There is a long history of attempts at
replicating neural systems either in software or in conventional
semiconductors, such as the FACETS project (not to
mention the creation of conventional logic gates from lab-grown
biological neurons!) According to a USC
Viterbi news release, researchers at the BioRC project, whose goal
is research on
an artificial cortex, have succeeded in creating a functioning synapse
from carbon nanotubes.
The new research was presented by Alice C.
Parker in the paper "A
biomimetic fabricated carbon nanotube synapse for prosthetic
applications" at the Life Science Systems and Applications Workshop
in April 2011. (unfortunately the actual paper is behind a paywall but
the abstract is readable). An earlier paper, "A Biomimetic Carbon
Nanotube Synapse Circuit", describes the proposed
design of synapse including schematics and comparison with biological
project has been working on this project for a while. To learn more
about their goals and history, see "Biomimetic
Cortical Nanocircuits: The BioRC Project (PDF format)"
Ok, that's the end of the actual news. :) Because it's been a while
had any really interesting articles on
consciousness, I couldn't help letting my mind wander as I wrote
this. As a result, I'm throwing in some personal ramblings and
observations on this one.
A common objection raised to most previous software and hardware
attempts at using Biomimetic neural technologies to build intelligent
robots is the "not made of meat" argument. The reasoning, as I
understand it, usually boils down to one of two complaints: 1) there is
something magical/religious/spooky about meat that grants machines built
from it conscious or 2) consciousness is a side-effect of the
particular structure that evolved to implement meat-based intelligent.
Argument 1 is more common in my experience but argument 2 is the more
serious complaint. The reasoning is that consciousness comes from the
structure, rather than the function, of the machine. So it doesn't
matter how well an artificial neuron executes its function, at least
with regard to
consciousness. If any part of the structure is missing (or "simulated"
vs "real") then the entire system will not attain consciousness. Your
robot will end up as smart and empty-headed as the philosopher's zombie.
A cognitive system implemented using silicon neural hardware is
unlikely to be conscious by this reasoning, because at least some of the
neural components are (considered by someone) to be mere "simulations"
of their biological counter parts. And, of
course, a cognitive system built using neural network software is right out
(these folks usually argue that everything done on a computer,
even basic math such as adding 2 plus 2, is merely a simulation and not
real in the sense they wish it to be). The only thing that will satisfy
them is a cognitive system based on hardware that has no component they
deem as "simulated".
I think most of this line of reasoning is nonsense based on
semantics. But, on the other hand, none of us really know what we're
talking about until someone defines
intelligence and consciousness in a meaningful way. So the significance
of this debate should not be underestimated. After all, there have been
theories of consciousness
built around single neurons.
What's different here is that BioRC's approach is to build a fully
functional neural structure in hardware using whatever materials are
necessary. Whether the "meat" crowd will consider the BioRC hardware
"real" enough to meet their requirements or whether they'll still find
some portion of it to point to and say, "ah-ha, just a simulation!" remains
to be seen. But it will be entertaining if not instructive to watch the
further progress of the BioRC project and any possible practical
applications of the hardware being developed.
I should also point out that most of this debate is going on among us
spectators and is not really the main thrust of BioRC's research. Their
reason for choosing carbon nanotubes over, say conventional CMOS
circuits, is that the size and density of connections required to build
a human-level cortex in CMOS would require decades to build and be
prohibitively large. They hope carbon nanotubes offer a faster, smaller
way to achieve that goal.