Computer Hope
Other => Computer News => Topic started by: Computer Hope Admin on March 31, 2009, 11:32:30 AM
-
An international team of scientists in Europe has created a silicon chip designed to function like a human brain. With 200,000 neurons linked up by 50 million synaptic connections, the chip is able to mimic the brain's ability to learn more closely than any other machine.
Although the chip has a fraction of the number of neurons or connections found in a brain, its design allows it to be scaled up, says Karlheinz Meier, a physicist at Heidelberg University, in Germany, who has coordinated the Fast Analog Computing with Emergent Transient States project, or FACETS.
Link (http://www.technologyreview.com/computing/22339/?a=f)
-
Dang....That's mind boggling
But the hard part is, the AI for it.
They may be able to recreate the human brain as an electronic microprocessor but the AI has to control it.
Like saying, you may have the brains, but do you have the knowledge.
-
You're totally correct. But I'd imagine the hardware part would be a lot more difficult that the software.
-
Dang....That's mind boggling
But the hard part is, the AI for it.
They may be able to recreate the human brain as an electronic microprocessor but the AI has to control it.
Like saying, you may have the brains, but do you have the knowledge.
You have it backwards. the idea of a neural network is to allow for AI without requiring explicit programming that would be required in software- it allows for fuzzy logic decisions to be made- for example, people make fuzzy logic decisions all the time,
if we ask a person, "is this colour red" they don't internally compare the colour to equal red exactly, they apply learned fuzzy logic that essentially boils down to, "is this colour closer to red then any other colour I know". with sufficient learning, (which I might add is how a neural network gets "knowledge", and learns patterns much like the human brain does).
Some people think neural networks are a waste of time- however, they can provide far more intuitive work with computers- take for example OCR, which as it stands now is fairly strong. by programming a neural network one could acquire virtually limitless and flawless. recognition of any character to a standard unicode character.
"Training", much as with a human brain, is a necessary part of creating a neural network with the proper "connections" and, as you say, "knowledge". Training is similar to that used with people- by asking a neural network, for example "is this face male or female", it will basically need to provide a complete guess the first time. However- no matter wether the guess was right or wrong, it can then save that "knowledge" about that particular face so that it may provide more accurate "guesses" in future queries. By training the network in this way with a variety of different faces, it can achieve over a 99% accuracy on something as complex as recognizing a human face.
Some might say, 99% accuracy isn't 100%. And that, of course, is true. But as much as we hate to admit it sometimes- even we have trouble distinguishing genders- which results in some uncomfortable situations, to say the least ;D
-
True but still like the article said, the chip is serial not parallel like the brain.
This could still be a limit on thinking capabilities, take Trivalent Logic:
Jan Łukasiewicz showed that the thinking of AI cannot be limited to just true and
false as shown in human thinking, that there is a third value or more that could extend
the thought capabilities to create that "fuzzy logic" that you said BC.
The chip is still just limited to TRUE/FALSE, On/Off thinking which limits it to not be able to have that "fuzzy logic" capability.
Analog computing is probably the way, to have more values than just On and Off but to have more than two values with different voltage levels or amount of current passing through.
Even in the TRUE/FALSE computing limit, AI can be developed through sub-dividing the values beyond TRUE/FALSE with say numbers, words, characters.
-
;D Glad that you had posted this, will be useful for my paper.
-
Next thing you know, people are being harvested for energy in massive fields while they unknowinly are in a virtual universe.
-
Next thing you know, people are being harvested for energy in massive fields while they unknowinly are in a virtual universe.
I hear that happens every fourth year in november in the U.S.
-
Next thing you know, people are being harvested for energy in massive fields while they unknowinly are in a virtual universe.
NetNavi's?
-
It's about that time then.
-
The chip is still just limited to TRUE/FALSE, On/Off thinking which limits it to not be able to have that "fuzzy logic" capability.
With the proper algorithm, anything is possible.
A Neuron is, after all, simply a concrete object that reacts to electricity coming through it's axons by sending electricity through it's Axons. Electricity either flows through a neuron, or doesn't. there is no "fuzzy" middle ground with the signal. a Nueral network is simply a giant conglomeration of these neurons-. stimulate one here or there and whatnot for the input, and then read the outputs of particular axons of neurons on the opposite site to get the output.
The trick is of course how to pass in data such as a human face and how to parse the output.
This inevitably makes one wonder exactly how the human brain works in such a mysterious way- images can essentially be recalled on demand, or even invented- it's even able to create Virtual worlds (dreams) that are manipulated by thoughts and feelings. neurons are constantly firing- new axons are generated between neurons to form the process of learning.
I would imagine that something being learned such as "hand eye coordination" (obviously learned through various activities), would involve adding connections between the cerebellum and the Visual cortex (ack, I think that was it's name is...)
-
pretty much have specialized chips for different functions.
-
isnt that no real differnt then a cpu almost?
-
Well if you wanted to sub divide a chip into specialized functions(Speak, Voice Reg, Core, I/O) you would have to use different CPU's or just a Quad-Core Intel ;)
-
I suppose??? But dont most do four functions?
-
I know...I'm saying if you used a Quad-core for a real AI then you could split up each function to the different cores, like the first core gets speech, the second gets speech reg, and so on and so forth..
-
I didnt mean quad I ment like a single cpu. Atucally it does two functions I think so in sense that the quad would preforme better any way.
-
Oh...
-
like it has a math funtion and the other funtiction..... ::) I dont recall too much..... but I am sure of this more or less.
-
to many functions that we cannot think of to add-in...like a Drawing, Blogging, Video Watching, Fedding Dog, etc...
My head hurts... :P
-
That wouldnt be a feature of the chip unless you wanted to be like solid state.
-
Exactly- that would be learned bahaviour in the neural network.
a conventional processor can't "Learn", for example, if that were the case it would learn quickly not to deference a null pointer, since it will crash if it tries to do so.
Also it would probably refuse to execute the HALT instruction.
Additionally diagnosing CPU problems will be more difficult, since each processor will have had different learned behaviours, which essentially invalidate previous experience with other processors.
-
NetNavi's?
The Matrix
-
have you been?
-
The Matrix
Nope, Megaman :D
-
admin save you know.
-
admin save you know.
What does this mean ? ?
-
G-Man - Admin (Half Life)
Or is he talking about Nathan?
-
That wouldnt be a feature of the chip unless you wanted to be like solid state.
I'm still working on this one so i can't say at the moment...
-
G-Man - Admin (Half Life)
Except the admin is Wallace Breen....
-
Ahh..So you know ;D
-
yeah I ment him. he probably has been.