US Scientists have developed a trashcan-shaped robot
that has 20,000 brain cells. The robot named "Darwin VII" has a
mobile base and can see with a CCD camera, hear with microphones, taste
via conductivity sensors, can turn it's head and even grab things. It
works by tasting everything and discovers what tastes good or bad. For
example, the robot finds that tasting blocks with one type of pattern
tastes good and blocks with
another pattern tastes bad. The robot learns to stop
tasting the blocks that have the pattern that indicates bad taste and
continue tasting the blocks that have a pattern associated with good
taste. Researchers at the Neurosciences
Institute (NSI) in La Jolla, California say the robot operates on
biological principles and without any pre-specified instructions.
I'm unable to find this article but will continue to look.
The brief states that it selects "what tastes good or bad" without
"pre-specified instructions". There would have to be initial tweaking
of the neural net to facilitate this and/or some feedback loop where the
determination of "good or bad" could be made. Biological systems come
with plenty of neural presets.
Looks like these articles all picked up the story from New Scientist.
Their article is a pay-per-view, so I don't know where it came from. If
there was a new paper released, I haven't seen it yet.
Gerald M. Edelman is a well known theorist on the subject of mind and
consciousness. He disagrees with Daniel Dennett on the subject of
qualia but otherwise both are materialists (e.g. they believe mind and
consciousness arise from real things found in your head as opposed to
souls, spirits, quantum weirdness, etc). He's published several books
and lots of papers
on his theories.
Somebody at robots.net (motters?) was working on his own implementation
of Edelman's theories a while back. Not sure if anything interesting
came of it.
Jeffrey L. Krichmar
You can probably get a rough idea of their current project from some of
their papers on the older
Details about DARWIN can be gleaned from this 2004
paper on DARWIN VIII (PDF format). I don't see any mention of
specifically "20,000 brain cells". They do mention that DARWIN VIII has
a cortical area of 53,450 simulated neuronal units with 1.7 million synpatic
connections. Computation was carried out on a Beowulf cluster
of 12 1.2GHz Pentium IV linux boxes, which allow complete recalculation
of connection states every 100ms.