From: Eugene Leitl (Eugene.Leitl@lrz.uni-muenchen.de)
Date: Sun Jan 05 1997 - 11:40:55 MST
JAPAN'S "J-BRAIN" PROJECT (2000-2005) VISION
The so-called "J-Brain Project" (J=Japanese) is an ambition of mine to
see Japan invest in a large (2000 scientists and engineers) national
project to build/grow/evolve an artificial brain with 10 million
evolved neural net modules, over the time period 2000-2005. This in
NOT an official announcement of such a project, its only an ambition
of mine, and is something I am working towards. Its one of the main
reasons why I came to Japan in the first place.
How realistic is such a project?
You will be able to see from other sections and papers on this web
site, that our Brain Builder Group has already put about 10 million
artificial neurons together. (See the new 2D model on the Image File
2). We are confident that by 2001, we will have achieved our target of
a billion artificial neurons. The numbers are no longer the issue. The
next problem to be confronted was how to evolve each neural module
quickly. This problem will be solved by the summer of 1997, with the
completion of the CAM-Brain Machine (CBM), which uses FPGAs (Xilinx's
XC6264 chips) to evolve a neural net module in about 0.1 second (with
about 30 neurons embedded in a cellular automata space of about 2000
CA cells). As each module is evolved, the resulting CA circuitry is
downloaded into a tiny fraction of a gigabyte (or more) of RAM. The
same programmable CBM can then be reprogrammed to update the whole
huge space in a small fraction of a second. At the moment, the idea is
to imagine the RAM as a large apartment block, with empty cubes where
the evolved module's CA circuitry is downloaded. Between these hollow
cubes are software generated CA-trail communication highways
connecting modules. The high level design is human dictated, the
modules are evolved with human specified functions. Thus the approach
is a compromise of top down human design (at architecture level), and
bottom up evolution (at module level).
Imagine it takes an evolutionary engineer (an EE) (see the talk on
this web site "Million Module Neural Systems Evolution : The Next Step
in ATR's Billion Neuron Artificial Brain ("CAM-Brain") Project") one
hour to conceive and evolve a neural net module (and to consider its
function in relation to a global architecture of all the modules and
their connections). How many modules could a small team of 5 people
evolve in a year? The answer is about 10,000. Such a brain building
project could be undertaken by university groups and small start up
companies, so long as they had a CBM and a lot of RAM in their
workstation. With 30 people over 3 years, roughly 200,000 modules
could be evolved. (If sheets of modules are needed, e.g. a retinal
layer, then the CBM could be run under software control to evolve
large numbers of modules, each only marginally different from its
neighbors. This would significantly increase the efficiency of the
project). I hope my Brain Builder Group will start building a 10,000
module system in 1997. If that is a success, then it is likely that
the size of the Brain Builder Group at ATR will increase to division
size, so that about 30 people could work for 3 years, to build a
200,000 module system. We would become true "brain architects". If the
larger project is a success, then I think the time would be ripe (i.e.
around the year 2000) to launch a major national brain builder
project. Hence the idea of the "J-Brain Project".
International Competition
Brain Building I believe will become one of the main technological
activities of nations in the early 21st century. It will be the
vehicle by which nations measure their scientific and economic
prowess. Since Japan has roughly 4 times the savings rate of the US,
and invests in itself far more heavily than the US, I prefer to see
Japan become the world leader in this new field. In fact I'm rather
cynical that America will be able to even pay for such an undertaking.
The US consumes more than it produces, with Japan paying most of the
shortfall in the form of buying US government bonds and the like. The
interest on this debt is now impacting badly on America's ability to
fund blue sky research. The US brain drain to Japan is already
underway. I see it every day, living in Japanese research labs. Since
I plan 10 to 20 years ahead, I reason that Japan is the best place to
brain-build. Thats why I live in Japan.
The only way I see the US spending big money on brain building is if
the field is seen as having strategic (i.e. national security)
implications. America's DoD (Department of Defence) could spend big
money if it felt it necessary. Our Brain Builder Group now closely
collaborates with the Chinese, so if the Americans do feel that the
field could become strategic, then they cannot afford to let the
Chinese get too far ahead.
I see Japan as America's main economic rival and China as America's
new political rival, perhaps even a future enemy. I have a similar
love-hate relationship with the Chinese as I have with the Japanese. I
have a great pride in Japan's economic achievements and its likelihood
of being number one for a generation (2000-2030) before China takes
over and dominates (with its 1.2 billion people and 10+% growth
rates). China was civilized millenia before either the west or Japan.
Europe's dominance for the past 5 centuries is a mere historical
glitch from the the Chinese 5000 year cultural perspective. They know
they will dominate next century. So do I. I want to be part of the
action, even though I am disgusted at their nepotism. (The problem
with being cosmopolitan, is that its impossible to share the same
prejudices and ignorances of a local population. The greater the level
of cosmopolitanism, the greater the cynicism, I find).
So - if any US DoD or NSA types read this and get alarmed, dont
assassinate me, just ask me to collaborate with you. I will tell
anyone who seriously wants to collaborate with our Brain Builder
Group, everything that I know. We already collaborate with groups in
Japan, China, US, and Europe. I have no particular loyality to any one
country, because I have lived in 6. I guess deep down (and in the long
term), I'm a Cosmist. (I think it would be a tragedy if humanity does
not develop the next VASTLY superior species in the coming centuries.
That would be a cosmic error.) So playing with potential international
rivalries only accelerates the rise of Cosmism. I'm quite upfront
about this. I believe this new Cosmist doctrine will dominate 21st
century global politics. (See the Cosmist essays on this web page).
Dr. Hugo de Garis,
Brain Builder Group,
ATR, Kyoto.
degaris@hip.atr.co.jp
http://www.hip.atr.co.jp/~degaris
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:43:57 MST