Re: Neural Networks

[Jack Park <***>]

I suppose I can volunteer to be the geek who fell victim to two
different camps: traditional (whatever that means) AI, and the
connectionist camp, which includes the neural net stuff. I tend to
think there has been an enormous amount of scholarship going on in
both camps. Is either right? Define "right."

In the beginning, if you count funding sources as the roots of all
scholarship these days, ARPA, later DARPA started up the AI program,
but a fellow named Lickleider couldn't wait for that to happen, so he
started the Augmentation program. Again, in the beginning, the AI camp
spoke in terms of replacing humans, performing real (define "real")
thinking (define "thinking"). Of course, we all know such boasting
turned out to attract lots of flying, uncooked egges, most of which
hit home. It was the augmentation program that seems to have won out,
largely by having the AI jockeys start talking in terms of augmenting
human cognition (define "cognition"), and a lot of really useful
software began to fall out of the AI schools.

As I see things, the neural net folks don't have a stranglehold on
received wisdom either. As Iben Browning once said, they ignore the
opportunity for phase encoding of information that is just as liable
to be going on in brains as not. This happens when you perform all
your computational magic in a thresholded analog domain and ignore the
pulse encoding we know to exist in nervous systems. Browning built a
pulse-encoded neural net that solves interesting problems. I
replicated some of those maybe 15 years ago.

Another story: I built a traditional AI program (expert system)
coupled to a data acquisition system and a process controller -- right
out of the Rosen Modeling Relation diagram, and successfully changed
the way polymers are cured in autoclaves. A patent (not my idea) was
granted on the architecture that performed this magic, so I built a
nine-node "neural net" that successfully duplicated the cure cycle on
an autoclave simulator. I don't think any one technology has a
stranglehold on received wisdom. In the end, my take is this. Given
that we are saddled with the "von Neumann" bottleneck of single
processor computers, until we have orders of magnitude more single
processor computers doing the computations, the present paradigm will
not be the one that delivers up what the original AI jockeys once
claimed. Indeed, arguments for "embodied" cognition are, in every
sense (my opinion) just arguments for relational thinking in much
higher dimensions than we now are capable of performing.

Having said that, my claim remains that the augmentation program
rocks! I am saying that we need now to jump on every possible
augmentation program available to us in order to tackle the complex
and urgent problems humankind now faces (and continues to create!). We
need world leaders capable of thinking in higher dimensions than we
now have, and if we can't find them, we need to augment the
capabilities of those we can find. The time is long past (the train
left the station a long time ago) for contemplating navels, worrying
about which technology is better. Time to find ways to semantically
integrate all possible technologies and get on with problem solving in
an anticipatory rather than reactive fashion.

My 0.02 EUROs for the day.
Jack