Building an artificial brain: 86B neurons, 500T synapses, and a neuromorphic chip

Рет қаралды 11,621

Күн бұрын

Is neuromorphic computing the only way we can actually achieve general artificial intelligence?
Very likely yes, according to Gordon Wilson, CEO of Rain Neuromorphics, who is trying to recreate the human brain in hardware and "give machines all of the capabilities that we recognize in ourselves."
Rain Neuromorphics has built a neuromorphic chip that is analog. In other words it does not simulate neural networks: it is a neural network in analog, not digital. It's a physical collection of neurons and synapses, as opposed to an abstraction of neurons and synapses. That means no ones and zeroes of traditional computing but voltages and currents that represent the mathematical operations you want to perform.
Right now it's 1000X more energy efficient than existing neural networks, Wilson says, because it doesn't have to spend all those computing cycles simulating the brain. The circuit is the neural network, which leads to some extraordinary gains in both speed improvement and power reduction, according to Wilson.
Links:
Rain Neuromorphics: rain.ai
Episode sponsor: SMRT1 smrt1.ca/
Support TechFirst with $SMRT coins: rally.io/creator/SMRT/
Buy $SMRT to join a community focused on tech for good: the emerging world of smart matter. Access my private Slack, get your name in my book, suggest speakers for TechFirst ... and support my work.
TechFirst transcripts: johnkoetsier.com/category/tec...
Forbes columns: www.forbes.com/sites/johnkoet...
Subscribe to the podcast: anchor.fm/techfirst
Keep in touch: / johnkoetsier

Пікірлер: 35

@teaguewilson7372 2 жыл бұрын

Plot twist: Gordon himself was a fully functional neuromorphic AI the whole time

@444haluk 2 жыл бұрын

Give us the juicy bits, every thing you said is the known arguments in the neuromorphic community. Give us async neurons, synaptogenesis methods, how to code them, Assembly?, What is the power, can they be stacked on top, is there neuroregulation methods, the juicy bits!

@Notthatguy23 2 жыл бұрын

Finally speach to text will work fluently in every language! Universal translator!

@myelinsheathxd 2 жыл бұрын

We also need better real time physics simulation processing unit to simulate them in an environment!

@eyal.herlin 2 жыл бұрын

Thanks a lot for this. Eye opening, super interesting and very clearly delivered.

@johnkoetsier 2 жыл бұрын

I couldn't agree more. Gordon is a really good communicator.

@eyal.herlin 2 жыл бұрын

@@johnkoetsier I also liked his excited but realistic and non-hyped attitude while discussing such a grand vision. I happened to watch @Veritasium's "We're Building Computers Wrong " episode yesterday which showcased a solution that required digital steps between the analog ones to avoid accumulated distortion. Gordon mentioned this kind of solution and clarified that their solution is fully analog (highlighted by him saying the signal/computation flows at wire-speed end-to-end). Impressive.

@DivineMisterAdVentures 5 ай бұрын

Been thinking about this since '74 myself. Came up with the discreet Memsistor idea before it was invented. Analog circuits are basically able to directly measure, manipulate, and store complex values with very minimal circuitry compared to digital - and commensurately faster and precisely. Rounding errors are a factor reduced only by huge amounts of digital overhead, as well. So for me, the takeaway here is how FEW views there have been of this interview over the past year. Not many people are really aware of this going on, it seems. It was only a post-openai blow-up news article that mentioned Altman's investment that clued me to RAIN - cause I am not a privileged elite. Just an inventor of copacetic ideas and models.

@That_BIack_Cat 6 ай бұрын

Someone needs to find out if Gordon Wilson has met Michael Levin from Tufts University yet. If not, they should. These two gentlemen might be able to expedite each others work. Just my opinion.

@BeyondBorders00 6 ай бұрын

I would love to hear more about brain organoids and in vitro neurons used for calculation. For example, Dishbrain from Cortical Labs in Australia. There are a number of university research groups also using neurons either in organoid form or in vitro to perform calculations and solve problems. Please cover this topic

@shawnvandever3917 Жыл бұрын

I think the answer will be digital and analog together . What these guys and others like them are doing is amazing

@monad_tcp 2 жыл бұрын

What they need is a way to make the chips faster. I have some ideas on how to make a hybrid analog compiler capable of translating high level data structures with their analog algorithms to low level CMOS mask, bypassing most of the expensive and complex eletrical engineering steps, synthesizing hardware.

@fincrazydragon 6 ай бұрын

FPGA and FPAA?

@monad_tcp 6 ай бұрын

@@fincrazydragon no FPGA or FPAA, I'm literally going to calculate the logic and then compile it to CMOS mask without using pre-defined "devices" or database of CMOS masks. I'm going to draw the CMOS mask in the compiler, its is not programmable in any form, its ASIC, but the entire process is done via compiler.

@monad_tcp 6 ай бұрын

I am going to even create my own photolitography machine to be able to print custom different "masks" without an actual physical photolith film using DLP techniques, so it can be changed on the fly. If anyone has interest ping me.

@user-gh6sn3wd7v 6 ай бұрын

got any links to the papers that inspired you to do this? sounds interesting @@monad_tcp

@v1ryus 6 ай бұрын

@@monad_tcpwhat's your lowest die size?

@PeterMorgan100 Жыл бұрын

Nice.

@seanburton5298 6 ай бұрын

How would you make the chips respond to frequency and shielded from electromagnetic pulses?

@v1ryus 6 ай бұрын

His figures are lower than the latest NPUs - at least performance wise. His efficiency factors are their value prop.

@johnlucich5026 Жыл бұрын

Photonic Neuromorphic Chips Computing is still 10-15 year away, but you can be out in5 years. . GOOD LUCK GENISES !

@seanburton5298 6 ай бұрын

Does the Local learning rule have an x,y,z coordinate to it?

@mt-qc2qh 6 ай бұрын

x,y,z,plus t in order to mimic the plasticity of the human brain.

@seanburton5298 6 ай бұрын

@@mt-qc2qh cool 😎 thanks!

@TheMajickNumber 6 ай бұрын

Um... Anyone just a little bit concerned. Am I seeing things begin to fall in place at a more and more rapid pace? Across multiple domains?? Almost as if.... Are we f*cked? Am I f*cked? Do I need a vacation?

@JosephDefendre 6 ай бұрын

We are in fact beyond fucked

@RajeshGupta-vv5sc Жыл бұрын

A lot of talk what no plan. The things discussed here are known to a lot of people but the plan is to build a network with 86 billion artificial neurons. i.e 86000000000. Whether the same will be analog or digital does not matter.