FleshFactor FleshFactor


FleshFactor: Wet, dry, mechanical, electronic, opto-electronic, digital, analogue and others!

A E C  F O R U M - "F L E S H F A C T O R"

Well, I tend to agree with Derek Robinson when he says we won't get
anywhere by trying to replicate a brain in detail. That is why I was
stressing the "in theory" in my previous articles, and why I made a
distinction between theory and practice. But, as pointed out in the
articles about chaos, it seems to me that even if we could replicate every
little detail in practice we still wouldn't end up with the same brain
because in the five minutes between finishing the replica and observing it
together with the original, the two will have diverged in their internal

That experiment with the dots on paper sounds totally fascinating! And I
agree with Derek's comments on quantum effects _ I find that a bit hard to
swallow too, even though I trained as a physicist! Perhaps it is because I
was an applied physicist rather than theoretical? 

I once witnessed a debate on this subject between Roger Penrose and John
Taylor. To those who don't know one or either of these people, Roger
Penrose is a great exponent of quantum consciousness while John Taylor,
being a mathematician involved in algorithm development for computer
simulations of artificial neural networks, is a staunch determinist. My
alliance tends towards John Taylor. Having said that, I only have a
superficial knowledge of Roger Penrose's arguments and my objection to
them is based almost purely on 'gut feeling'.  Anyway, this was a most
interesting debate between two people on opposite sides and it degenerated
into a brawl because really, these are just theories at the moment. Great
fun watching great minds battling! 

In answer to the original wet/dry question, I believe that ultimately any
'machine' which is to resemble our brain in function or processing power
(and I say resemble, NOT emulate) will have to be of biological origin. I
think this because such a large diversity of structures and so many
neurones, so highly and densely interconnected, will be needed. I cannot
see (at least in the foreseeable future) any technology we have at the
moment even beginning to crack this problem.  Computer simulations and
dedicated electronic circuits are useful in research and industry but they
don't yet even have the processing power of a house-fly! (This is where
someone pipes up saying that such and such a machine DOES have the
processing power of a fly. In that case, change my house-fly to a frog and
there you go!) 

I was involved in building a demonstration opto-electronic neural net
system, using holograms and spatial light modulators (i.e. transparent
mini TV screens). This one was a digital system (i.e. all uncertainty
removed, at least in the hardware level. That is not to say that chaos
would not play a part in any information processing derived from this
system; after all chaos is modelled on a digital computer.) where many
tiny light beamlets replaced electric wires. The advantage of light is
that beamlets can pass through each other without interfering whereas
electric wires cannot, thus allowing a much denser and faster
interconnection scheme. 

In theory opto-electronic systems are better suited to neural nets, and
parallel processing in general, than purely electronic systems but the
technology is now at the stage that electronics was in the 1930's!  We
would need many billions of any currency invested to bring
opto-electronics to the level of sophistication that pure electronics is
at and this is going to take a long time. Take the motor car as an
example: a top of the range BMW or Aston Martin or anything still has
basically the same engine as the model T Ford. Refined ad-infinitum, much
more efficient, much better than the T engine, but still not a good as
some alternative designs which haven't made it, simply because of the
industrial inertia behind the conventional engine where a small
improvement in the short term is deemed more desirable than the risk of
trying for a different design which may be much better in the long run. I
think this is happening with opto-electronic processing too.

Please excuse me for these long paragraphs, I do get carried away
sometimes! So, my point is that to build a 'brain-like' machine we will
need a technology that has a great capacity for many different types of
processing elements (neurones) and for interconnecting them in many
different ways and for implementing the transfer of learning information _
i.e. for making the neurones adapt to their inputs and organise themselves
and their interconnections. We could try to slog laboriously through this
process using 'hard' technologies such as electronics or opto-electronics
(digital or analogue) but it would be much more a practical to devise a
biological model which would grow itself. This approach has similarities
with the fractal methods mentioned by Laurie McRobert.

Indeed, as I wrote in my first article, I believe that this is the only
way we will ever manage to create something even close in "intelligence"
(whatever that means!) to ourselves. 

Dean Pignon

London, UK

to (un)subscribe  the Forum just mail to
fleshfactor-request@aec.at (message text 'subscribe'/'unsubscribe')
send messages to fleshfactor@aec.at

[FleshFactor] [subscribe]