*VLSI/AGI minimum system requirements*
Boris, let me state some of my intellectual standards - a practical VLSI -
AGI - SIGI prototype this year, at worst - next, with minimum system
requirements ~ Dual Core Core 2 Duo 2,5 - 3,33 GHz, 3 GB RAM (32-bit OS),
any OpenCL-capable GPU, a web cam, microphone and Internet. A better
configuration will include two or more cameras, Xtion or Kinekt, and more
microphones.
The limitation due to CPU would be only in sensory resolution and memory,
but human-level is not quite demanding. I suspect that it could run with
several times smaller resident RAM.
And the road is near, if everything is fine I may have a substantial part
of my basic augmenting cognitive infrastructure rolling next month, then it
would provide accelerating development and research speed, and will be the
beginning of the incremental learning process of those prototypes.
In general, I expect that a Core i5 with 32 GB RAM and a mid-range or
high-range GPU (say from 470 GTX up) should be more than enough for a super
human VLSI that evolves like a rocket, and is limited only by its sensors
and actuators.
* Human-Level is a low bar*
IMO "human-level" is a pretty low bar, with its big latency and silly
memory. Sure, I have to prove it.
Matt's bullshit about the 10^9999 PFLOPS and 25 billion dollars are based
on his (and of the majority of people) multi-domain blindness and
cognitive thresholds, they don't understand stuff that's more than obvious
and trivial for ones who do understand. "The hard problems of arts,
vision, ..." - what hard problems, where are they, why I am capable in all
arts, all technical fields, in languages, vision, motion, etc. etc. and
can't see anything hard, and "intellectuals" who are less than laypeople in
those fields find it hard? Obviously why - they can't deal with those
fields. Sorry - some people can.
IMO many of those "hard problems" would be *incrementally* solvable even on
80-ies or early 90's PCs - just after the right software was developed.
The problem is that the ones who do understand all domains and can see and
feel this, and bridge it, such as myself, are very rare, the others try to
figure out things which are beyond their cognitive thresholds, horizons and
memory capacity, so they cycle-recycle-cycle, but the growing computing
power allows step-by-step going closer to the horizon, even with
intrinsically inefficient methods.
*Boris, your compuational requirements?*
Boris, you also don't have any idea what computing power you need - "a
zillion"; you have a formal incremental theory - but it's pseudocode which
you still cannot map to reality or to computational bounds ("piece of cake"
- bullshit), i.e. you're not sure what you really increment and can't boot
it, you don't have a boot-"sector" instead in pseudocode - that you cannot
run, except only only for you to understand. That's partially absurd and
contradicting your own methodology.
*Syntactic Overhead Nonsense
*
The excuse of yours is "syntactic overhead" ... However,
* The syntactic overhead can always be compensated with a shorter
representation of the payload data, either compressed or reduced; or
reduced number of represented items; and since "generalization is
reduction", that implies that the payload data is small anyway. And most
concepts with which humans operate are *extremely* trivial.
As of the "offenses" to me:
*Newborns are not interested in everything*
* Newborns are not interested in everything, they have too small cognitive
capacity, low resolution and too little experience in order to be, and
actually are dead slow learners - one year to utter the first intentional
word? Give me a break, that should happen in the first day or the first
week if the brain was a fast learner and had appropriate input bandwidth.
On the other hand a twenkid/VLSI improves and learns like a rocket in all
directions, with almost no learning curve, and ideally in an increasing
pace. Of course that also requires appropriately good memory - which most
people lack, their buffers are too short for Versatile Limitless Self
Improvement/Truly General Intelligence.
*The more you learn and invent, and the more you generalize the knowledge,
the faster you should learn and invent. *
* The more you learn and invent, and the more you generalize the knowledge,
the faster you should learn and invent. Why not study another science or
technology, if you can master it in a few days or by reading a few papers
and textbooks, in the mean-time, while you are in the toilet, for
example.And if you're less than a layman or a layman in it, on what is
based your confidence that it's "dumb" and doesn't give proves or disproves
for the methodology etc.
* The more you learn and invent, the more technologies to accelerate the
work you're supposed to develop and to increase the bandwidth and to unload
your brain.
*"Boredom" reasons poles: Saturated Progress from the bottom to the top
or... Zero Progress forever*
* There are two major branches of cognitive reasons to stop or pause
studying, or - a more precise term - improving - in a particular
domain/modality/field/direction, besides pressure from competing domains:
* 1. You have exhausted/saturated it, you've progressed all the way
through and there's no more stairs up.* It's already obvious, too easy/was
too easy; or it is now too hard/demanding too much resources to make a
further progress, and the expected new progress is too insignificant,
compared to the path covered already. You have galloped through it or
understood it at a glance and you're a master of it, while you see the
others have struggled and have progressed very slowly.
*2. You haven't even scratched it. You haven't progressed at all - was
too hard from the start. You've been exposed to the domain, but you've made
no progress.* You've struggled, while the others were galloping through it.
Then the domain was abandoned - it was "boring", "piece of cake",
"deductive reasoning", "blah-blah, 999999 PFLOPS for meaningful results,
too much overhead from the syntax (bullshit), 3848394*10^56 TiBiGibiBytes,
the hard problems in vision, language, arts, blah-blah-blah, I'm
intellectual, PhD, generalist, "it's a piece of cake", I have 256
publications in Journals, but I'm so narrow-"general" brained, that I can't
learn to draw a face or can't play 30 tones with a guitar, or can't
improvise and play another character, or study something for 1 year, which
can be learnt in 5 days, blah-blah-blah".
Yeah, in case of 2, someone may invent elaborated "abstractions" which
express instructions and concepts which are obvious pseudo abstractions
from the POV of 1, but are beyond the horizon of 2.
=== Todor "Tosh" Arnaudov ===
... Toshko 2 - Bulgarian Text-To-Speech Synthesizer -
http://twenkid.com/software/toshko2/
.... Twenkid Research: http://research.twenkid.com
.... Author of the world first University courses in AGI (2010, 2011):
http://artificial-mind.blogspot.com/2010/04/universal-artificial-intelligence.html
.... Todor Arnaudov's Researches Blog: http://artificial-mind.blogspot.com*
*
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com
