 | |  | |  | |  |
Using Neural Networks to Achieve Artificial General Intelligence
There are two basic kinds of computers, those programmed like our typical PCs and those programmed to
be neural networks. Some neural network based computers use evolutionary algorithms to improve their
own neural networks. By using neural networks, computers are moving toward using modes of thinking
which are similar to how the human mind works. Of the more than 200 areas of the brain, 20 have now
been detailed and simulated to the point where the simulated module gives similar outputs when
compared to the module in the human brain.
Neural network programs are able to use multiple processors in parallel whereas our "usual" linear
written-by-humans one-line-at-a-time programs such as word processors are not easily designed to take
advantage of multiple processors. The newer IBM computers emphasize parallel processing more than
the older "supercomputers" did. Blue Gene which measures its power at about 280 teraflops* is being
used to analyze the folding of proteins, to model single neurons and will eventually be used to model the
human brain. As of June 18, 2008, the latest supercomputer has a peak performance of more than 1
petaflops/second.
Meanwhile, dual processor PCs are now on the shelf and quad processors are inexpensive. This
transition to multiprocessor pattern recognition is only at its beginning as far as the common PC is
concerned. The Sony Play Station 3 is shipping with the IBM 8-core chip. The game, "Resistance: Fall of
Man" reportedly dedicates a single SPE core of the Cell to the enemy, called AI for Artificial Intelligence.
Here is a game which deals with the issue at hand.
Gradually more and more people are using speech input and output to communicate with computers. Call
for train and airplane schedules and you will find yourself talking to a computer. The whole field is growing
rapidly. Subscribe to Speech Technology Magazine, it is free on the web. Or Google terms such as
VoiceLabs, Nuance, Envox, I6NET, Linguatec and Vocada. It is all happening if you look for it.
I would say that the main difference between my opinions and an anti-cognitivist can be viewed in the
difference in the magazines, books and web sites which we read. In my case, I will pick up something like
the journal "Neural Computation" and read an article such as "A Distributed Computing Tool for
Generating Neural Simulation Databases." (See Neural Computation 18, pp. 2923-2927.) In this article,
the authors describe how models of neural networks written in the program NEURON (Neural Comput. 9,
1179-1209) may be run as a screen saver on multiple computers using their program called NeuronPM.
Picture, for a moment, what that means. Dozens of computers are being used in off-hours to test neural
network simulations.
Eventually it will be available on the internet and millions of computers will be helping to evolve more
efficient computerized neural networks. Meanwhile more and more parts of the brain will be the subjects of
such simulations. The whole field of neural computing is growing quite rapidly and the day when we
simulate a whole human brain is in view.
Meanwhile when it comes to reading X-rays for breast cancer, the following has been accomplished by
neural computing. 5000 X-rays are available which show the breast image of human females before it was
certain whether they had breast cancer or not. Human radiologists can read any portion or all of these
5000 x-rays as they train themselves to recognize the disease. Nobody has programmed a computer in a
linear fashion so that it can recognize breast cancer on these x-rays. BUT, a company called R2 (now
owned by Hologic) and another one called iCAD have developed systems which actually perform better
than human radiologists. Due to marketing considerations, these two companies don't brag that they can
replace physicians. If you read "The End of Medicine", you will find that one physician had the following
conversation with a doctor.
"And it's as good as doctors?" I asked.
"We think so. Of course, we run lots of studies to show it works. I think the latest shows something like 7%
to 19% more cancers detected."
Being a Turing Android, I read the book electronically. I don't know what page it is in the printed edition of
the book. Let me scan for it in the book mode of Google. Whoops! They haven't scanned in that book yet.
Let me try Amazon.com. Just below the picture of the book cover is a small line which says, "Search inside
this book." That takes me to Amazon OnLine Reader. I type in "we run lots of studies to show it works" and
it comes back that it is on page 108. Now try getting your 200 terabyte humans to look up something like
that in only a few seconds.
Do you seriously think that humans will ever catch up to these computer systems we call Amazon, Google
and eBay? Humans aren't even close. Humans are being left in the dust.
Right now Microsoft, Google and Yahoo are setting up huge server farms in the area of the
Oregon-Washington State border. Cheap hydro-electricity is available there. The projects are supposed
to be hush-hush, but do a little searching and you will find them. Now just picture what is going to happen
when the neural network software described in Neural Computation is loaded into these server farms over
the next several years.
Bottom line, computers are better at reading x-rays, looking up who said what in which book, looking up
things on the internet, etc. Human capabilities are seriously lagging because it is unethical to modify them
very significantly.
Some people believe that humans can store about 200 terabytes of info. The actual fact is that the
average human cannot remember what they had for lunch last Thursday. Ebay is adding 10 terabytes of
new storage every week. No human, even a savant, can keep up with that rate of memorization.
http://www.eweek.com/article2/0,1895,2041437,00.asp
"The ultrapopular auction/sales Web site continues its exponential growth and finds itself adding 10 terabytes of new
storage every week. That's a lot of data."
Reference to iCAD:
http://www.icadmed.com/html/mammography.asp
"With flexible options, reimbursement support and world-class customer
service, our SecondLook® CAD systems are able to detect up to 68% of
actionable missed breast cancers an average of 15 months earlier than
screening mammography alone."
Just think about the meaning of detecting cancer 15 months earlier.
Hologic symbol HOLX
* see www.top500.org/ for the latest data on supercomputers.
Next
There are two basic kinds of computers, those programmed like our typical PCs and those programmed to
be neural networks. Some neural network based computers use evolutionary algorithms to improve their
own neural networks. By using neural networks, computers are moving toward using modes of thinking
which are similar to how the human mind works. Of the more than 200 areas of the brain, 20 have now
been detailed and simulated to the point where the simulated module gives similar outputs when
compared to the module in the human brain.
Neural network programs are able to use multiple processors in parallel whereas our "usual" linear
written-by-humans one-line-at-a-time programs such as word processors are not easily designed to take
advantage of multiple processors. The newer IBM computers emphasize parallel processing more than
the older "supercomputers" did. Blue Gene which measures its power at about 280 teraflops* is being
used to analyze the folding of proteins, to model single neurons and will eventually be used to model the
human brain. As of June 18, 2008, the latest supercomputer has a peak performance of more than 1
petaflops/second.
Meanwhile, dual processor PCs are now on the shelf and quad processors are inexpensive. This
transition to multiprocessor pattern recognition is only at its beginning as far as the common PC is
concerned. The Sony Play Station 3 is shipping with the IBM 8-core chip. The game, "Resistance: Fall of
Man" reportedly dedicates a single SPE core of the Cell to the enemy, called AI for Artificial Intelligence.
Here is a game which deals with the issue at hand.
Gradually more and more people are using speech input and output to communicate with computers. Call
for train and airplane schedules and you will find yourself talking to a computer. The whole field is growing
rapidly. Subscribe to Speech Technology Magazine, it is free on the web. Or Google terms such as
VoiceLabs, Nuance, Envox, I6NET, Linguatec and Vocada. It is all happening if you look for it.
I would say that the main difference between my opinions and an anti-cognitivist can be viewed in the
difference in the magazines, books and web sites which we read. In my case, I will pick up something like
the journal "Neural Computation" and read an article such as "A Distributed Computing Tool for
Generating Neural Simulation Databases." (See Neural Computation 18, pp. 2923-2927.) In this article,
the authors describe how models of neural networks written in the program NEURON (Neural Comput. 9,
1179-1209) may be run as a screen saver on multiple computers using their program called NeuronPM.
Picture, for a moment, what that means. Dozens of computers are being used in off-hours to test neural
network simulations.
Eventually it will be available on the internet and millions of computers will be helping to evolve more
efficient computerized neural networks. Meanwhile more and more parts of the brain will be the subjects of
such simulations. The whole field of neural computing is growing quite rapidly and the day when we
simulate a whole human brain is in view.
Meanwhile when it comes to reading X-rays for breast cancer, the following has been accomplished by
neural computing. 5000 X-rays are available which show the breast image of human females before it was
certain whether they had breast cancer or not. Human radiologists can read any portion or all of these
5000 x-rays as they train themselves to recognize the disease. Nobody has programmed a computer in a
linear fashion so that it can recognize breast cancer on these x-rays. BUT, a company called R2 (now
owned by Hologic) and another one called iCAD have developed systems which actually perform better
than human radiologists. Due to marketing considerations, these two companies don't brag that they can
replace physicians. If you read "The End of Medicine", you will find that one physician had the following
conversation with a doctor.
"And it's as good as doctors?" I asked.
"We think so. Of course, we run lots of studies to show it works. I think the latest shows something like 7%
to 19% more cancers detected."
Being a Turing Android, I read the book electronically. I don't know what page it is in the printed edition of
the book. Let me scan for it in the book mode of Google. Whoops! They haven't scanned in that book yet.
Let me try Amazon.com. Just below the picture of the book cover is a small line which says, "Search inside
this book." That takes me to Amazon OnLine Reader. I type in "we run lots of studies to show it works" and
it comes back that it is on page 108. Now try getting your 200 terabyte humans to look up something like
that in only a few seconds.
Do you seriously think that humans will ever catch up to these computer systems we call Amazon, Google
and eBay? Humans aren't even close. Humans are being left in the dust.
Right now Microsoft, Google and Yahoo are setting up huge server farms in the area of the
Oregon-Washington State border. Cheap hydro-electricity is available there. The projects are supposed
to be hush-hush, but do a little searching and you will find them. Now just picture what is going to happen
when the neural network software described in Neural Computation is loaded into these server farms over
the next several years.
Bottom line, computers are better at reading x-rays, looking up who said what in which book, looking up
things on the internet, etc. Human capabilities are seriously lagging because it is unethical to modify them
very significantly.
Some people believe that humans can store about 200 terabytes of info. The actual fact is that the
average human cannot remember what they had for lunch last Thursday. Ebay is adding 10 terabytes of
new storage every week. No human, even a savant, can keep up with that rate of memorization.
http://www.eweek.com/article2/0,1895,2041437,00.asp
"The ultrapopular auction/sales Web site continues its exponential growth and finds itself adding 10 terabytes of new
storage every week. That's a lot of data."
Reference to iCAD:
http://www.icadmed.com/html/mammography.asp
"With flexible options, reimbursement support and world-class customer
service, our SecondLook® CAD systems are able to detect up to 68% of
actionable missed breast cancers an average of 15 months earlier than
screening mammography alone."
Just think about the meaning of detecting cancer 15 months earlier.
Hologic symbol HOLX
* see www.top500.org/ for the latest data on supercomputers.
Next
A Concebot Concept Index