om
Since I'm too lazy (not superhuman enough) to master the x86 PC enough
to write an operating system for it I started looking into what it would
take to write an AI to do it for me. ;)
(Specificly, I have an "OS-TEST" machine that I'm trying to scrape
togeather an OS for... I am having trouble getting a minimal leenooks
system installed on it properly cuz leenooks sux. =(
The two main issues seem to be A: establishing a cybernetic model for
the initial AI design. and B: establishing motovators that will cause
the AI to behave like an operating system and/or properly maintain the
services that the user requires.
In considering the guts of my dual athlon I came to the conclusion that
any attempt to discover the functionality of the system by raw probing
would only be partially sucessful at best (many of the behaviors such as
video and sound can only be understood by a feedback mechanism outside
the machine.) Beyond that chipsets are known to have hidden "features"
that are functionally equivalent to the fabled "HCF" opcode -- the
machine could be physicaly damaged.
So the development would necessarily have to start in some padded
environment such as Squeak or a simplified virtual machine such as
Kunth's MMIX.
OK. (om)
MMIX is good because it is simple and should be very easy to learn, or
at least not require inordinant complexity on the part of the AI to
manipulate. (necessary if a hand-audit of the AI's text is required).
This would need some kind of modality for the registers, and memory
locations... Eliezier has made some noieses about a modality for
sourcecode but he has shown no interest in sharing that with the
community as he seems hell-bent on dominating the universe himself. =\
In considering registers and register values I was reminded of the work
of Lenant and his work on an automated theorem prover before he started
thy cyc project. Numbers seem pretty stupid (simple) at first but people
have managed, in a pseudoschizophrenic way to find all kinds of patterns
and magical properties. =P
Lenant's program WAS succesful in producing some rather boring results
but it soon ran out of steam. So how do mathemeticians work? They work
by coming up with concepts (<< GREEN LIGHT WORD) in the form of formal
systems which they create at and manipulate. Examples of these
conceptual creations are calculus, set theory, graph theory,
quantification theory etc... The key feature is the creation of concepts
such as the existential quantifyer which leads to discoveries.
The contents of your computer is basically an enormously large number.
It would seem that number theory is like the wellspring from which all
other logics, thinking, and perception arise.
In writing essays for singularityactiongroup.com (and even too this
message!) I used a process where I picked four or five ideas that I
wished to express. These ideas were then intermixed in a highly complex
way to form the themes of the essay. It would seem to me (though I can't
prove it just yet) that these ideas are like huge prime numbers and the
final essay is the synthesis of these. It would seem that if you were
able to distill the ideas enough identify these primes and then multiply
them togeather they would form an idealized version of the essay
(through the inverse of the distilation process) you would get God's
version of the essay. ;)
So how does this help us create a modality for a computer memory? Well,
it would seem to point to a complexity heirarchy for cybernetic systems:
Lets do the reverse of chomskey and work our way upwards:
CLASS 1: TRIVIAL SYSTEMS
Systems that perform little or no computation on inputs -- one or two
fixed operations at most. An example being the thermostat and the
streach reflex in your muscles.
Class 2: COMPUTATIONAL SYSTEMS:
Systems that perform some complex, or variable computation on input
parameters. They may include state machines and memories but don't
provide a general computing environment. Examples include the adaptable
autonomic functions of your mind including your emotions and complex
reflexes. To pick a current and most unfortunate example, the space
shuttle's flight computer. In addition to its trivial reaction of moving
the flaps to steer it changed its behavior to using the steering jets as
things started to come appart. The engine controll computer on ascent
does complex matrix manipulations to calculate the proper alignment of
the engine nozles.
Class 3: Intelligent systems
Systems that are able to self-program and generate task-specific
computational systems to solve problems that are intractable for purely
computational systems or to provide solutions in an unpredictable
environment. Examples would include good Chess and Go playing programs.
Systems which can navigate in real-world conditions. Systems which can
generate appropriate pseudo-natural languages for a given complex
system. And, apparently, being a number theory nut.
(I'm only hard on the mathemticians because I envy them so much!)
Since mathematics is the one science that one could study should the
universe cease to exist it is very well suited to serving to develop AI
on a machine with little input or output capability. It would seem that
mathematics IS the archeotype of all other modalities though it is not
clear at all how useful this might be in designing good high-performance
real-world solutions...
Math is a highly abstract system that would be very difficult for a
person to work on. An easier approach would seem to be "raising" an AI
in a very friendly environment such as the Squeak system. (which is why
I need linux to work on that damn Dual Athlon, damnit!)
Lets immagine that we have crude but working modalities for the AI and a
cognitive matrix that seems to be up to the task. How do we direct the
thing's behavior to productive tasks? (such as being the user's loyal
minion).
In general, I think it is safe to say that we are most interested in
*motovators* and *inhibitors*.
So far I have come up with the following motovators:
GENERAL AROUSAL MOTOVATOR.
Humans seem to require a general arousal motovator that causes the
brain to respond to its environment, without which the person becomes
comatose. This should be investigated thouroughly to determine wheather
an AI entity will require it too.
NVELTY MOTOVATOR
Some kind of drive/reward system which monitors the cognitive matrix
and gives an appropriate reward when a new concept is added, and induce
boredom when few new concepts have been added reciently.
"om" motonawor.
A thought-induction motovator that may simply be another face of the
general arousal motovator. This motovator should be engineered to cause
the AI to constantly seek greater wisdom (om) and increasingly sound
concepts and logic.
(these are being copied directly from a scratch sheet hence their not-so
subtle humor. ;) This post has taken me most of a week to work out
though I'm only writing it tonight.)
D0S MOTOVATOR. ;)
Since I idolize and revere DOS I use the name to describe the motovator
which would drive the AI to present itself in an understandable and
useful way to the user and be a trustworthy steward of the user's
digital affairs. -- A motovator to make it behave like an idealized
operating system. =) ("Good morning Dave".)
Since a motovator of this nature is so alien to anything found in
nature it is unclear how this idea will evolve before it will become a
working piece of software.
Humans seem to have an innate drive to talk and to learn language. So it
would seem that a COMMUNICATION MOTOVATOR is required. This would
probably have to be reversed engineered from the human "language
acquizition device"
Finally, it would seem to be desirable to have a "QUEST MOTOVATOR" to
focus and direct the AI's attention to complex and long-term projects
(such as this one of developing AI) on behalf of the user or, in the
abscence of user input, towards some other long-range goal. This might
not actually be necessary. This motovator does require a
user-controllable kill-switch to stop the AI from working on projects
that are not a part of the user's agenda.
---- INHIBITORS ---
The first inhibitor is, naturally a "Harm Humans Inhibitor" (probably
paired with a "DISOBEY HUMANS INHIBITOR". This inhibitor would work by
actively suppressing lines of thinking that might go against the human
user's wishes or be otherwise illegal.
It would seem, at first, that a "FUTILE BEHAVIOR INHIBITOR" would be
entirely a good thing, stopping the AI lots of time. However this could
backfire badly if the AI was trying to educate some clueless newbie with
little sucess, hence it must be disabled when dealing with humans and
strongly bound to the disobey humans inhibitor.
The only other inhibitor that I could think of was an inhibitor against
doing things that would screw up its underlying system. For well known
reasons this should not be an active motovator.
Finally, the AI would require an "input regulator" or some other device
(such as the thalamus) which serves to assign the varrious cognitive
tasks appropriate resources.
I am becoming highly anxious to implement this stuff. ;) As I mentioned
above I do need help getting a working "Squeak on thin linux" platform
working. With some hax0ring squeak's rich multimedia features and RAD
optomizations will make it an ideal platform for AI development.
Squeak supports multiple mouse pointers so giving the AI a hand should
be no challenge at all. =) Squeak also has some impressive facilities
for reading the screen buffer and evenn has a tool that can read the
area of screen surrounding the mouse and magnify it in a seperate
display object. This would be perfect for creating a roving eye... As it
would be impractical to pattern-analyze the entire screen in hi-res (and
because it would be useful to know what the AI is looking at) the screen
buffer could be analyzed at a lower resolution (say 1:4) while the
information near the AI's eye can be analyzed at full resoultion and
then integrated into the low-resolution impression of the screen.
I CAN'T WAIT TO GET THIS DONE!!!!! =))))
PS: I have a .bmp picture of myself that I need translated into jpeg and
perhaps png for wider distribution. Can anyone help me with that?
--
I WANT A DEC ALPHA!!! =)
21364: THE UNDISPUTED GOD OF ALL CPUS.
http://users.rcn.com/alangrimes/
[if rcn.com doesn't work, try erols.com ]
-------
To unsubscribe, change your address, or temporarily deactivate your subscription,
please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
