20 billion transistors on a SiGe chip, 100 billion neurons in the brain = "Ava" AI?

Status
Not open for further replies.
20 billion transistors on a SiGe chip, 100 billion neurons in the brain = "Ava" AI?

I see in the news today that IBM has announced a working Silicon-Germanium research chip with 20 billion transistors. It is apparently (still) planar, they are not even using stacking (yet):

http://www.nytimes.com/2015/07/09/t...hips-more-powerful-than-any-in-existence.html

The human brain has arouund 100 billion neurons:

Neurons & Synapses - Memory & the Brain - The Human Memory

So the speculative question is... what are the odds that five of the new high density chips ever be able to be configured to make an "Ava" style AI in a physical space similar to the brain? 😀

http://www.rottentomatoes.com/m/ex_machina/

I'm equating one transistor with one neuron here, which may be a bad comparison. Some schools of thought tend to equate on transistor with each connection between neurons, rather than each neuron, which would add another order of magnitude or two to brain density. Then there is the speculation that the brain may be a quantum computer at some microscopic level...
 
Last edited:
I'm not a neuroscientist, so a grain or two of salt for what I'm about to say is entirely appropriate. My understanding is the individual neurons in our brains employ multiple connections with it's neighbors so the brain isn't like digital computers where simple switches are only on or off. My guess is there would need to be at least an order of magnitude more switching elements than the human brain utilizes, but again, it's just a wild guess from ignorance.

Mike
 
I'm equating one transistor with one neuron here, which may be a bad comparison.
Neural networks use at least one processor/simulated neuron, and more generally a cluster of processors/neuron (except for those working purely in soft, but the speed scale is of a different order).
Some very simple animals like worms and some arthropods only have two or three neurons, yet they manage better faced with a complex environment than our best hi-tech alternatives.
I am sure we will some day be able to catch up, but there is still >some< work to be done....
 
Hi,

Short answer is very much no, not a chance.

The complexity of a neuron is way, way beyond a transistor,
and the interconnectivity mind boggling higher, with up to
~ 1000 inputs and up to ~ 10,000 outputs per neuron.
All fed with a parallel analogue bus that is the blood supply.

FWIW a human brain divorced from all the feedback and
sensory mechanisms supplied by the nervous system
of a human body would simply "crash", i.e. die.

Current AI is miles off any sort of apparent sentient intelligence.

rgds, sreten.

My favourite take on the subject is how would you build a cats brain ?
 
Last edited:
Not even close. Our "software" is electrochemical in nature with a thousand
states between "on/off".
Times the 10,000 different interconnects and the billions of neurons and
you have the equivalent of over 10 trillion transistors.

Then it would be up to us electrochemically driven monkey's to develop
software sophisticated enough to emulate within those 10 trillion.

A better bet would be a quantum state processor to give birth to
the first AI.

OS
 
I'm with Streten on this, numbers of transistors has absolutely nothing to do with numbers of neurons.....absolutely nothing. They cannot be equated in any way. I feel that the whole notion of artificial intelligence in computational devices being compared to the consciousness of human and non human organisms is pointless. They are worlds apart!
I am sure however that at some point computers/robots will be able to display characteristics that susceptible people will regard as absolute proof of their artificial intelligence and consciousness.
They will be wrong!
The brain inevitably 'has' to work at a quantum level as discussed by Roger Penrose in his book.....the title of which escapes me at this precise moment!
 
The brain inevitably 'has' to work at a quantum level as discussed by Roger
Penrose in his book.....the title of which escapes me at this precise moment!

Hi,

At the very lackadaisical speed neurons work RP is
well out on a limb and very probably totally wrong.

All life is quantum, by default, but is brain function ?

rgds, sreten.
 
Last edited:
Agreed...he may well be wrong but he is correct in 'discussing' the possibly inevitable influences that quantum effects will be shown to have on the source of the complexity of the human (in particular) mind. Who knows?
It is with these sorts of discussions that we must remember the limitations of our current knowledge and put it in some sort of perspective.
For instance:
What do we think about what was known by us humans one thousand years ago? And therefore what will humans think about what we know now in one thousand years in the future?
 
Whist were are on the subject; has anyone seen 'Ex Machina' ?
Great new film concerning AI,( especially if you like a bit of full frontal nudity and robots.)
It's a subtle and intelligent film that's a slow burner with good character development and a neat plot. Stylish to!
 
The difference is that a computer uses a massively fast, digital, serial cpu while the brain is a surprisingly slow, analogue and massively parallel cpu.
Any comparison between numbers of transistors and numbers of neurons is meaningless.

That's not accounting for the self-correcting, continuously adaptable software the brain uses.
 
Status
Not open for further replies.