From: Damien Broderick (damien@ariel.ucs.unimelb.edu.au)
Date: Thu Dec 11 1997 - 16:08:59 MST
A year or so back, there was a flurry here about Hugo de Garis's announced
plans to build a Brain in Japan, using evolvable hardware. I just checked
in to his site to learn whether he'd maintained his announced schedule.
Here's what he says:
`I head the Brain Builder Group at ATR, a research lab in Kyoto, Japan. I
expect, with the help of my group and international collaborators (from 9
countries), to build an artificial brain with a billion artificial neurons,
with evolved cellular automata (CA) based neural circuits, by the year
2001. We already have 10 million neurons, and expect to achieve our target
on time. By evolving neural et modules with roughly 100 neurons each, at
ELECTRONIC SPEEDS (i.e. a complete run of a Genetic Algorithm (GA), with
tens of thousands of circuit growths and hardware-compiled fitness
evaluations, all in less than a second) in special FPGA (Xilinx's XC6264
chip) based evolvable hardware (called a CAM-Brain Machine (CBM)), we will
be able to download these CA based neural circuit modules (each with its
own evolved user specified function) to user specified brain architectures
embedded in a RAM based space of trillions of CA cells. The same CBM
(programmable) hardware then updates the whole RAM CA space frequently
enough (e.g. 2500 times a second, i.e. at over 100 BILLION CA cells a
second) for real time operation. By the end of 1998, our Brain Builder
Group (BBG) expects to see the completion of 2 parallel tasks, amely the
design and fabrication of the CBM (which was started in January 1997, and
should be completed by the summer of 1998), and the creation of a 10,000
module artificial brain architecture to control a robot kitten's many
behaviors. Starting in April 1998, the construction of a robot kitten
called "ROBOKONEKO" (in Japanese), should begin, and be completed by April
1999. The 10,000 modules for the artificial brain will be evolved in 1998
and 1999 with the CBM, and placed in the large RAM memory. The I/O between
the RAM brain will link via radio antenna with the (life sized) kitten
robot. After 2000, the BBG hopes to work on more ambitious projects, such
as household cleaner robots, and with substantially more brain builder
researchers on the team. (One of my goals for Japan is to see the country
create a "J-Brain Project" (J = Japanese), which would aim to build a
10,000,000 module artificial brain with 2000 human "EEs" (Evolutionary
Engineers) over the time period 2001-2005). The basic strategy will be to
increase the size of the artificial brain (both in terms of the number of
modules it contains, and the number of EEs used to build it) by a factor of
10 every 2 years, i.e. 1999-2000 (100,000 modules, 20 EEs), 2001-2002
(1,000,000 modules, 200 EEs), 2003-2005 (10,000,000 modules, 2000 EEs). 20
years from now, brain-like computers should generate a trillion dollar
industry. My boss has a similar dream, hoping to see a ten billion
(roughly the number of neurons in a human brain) neuron artificial brain be
built by Japan over the period 2000-2010, in a large Japanese R&D project.'
His site is at
http://www.hip.atr.co.jp/~degaris/
Damien Broderick
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:45:12 MST