Return to top; or to the
weblog.html Weblog.
The conception, creation and diaspora of artificial Minds in various
computer programming languages - such as
C++ -
Java -
Lisp -
Perl -
Prolog -
Python -
Ruby -
Scheme -
Visual Basic - etc. -
requires a prescriptive mind-model of what is to be implemented.
As in physics, a Standard Model gradually emerges from the
various candidate theories of mind:
/^^^^^^^^^^^\ How A Mind Generates A Thought /^^^^^^^^^^^\ / EYE \ CONCEPTS / EAR \ | _______ | | | | __________ | | | / cat \!!!!|!!!|!| | / \ | | | / image \---|---|-+ | ( Sentence )-------|-------------\ | | \ recog / | |c| | \__________/ | | | | \_______/ | |a| | | \ ______ | auditory | | | recognition | |t| | | \/ Verb \ | | | | of a cat | |s|e| | ( Phrase ) | memory | | | initiates | | |a| ___V__ /\______/ | | | | spreading | f| |t| / Noun \/ | | channel | | | activation | i| | | ( Phrase ) | | ________ | | | _______ | s| | | \______/ _V_____ | / \ | | | / new \ | h|_|_| | /English\ | / "cats" \| | | / percept \ | / \ __V____ \ Verbs /-|-\ "eat" / | | \ engram /---|--\ Psi /--/ Nouns \ \_____/ | \ "fish" / | | \_______/ | \___/ \_______/-----------|---\______/ |
The mind-model diagrammed immediately above is proposed here as a
Standard Model by default, because it was the prescriptive genome
of early AI Minds such as
Mind.Forth and the
JavaScript Mind-1.1
tutorial program listed with full source code in the AI textbook,
AI4U: Mind-1.1 Programmer's Manual published in 2002.
The mindcore array "Psi" of the Open Source Artificial Mind shall
include standard concepts. For example, there may be an initial
set of 64, 128, 256 or 512 "
psi" concepts in the mindcore array.
These mindcore concepts may be agreed upon by convention among AI
designers, and may be suggested by various on-line or hard-copy
listings of the "most frequent words" of natural human languages.
(9-16) Octave two: adverbs
9. also
10. else
11. how
12. not
13. then
14. when (also under conjunctions)
15. where
16. why
(17-24) Octave three:
conjunctions
17. and
18. because
19. either
20. if
21. or
22. that
23. when (also under adverbs)
24. whether
(25-32) Octave four: interjections
25. good-bye
26. hello
27. no (also under adjectives)
28. oh
29. ouch
30. please
31. thanks
32. yes
(33-40) Octave five:
nouns
33. Andru (a name for a robot AI)
34. bugs (e.g., "Fish eat bugs.")
35. cats (e.g., "Cats eat fish.")
36. fish
37. people
38. persons
39. robots
40. things
(41-48) Octave six: prepositions
41. at
42. for
43. from
44. in
45. of
46. on
47. to (also part of infinitive verbs)
48. with
(49-56) Octave seven: pronouns
49. he, she, it (may require further breakdown)
50. I, me (the concept of self)
51. nothing
52. they
53. we
54. what (necessary for questions)
55. who (necessary for questions)
56. you (part of the concept of "other")
(57-64) Octave eight:
verbs
57. am, is, are
58. be
59. do
60. hear
61. know
62. see
63.
think
64. understand
Any AI designer who wishes to incorporate additional core concepts
into the AI bootstrap may simply start the learning of new concepts
at a higher number, such as 129 or 257 instead of concept number 65.
The mindcore concept range of numbers 65 to 128 may be considered a
"free zone" for neurotheoreticians to try out candidate concepts.
CPU width:
- 64-bit
- 32-bit
- 16-bit
- special CPU's, e.g., the Sony "Emotion Engine"
Proposed AI Standard: 64-bit CPU architectures
If your environment (read: budget or equipment already available)
affords you a choice of a 64-bit (or higher) platform on which to
implement your AI, please favor the 64-bit over, say, a 32-bit
platform -- for several reasons. The foremost reason is the
enormous memory space available to a 64-bit operating system --
which is crucially important to an artificial mind consisting
mainly of experiential memory channels under the control of a
arelatively small superstructure of information-routing mechanisms.
A 32-bit platform may be just right or even overkill for a word-
processor, but ultimately too confining in for an artificial
128-bit or higher platforms are a valuable, special case and should
not at all be avoided in favor of 64-bit systems. A platform with
the majestic power of a 128-bit CPU deserves to have an AI
specially written for it.
64-bit systems are the wave of the future and AI is the most futuristic
of all computer applications. Therefore, if we have a choice, it is
better to code 64-bit AI than to linger amid the inertia of substandard
albeit prevalent 32-bit architectures.
Standardizing the AI
sensorium will allow manufacturers to make input
devices in standard ways to ensure usability with standards-compliant
AI software. Early adopters have a chance to create the initial
standards
for exotic sensory input channels that only a robot may have, such
as:
barcode scanners -- database access; vision substitute;
compasses -- for a built-in sense of direction;
Geiger counters -- for a sense of radioactivity;
global positioning system (GPS) -- for a sense of geographic
lcation;
infrared vision -- for a superhuman visual range;
radar -- for a psychic "air traffic control tower";
sonar -- for an underwater range-finding sense;
voice stress analyzer -- for sensing lies in humans;
webcams -- for telepresence.
Standardizing devices for
motorium output hardware will permit
various robot attachments and other hardware to be used
interchangeably with standards-compliant AI.
The achievement of Massively Parallel Processing (MPP) in AI is a
journey, not an instantly attainable state of Mind. In the first
upswoop of our incipient Singularity, we disregard or simulate
massively parallel (maspar) data-flow by such techniques as using
the first result in a search where a biological brain-mind would
have used a winning result from a massively parallel competition;
by iterating serially through an entire loop of single operations
that would be performed all at once in a maspar mind; and by
substituting computer reliability in executing a single software
instruction in place of the much greater reliability attainable
in a biological brain-mind by dint of the sheer redundancy of
massively parallel associative tags forming a maspar mindgrid.
The
Concept-Fiber Theory of Mind describes a massively parallel
mindgrid. As an exercise in understanding or teaching AI theory,
it is possible to show (or require the showing of) how maspar
data-flow is preserved at any stage of the thinking that occurs
between non-maspar input and non-maspar output. For instance, the
maspar recognition of sensory inputs across maspar associative tags
leads to the maspar activation of massively redundant gangs of
concept-fibers, subject to the maspar control of a superstructure
that generates thought by flushing out the most active chains of
association snaking and meandering across the maspar mindgrid.
Some items for discussion among software standards are:
- programming techniques;
- documentation;
- arrays;
- naming standards;
- variables.
Most standards issues involve not the code but the need to
design the artificial mind in such a way that it conforms
to commonly accepted practices in all the sub-disciplines
of the multidisciplinary project.
The most basic software standard among AI Mind projects is
the order of calls to subroutines (modules) from the main
Alife loop, as already widely promulgated on-line at the
http://www.cpan.org/authors/id/M/ME/MENTIFEX/mind.txt
site of the Comprehensive Perl Archive Network (CPAN):
[tabularasa ("clean slate") -- to clear out memory];
[bootstrap -- to load up some standard mindcore concepts];
Sensorium -- to start the flow of sensory input;
Emotion (when implemented) -- for emotional reaction to
input/reentry;
Think -- to generate a thought based on input +/- emotion;
Volition (free will) -- to let thought result in action;
Motorium -- to take action based upon thought +/- emotion.
Since making a loop is one of the easiest things in any given
programming language, any would-be AI coder may have a potentially
dramatic effect on the further development of the basic Seed AI in
a particular programming language by implementing the above loop
and by either posting the code to Usenet or hosting the code on
a Web site -- with a facsimile of the public domain "AI Standards"
document or a link to this page. Having AI creation standards as
unobligatory guidelines makes it easier for Seed AI participants
to understand other people's code and to write their own AI Minds.
nen [n(umber of an) En(glish)" concept] is a concept
number for English.
Since an AI Mind program may easily contain not just English but two
or three or more natural human languages (e.g., for machine
translation),
the variable "nen" is formed by a proposed convention of
joining "n"
for "number" and "en" for English; or "nfr" for French; or "njp" for
Japanese;
or "n" plus any one of dozens of other two-letter codes for names of
languages as found online at the website
ISO 639:1988 Extract: Codes for names of language.
http://palimpsest.stanford.edu/lex/iso639.html
There is a difference between the relationship of the artificial
mind to the world-at-large, and the relationship of the operator
or coder of the AI to the AI software as seen within the human-
computer interface (
HCI). The programmer uses the HCI to
create and enlarge the artificial Mind, which in turn users a
"reality" interface (virtual or non-virtual) to relate as a Mind
and as a person to the external and internal world.
/projects/mind
Last updated: 10 September 2003
Return to top; or to the
weblog.html Weblog.