Computer simulated "brain" comming soon

We had some interesting comments about it on this thread. I think most people agree that they still don't have anywhere near the computing power that they will need. But who knows how much computing power you really need?

Anyway, it's hard to argue that the whole thing isn't worth a try. We need to have a Brain v1.0 Beta before we can ever have a Brain v6 Final.

Hopefully they are modelling the brain of this one guy I know that drinks waaaay too much. They can realistically achieve his computing power. :P

Well, I know one thing.. Even the 80386 cpus could do multiplications faster than I could ever dream of.. But yeah, there's pretty much to a brain really.. But then.. It sure would be a differense to emulate a 2months old child, or someone like Albert Einstein.. :p

Quote:

Original post by DvDmanDT Well, I know one thing.. Even the 80386 cpus could do multiplications faster than I could ever dream of..

That maybe so but the interaction with the physical world that your brain has to deal with is massively computationally expensive. I think the estimate that some people have bandied around is around 100-200 TFLOPS of processing power. To get this sort of processing power in an object the size of a brain is way beyond anything we are capable of today, I think Blue Gene/L is probably 10's of 1000's of times the size of a brain :)

I'm not sure this is possible. In fact, a brain and a computer ins't the same thing at all and trying to program something like a brain is just impossible since the computer cannot interact with the world like we do with our 5 sense.
Our brain do mutch more that just making calculation. That my opigion.

I guess the whole experiment is an attempt to "visualize" the way electrical pulses flow trough the neurons and maybe sucha visualization can give a clue about how the hell they work as a whole.

Its definitly possible to create a simulation of a brain, AFTER we understand how the brain works :D. Even if it requres a physics model of the atoms that make up the brain in order to create an accurate model, it is possible. However the computing power for somthing such as this is out of this world.

I realise it probably isnt the best comparason, but I do remember reading that each neuron is estimated to work at an equivilant of 100mhz. However I cannot relate this back to biological foundations at all. Is this simply that each neuron 'updates' it self 100,000 times per second, or that the estimated computer power to emulate each neuron is rated at 100,000 calculations per second, but then this would come down to the efficiency of the code.

But to put that into processing perspective a desktop pc has no chance of emulating more then a handfull of neurons... However it could always be run in slow motion

Im sorry I no longer have a link, nor can I even remember what I looking for at the time.

I posted an earlier version of this paper a long time ago...

Here it is again: Singularity

The key to greater-than-human intelligence is quantum computing. That means heatless processing (so that chips can be 3d and processor power grows with size) with virtually infinitely parallel processing (pure computational speed).

A greater-than-human intelligence is possible...and you'd better be ready for it.

Quote:

Original post by owl True or advertising? One way or another, it is an exciting idea me thinks... I would really like to read some comments from people with knowledge on the subject, and everyone elses oppinion.

Aww you ment human brain, so poor. Computing power for something smarter than human brain was enough aprox 6 years ago. It's not difficult to create something as stupid.
Also emulating parts of human brain is NO solution of creation an efficient inteligence. There is something like conversion factor by aprox 1000+ that renders such attempts as uneffective. There are more efficient ways, just you will get something than differ in behaviour from the norm.

Of course theirs work could be important for neurosurgery and simulation of a brain diseases. Not for AI.

And before this would be spread into a debate about speech recognition and visual recognition, lets that things to be done by a specialized hardware, thinking is something different.

Quote:

I realise it probably isnt the best comparason, but I do remember reading that each neuron is estimated to work at an equivilant of 100mhz

100 mHz = 1/10 Hz so
2 GHz = aprox 20 000 000 000 neurons. 1 P4 could do 5 brains. Luckily it needs more than one cycle per instruction. ~_^

(BTW if P4 would need 7 cycles per single instruction it would be alowed to work for a 20 cycles how many instruction would it do in 20 cycles?)

100Mhz, not 100mhz. (100,000,000hz.)

2Ghz = 2,000 neurons.

Not the best.

From,
Nice coder

Quote:

Original post by Raghar Quote: Original post by owl True or advertising? One way or another, it is an exciting idea me thinks... I would really like to read some comments from people with knowledge on the subject, and everyone elses oppinion. Aww you ment human brain, so poor. Computing power for something smarter than human brain was enough aprox 6 years ago. It's not difficult to create something as stupid. Also emulating parts of human brain is NO solution of creation an efficient inteligence. There is something like conversion factor by aprox 1000+ that renders such attempts as uneffective. There are more efficient ways, just you will get something than differ in behaviour from the norm. Of course theirs work could be important for neurosurgery and simulation of a brain diseases. Not for AI. And before this would be spread into a debate about speech recognition and visual recognition, lets that things to be done by a specialized hardware, thinking is something different. Quote: I realise it probably isnt the best comparason, but I do remember reading that each neuron is estimated to work at an equivilant of 100mhz 100 mHz = 1/10 Hz so 2 GHz = aprox 20 000 000 000 neurons. 1 P4 could do 5 brains. Luckily it needs more than one cycle per instruction. ~_^ (BTW if P4 would need 7 cycles per single instruction it would be alowed to work for a 20 cycles how many instruction would it do in 20 cycles?)

Ugh, reading this made my brain hurt! You can't compare these two very different systems this way. Go read a textbook on functional neurology! The "computing power" of the human brain doesn't come from the speed at which it operates--it's electrochemical and REALLY slow! An individual brain cell is rather insignificant. The power comes from the connections between neurons. Check into Hopfield networks and image storage for a little taste.

The brain is really, really, (no amount of overstatement possible) REALLY complicated. No one understands it yet. Anyone who says they do is trying to sell you something.

Comparing the computational power of the brain and a conventional CPU is meaningless. The brain isn't a general purpose computer. Roughly, it's a highly parallel self-modifying pattern recognizer. Not comparable. Even that description is suspect, since we barely know anything.

Consider: Where in your brain is memory stored?

In the strength of your synapses, distributed across the brain? In specific cells? In the chemical concentrations of the extracellular matrix? In fluctuating numbers of ion channels in the neuronal membranes? In self-sustaining electrical activity of recurrent neuronal networks. In the mostly unresearched syncitium of glial cells (astrocytes)?

We're making crazy wild guesses because we are totally in the dark.

So stop your damn speculating.

-Alex

You cant even make accurate comparisons of cpu in different families from arbitrary numbers(like clock speed) let alone different species lol. Where did the notion that a brain emulator would actually do anything come from? My best guess is that it would do absoluty nothing(just like you would with absolutly no outside input). Really I dont know anymore about this than anyone else though... On another note anyone see the picture of what the server will look like on that page? I wish I had a room like that...

I like what they're doing because it's important to gain understanding by observing the brain in action. There will be many other experiments in future and this one is just a small step along the bumpy road.

Quote:

Original post by Nice Coder 100Mhz, not 100mhz. (100,000,000hz.) 2Ghz = 2,000 neurons.

[headshake]
Comparing the speed of a CPU clock to the "speed" of a brain is a rather pointless exercise.
It's a bit like comparing the surface of a sailship's sails, with the surface of a plane's wings, and concluding that "obviously" the ship must go at least 20 times faster...

and secondly, a neuron is actually a rather slow way to transit information (Clinky), because it's an electro-chemical reaction that allows the signal to travel along a nerve cell.

The only time frequencies would be employed for anything would be to quantify the maximum number of action potentials per seconds emitted by a cell, which is a way to indicate the intensity of a stimulus (the higher the stimulus, the faster the nerve emits action potentials). The frequency at which the action potentials can be emitted is capped by the refractory period through which the nerve cell membrane goes after an action potential is emitted.
We are talking a few milliseconds here, 4-6 or so. Which would translate into a frequency around 160-250 Hz. Not exactly fast, is it?

But anyway, as I said, it hasn't got much to do with the "processing power" of a brain. If anything, average connectivity of the nerve cells is probably more relevant (how many cell the average cell is connected to).

Quote:

from cypherx The brain is really, really, (no amount of overstatement possible) REALLY complicated. No one understands it yet. Anyone who says they do is trying to sell you something.

We know some mechanisms rather well (the action potential mechanisms), others we are still studying and trying to really understand (the synaptic transmission mechanism). As technology allows better observation, there is hope that the synapses will reveal a great secret that has escaped us so far, but maybe we are mistaken and the secret resides in the fluids in which the brain is bathed, or the various neurotransmitters and their influence on synaptic transmission...

When we "translated" the neurons to the concept of Neural Networks, we left quite a lot out of the system. That's because ANN are a Mathematical beast, a mental exercise that allow us to understand the essence of what neurons are.
To think that you can just jump from ANNs to a human brain... [headshake]
It's akin to studying string parsing and suddenly thinking that you can create a universal translator.
Or studying how to plot pixels on a screen and thinking that this means you can now program the next Doom III.

What really poses trouble, in the end, is how Intelligence emerges from connecting nerve cells together.
It's a bit like looking at the formula for the game of Life,then try to predict the behaviour when we generate a grid on N dimensions (because a brain is NOT a 2D grid of nodes, gee, not even a 3D grid...)

It's the beauty of emergence [smile]

I agree that comparing is rather senseless. It is nice to see that, in history, there was much optimism about thinking computers. My room at the university holds a collection of old (old!) books -- one of them is Giant Brains by Berkely from 1950. It was written just after the first few supercomputers were finished. These were multiple room sized with processing power far below your average 386. The fun is that the author describes how these machines can think (it takes him exactly one paragraph to argue for this), how it will be possible in the near future to solve many problems only humans can and, in the end, he describes a scheme for speech recognition which should soon be possible. It still isn't super reliable. Just thought this might add to the fun!

Greetz,

Illco

actually, studying the brain led to the study of Neural Networks, which spawned Hopfield networks, which led someone to design a different type of memory altogether.
Our serialised hardware isn't really the best thing to emulate a human brain, and this is the sort of hardware that would make this task a bit easier, in the long run.
Or maybe it's just a gadget, what do I know [rolleyes]

Quote:

Original post by Cypherx Consider: Where in your brain is memory stored?

I read the other day an article on the newspaper about an experiment in which people was shown pictures of celebrities and the scientist monitored a group of neurons activity. The article is really funny and, if it is true, shows that science is shooting closer to where the memories are stored, or at least, which groups of neurons participates in patern recognition.

I think this is an english version of the article.