Grzegorz Kaczmarczyk
A strategies to build human-level-intelligence atrificial brain.
Version 0.03 (under developmend, is going to be continued); Gdańsk 2004

This article presents an approach to build human-level-intelligence atrificial brain based on resulst of recent scientific research. It discusses the computing power of the hardware needed to be used and needed memory capcity. The software that is necessary to make the solution usefull for mentioned goal is latter described. At the end the problem of obtaining conscience in computers according to latest scientific ideas is discussed.

1. The problem of computing power.
It is often estimated (Kurzwieil 2001) than
- computing power of Human Brain = 100 Billion (10^11) neurons * 1000 (10^3) Connections/Neuron * 200 (2 * 10^2) Calculations Per Second Per Connection = 2 * 10^16 Calculations Per Second
- memory capacity of Human Brain = 100 Billion (10^11) neurons * 1000 (10^3) Connections/Neuron * 10 bytes (information about connection strength and adress of output neuron, type of symapse) = 10^15 bytes = 1 PB = 1000 TB
Also, as Hans Moravec (1998) and others have speculated, very efficient simulations require about 1000 times less computation than the theoretical potential of the biological neurons being simulated.
The computing power per constant price is doubling every 14 to 15 months. The computing power of supercomputers is doubling every 14 to 15 months (results form http://www.top500.org/ are presented on attached SVG picture, the picture is described at the end of this article). However there are evidence that the growth of computing pover is described by double exponential curve (Kurzweil 2001).
Having the brain available on set of hard drives using todays supercomputers (ie. number 100 on the top 500 list) we would need one day to simulate one second of work of our brain. Such simulation would be interesting not only for neuroengineers but also for neurophysliogists. We could study milisecond by milisecond what is going on in the human brain. As the computing power and memory devices will get cheaper symulating human brain may became the main source of information about how human brain works (because usually it is not allowed to directly experiment on human brain with good time resolution).

2. The problem of software and hardware soulutons.
The software of the brain could be an application to perform simulation of the scanned human brain. We are not very far form the possibility of scanning human brain with resolution high enough to have information about neurotransmiter concentration.
The other approach is reverse engineering of the human brain. Doing this may take more time, but this may require less computing power. Kurzweil writes:
"The pace of brain reverse engineering is only slightly behind the availability of the brain scanning and neuron structure information. A contemporary example is a comprehensive model of a significant portion of the human auditory processing system that Lloyd Watts (www.lloydwatts.com) has developed from both neurobiology studies of specific neuron types and brain interneuronal connection information. Watts' model includes five parallel paths and includes the actual intermediate representations of auditory information at each stage of neural processing. Watts has implemented his model as real-time software, which can locate and identify sounds with many of the same properties as human hearing."
Instead of using supercomputers to simulate human brain we could use neuromorphic chips to perform identical calculations. The latest advances in using neuromorphic chips (Liu, S.-C. 2003) are very promissing.

3. The problem of consciousnes.
The probelm of consciusnes is widely discussed by Crick and Koch in ther recent papers. They have proposed a framework (2003) that cosists of testable hypothesies. This framework could be applied to real neurons, neuromorphic chips or even neurons simulated inside computer (or cluster of computers). More than ten years of their work shows that we are grting closer to solve the mistery of consciousnes in scientific way.

4. The problem of scanning of human brain.
Human brain scanning has already started. A condemned killer allowed his brain and body to be scanned and you can access all 10 billion bytes of him on the Internet http://www.nlm.nih.gov/research/visible/visible_human.html.
Carnegie Mellon University's Andreas Nowatzyk plans to scan the nervous system of the brain and body of a mouse with a resolution of less than 200 nanometers, which is getting very close to the resolution needed for reverse engineering.
Present chips are produced with resolution of 90 nm. Using AFM (atomic force microscope) we could obtain even a higher resolution.
Scanning of the brain may be used to migrate the mind from the brain to other medium like supercomputer, a cluster of computers, single computer or some kind of nanotechnology based medium.

5. Application to space program
The space program is the most expensive scientific program ever done, building of the space station Alpha costed 60 bilion dollars. The mission to send a man to Mars or build pernament Moon ststion may cost even more. Sending of man to space is risky and involves a lot of preparation, we could obtain the same results faster and cheaper by sending a human-level-intelligent robots to space instead of humans. Such robots could prepare conditions siutable for human habitation on Mars or on Moon. Investing in them we could make space exploration and colonization easier, cheaper and more effective.

References:
Crick, F.C & Koch, C. Consciousness and neuroscience. Cereb. Corex 8, 97-107 (1998)
Crick, F.C & Koch, C. A framework of consciousness. Naure naurocsience 6 2, 119-126 (2003)
Kurzweil, R. The Law of Accelerating Returns. http://www.kurzweilai.net/articles/art0134.html?m=1
Liu, S.-C. A wide-field direction-selective aVLSI spiking neuron , IEEE International Symposium on Circuits and Systems, May, 2003
Liu, S.-C. A silicon retina with controllable winner-take-all properties , IEEE International Symposium on Circuits and Systems, May, 2003
Liu, S.-C. Analog VLSI circuits for short-term dynamic synapses , European Journal on Applied Signal Processing, 7: 1--9, 2003
Lloyd Watts (www.lloydwatts.com)
Watts, L. The Mode-Coupling Liouville-Green Approximation for a two-dimensional Cochlear Model", Journal of the Acoustical Society of America, vol. 108, no. 5, pp. 2266-2271, Nov., 2000.
Moravec, H. When will computer hardware match the human brain?, Journal of Transhumanism. 1998. Vol. 1
Moravec, H. Robots: Re-evolving Minds at 10^7 Times Nature's Speed, in Cerebrum, v3n2, Spring 2001, pp. 34-49
Top 500 Supercomputer List (http://www.top500.org/)
The Visible Human Project.http://www.nlm.nih.gov/research/visible/visible_human.html


Figure 1. Exponential growth or double exponential growth of computing power? On X axis are years, on Y axis are 10-based logaritm of computing power (FLOPS for supercomputers and IPS for typical computers, 1 MIPS on the chart is 6, 1000 MIPS is 9) Blue circle are data probided by Moravec (1998) supplemented by my two points, the blue curve is double exponential curve specified by Kurzweil. The red circles are data from TOP500 list for supercomputers, the higest circles are data for sum of computing power for all top 500 supercomputers. The lower red circles represent computing power of computers number 1, 2, 5, 10, 20, 50, 100, 200, 500. The brown lines represent single exponential approximation to the data obtainged from top500 list. (For more details about the pucture see the description below the picture)



Letter to Ray Kurzweil.
According to the data from TOP500 list the growth of computing power available in spuercomputers can be described by single exponential growth. The correlation coefficient for sum of TOP500 supercomputers power (FOPS) is 0.99926. These data are not in conflict with the double exponential growth suggested by you in article on page http://www.kurzweilai.net/articles/art0134.html?m=1 (The Law of Accelerating Returns) for computing power per constant price. The rate of growth for TOP500 list (sum of computing power of 500 computers) is in perfect agreement with the rate of growth predicted by you for computing power per constant price.

On the above picture the X axis contains year and Y axis contain 10-based logaritm of FOPS for supercomputers and OPS for usual computers (for 1 MIPS it is 6, for 1000 MIPS it is 9, human brain 2*10^16 OPS it is 16.3 on the graph). The blue curve is the double exponential growth for computing power per 1000$, described in your article "The Law of Accelerating Returns". The blue circles are data made available by Hans Moravec on page http://www.transhumanist.com/volume1/moravec.htm under link http://www.transhumanist.com/volume1/appendix.htm with two additional points represented by computers added by me.

The red curve is the double exponentiol growth curve moved up for sum of computing power of all TOP500 supercomputers. For years 1993-2007 it is overlaped by single exponent approximation for the TOP 500 data. The red circles are computing power of supercomputers on places 1, 2, 5, 10, 20, 50, 100, 200, 500 and sum of computing power all 500 supercomputers. The brown lines represent the single exponential approximation for the data. The yellow part of the of the graph contains different estimations of the computing power needed to imitate human brain.

There are different numerical data according to the approximation:
R - correlation coefficient
A - level of growth
DM - number of months neede for doubling of computer power (at given level)
AY - year according to double exponential growth when the data when obtained level of growth is expected
YB - year when the computing power will ab at level of human brain (2*10^16 OPS)

The data are calculated for
0 - sum of computing power of TOP500 supercomputers
1 - the first supercomputer
2 - the second supercomputer
3 - the 5th
4 - 10th
5 - 20th
6 - 50th
7 - number 100
8 - number 200
9 - mubner 500
R0 is correlation coefficient for sum of computing power of TOP500 supercomputers

The 1000$ computers are about 5 years behing the 500th supercomputer.