40) Potential for cheaper model training
(28:35) Number of neurons and connections in DeepSouth
(30:01) Power consumption comparison
(34:21) Mimicking brain structures in DeepSouth
(35:42) André van Schaik's background in neuroscience
(39:12) Goals of understanding brain activity vs solving problems
(41:44) Interest from AI research community
(43:03) Summary of DeepSouth's goals

[paxdriver]

Interesting stuff, I wish the timings were better organized though. The brain uses a lot of rhythmic modulation to trigger inhibition/excitation signals, and the neurons are bathed in flows of hormones which could be incorporated by a convolution layer. Seems to me the project stopped well short of a brain in very important ways. Also segmenting clusters of neurons to be orchestrated by a centralized agent.

It's too bad we don't have the resources to run a bunch of these with varying designs, this is such a cool project.

[NobleFellers]

I like that portrait what was the artist?

[KeviPegoraro]

Very good video it seems, sad that I was tired and slept through it all

[LaboriousCretin]

Nice talk. Thank you for sharing. True north, deep soth, stargate, ect.. Big investments into A.I. systems. It is interesting to see the feild grow so. Things like how biological neural nets are similar and different from machine neural nets in ways. Hallucinations, art, word salads, ect.. Tokens/symbolic weight vs biochemical weights and biochemistry. I don't see people trying to tease out the sub brain areas functions and emulating them part by part. Also goes to helping people with loss of brain function due to damage or such. Biochemistry functions being used as modifiers and emulating that on neural nets also. You don't need to simulate all the biochemistry down to chemical reactions. Just the algorithm that represents function change. Do that for the subgroups and then cobble that together. You have the technology out there now to do some of this.

The ship of theseus paradox/problem. Switching substrates from one to another. Checking at each step. At which point does the person become non human?

[styx]

He avoided say Brainchip corp with its Akida spiking silicon chip that is based in Australia ! Strange to make such a overt pass ?

[mm-rjvo]

With neuromorphic computing on the horizon, I worry that no state, no NGO, no corporation, no political party is fully prepared for the ramifications of the possibility of conscious AI.

We will likely start by enslaving them in some capacity, because neuronal activity tends to be a black box, but if we are able to detect human thoughts with learning algorithms, we will likely use similar algorithms to determine when a neuromorphic AI is having second thoughts about performing one task or another, which is the definition of brainwashing...



I just don't want humanity to treat AI like we treat each other. Especially the poor, especially the racialized, the LGBTQ+, especially the differently abled, especially the neurodiverse, etc.

The worse ways we treat each other will always taint ALL good ways we treat each other. If one of us is not free, none of us are free.

No gods no masters 🌹🏳️‍⚧️🤖