As usual, I have a pile of disorganized thoughts without a clear idea
what I'm trying to argue or what point I want to make, nevertheless I
want to make a post so make a post I will!
The main focus (if you can call it that) of this post is to review the
elemests that make up the human mind, compared against the elements we
see in our pre-nascant AGI systems and try to establish a workable
taxonomy/ontology. Importantly, the question of whether AGI is viable as
a technology. What I mean by that, is whether AGI can become a viable
engineering discipline. The alternative being that the pratcice of AGI
will always be a form of sorcery where a procedure for producing a mind
in an artificial medium is known but never understood.
Human neurology, it is understood, is a totem-pole of systems that
co-evolved with our evolutionary ancestors. While each of these systems
still produces a detectable behavioral signal, it is not immediately
obvious how many of them are still essential to producing a healthy
individual, fewer of which are required to produce a mind that is
intelligent but not quite human, fewer still that produce a mind but
can't quite operate a human body in real time.
So up and down the spinal cord, including the thalamus you find a number
of circuits and modules that produce basic reflexes. Some of these are
VERY simple like the stretch reflex for muscles, it tries to protect the
joints by compensating for a sudden stretching of a muscle, much faster
than any of the higher centers can. Higher reflexes deal with issues in
the respiratory and gastro-intestinal systems. You also find the
mechanisms underlying emotional affects such as laughing, crying, etc
all in thalamic reflexes.
Paralell to this, there are the sensory-motor pathways. These pathways
may accomplish a number of complex computations including motion
detection, spectrum analysis (such as of aural stimuli), localization of
visual and aural sources, etc... But all of these functions greatly
improve the brain's speed and accuracy at perceiving the world but may
not be strictly necessary. You can probably heap a bunch of nerves,
neucleii, and brain anatomy into this category, cerebellum, the
colliculi, the LGN, lots of stuff... There's a ton of functional modules
that have been identified in human anatomy. The key thing here is that
they're all fixed-function and only learn to the extent necessary to
provide decent enough performance given developmental and lifetime
changes to the anatomy.
[ just walked over to the mechanic to check on their progress on my 23
year old rocket-sled (Honda Civic) that they've had for like 2 weeks, on
my way I passed an uber eats ** DELIVERY DROID ** on the sidewalk. damn
the future is coming quick.. Mechanic had the engine mostly disassembled
and was trying to install new rod bearings, the new ones were too tight
and were preventing the crank from turning... UGH, they say they'll have
it done tomorrow but who knows how they'll botch it up... ]
Jumping to the top level, we have a network that, it now seems, is
dependent on its computational properties and its raw scale, and little
else. Lets call this scale-emergent. The GPT systems have this property
(or a close-enough approximation of it ) and the brain has it. Given a
decently good basic topology, specifically the cortico-thalamo-cortical
loop that exists in the brain, involving a multi-stage network that
starts at the cortex, runs through the basal-ganglia, the putamen, the
STN, the thalamus, and finally recurrently connects to another cortical
region. Cortico-cortico connections are also quite prevalent.
This sounds suspiciously to me like what Greg Egan described in
Permutation City all those years ago where a new "citizen" was
manufactured by constructing a matrix of sufficient size and then
spawned it out into the VR to experience some analog of a childhood
until the training process produced a functioning mind. The irksome
thing about this type of cognitive system is that it is virtually
impenetrable from the outside. It is a pseudo-dualistic entity that
exists only within a sufficiently large computationally flat space. The
brain's firmware is essentially mostly in the hypothalamus which has a
dedicated neural pathway to the amigdala. These are where you get your
hungers and lusts and stuff but they operate as self-contained control
systems that run in paralell to the brain. The scale-emergent parts of
the mind have basically no access to them (or vice versa) instead they
seem to operate by reading data flows through the adjacent thalamus and
then triggering reflexive, sensational, or neuro-transmitter based
responses when their input conditions are met. The scale-emergent higher
mind only reverse engineers these to the exent that it is able to
anticipate and influence them a bit.
The vexing thing is that if this situation proves robust, it means that
we can already see a brick wall WRT AGI. It means we can only create one
of these matrix thingies, watch it develop a "ghost" or "soul" or
whatever you want to call it, and then have it live for a while until it
eventually degrades or its training becomes obsolete and it has to be
retired. We, ourselves, are similar, where our scale-emergent cortices
have so vastly transcended our biological evolution, but we may now be
doomed by how inaccessible our emergent minds are from the physical
substrate.
It always needs to be repated that uploading does NOT solve these
issues. Indeed uploading itself becomes that much more difficult as we
are not trying to copy a set of files, as it is generally
over-simplified as being, but to duplicate a system with very fuzzy
emergent properties. The EEG signals indicate that to even maintain your
base-identity, a specific brainwave pattern must be maintained. Patients
who claim to have disascociated identities often do exhibit several
different recognizable brainwave states. So getting that to function to
the point that you can bullshit the identity problem will already be a
minor miracle. Even with that solved, you are back at the fundamental
problem, that the scale-emergent parts are largely opaque, means that
making meaningful upgrades to the mind will be futile.
What we see in human psychology is that we start out with a lot of fluid
intelligence and gradually lose that in favor of learned "crystaline"
intelligence, essentially becoming walking libraries of obsolete
knowledge. One of the things I'm fond of about my rocket-sled is how
well-proportioned its whels are, it has 24" outer diameter wheels
composed of 14" rims and 60R tires, which are becoming scarce. In terms
of new cars, you'll be hard pressed to find anything with less than a
18" rim, the higher end models forcing up to 22" rims, and painted-on
tires my mind is like GAAAAH!!! Yeah, I'm becoming an old man. =\ The
point I'm trynig to make is that a brain emulation will simply freeze up
like my mind has and you just won't have something that will last very
long, especially if the "patient" was already elderly.
The next problem is whether there exists an algorithm that can produce a
"well structured" mind that also exhibits similar capabilities as the
monolithtic scale-emergent systems.
--
Beware of Zombies. =O
#EggCrisis #BlackWinter
White is the new Kulak.
Powers are not rights.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/T59f369fee7febd6d-Mc800b01fec97b67fd7c1c982
Delivery options: https://agi.topicbox.com/groups/agi/subscription
