AI summary
George
> I think AI is mainly about:
>
> Knowledge base indexing/searching
> Goals/interests/intentionality
> Problem solving, analysis, reasoning, logic, problem
> representation,
> managing states, plans, theories, conclusions, focus,
> attention
> management
> Information theory, representating/filtering/summarizing
> content,
> consideration/relevance theory and management, data
> filtering
> Knowledge, representation of, symbols, situation and world
>
> models, cross-referencing, explanations, maps, ideas,
> limits
> Logic, communication, general syntax/styles/considerations
>
> Linguistics, grammar, natural language processing,
> thought, discussion
> Learning, building successfully adapting neural pathways,
> conditioning
> Dynamic regulatory control systems (drives)
> Evolutionary algorithms
> Decision making, probabilities and certainty
> Preference management, familiarity, belief networks (the
> shoulds
> and should nots... A is better than B because...,
> etc.)
> Games, performance, excercises, challenges, competition,
> improving winning strategies, risk management,
> dominance,
> dealing with equality/trade issues, agents, autonomy
> Robot art, completeness, inspiration, creativity,
> appreciation, moods,
> passions, emotions, desires
> Robot consciousness and awareness
> Image and voice data architectures
> Etc.
>
> How boring.
>
> George
I may end up being an AI academic. Imagine. I would start
the class
telling the students with crossed eyes pointing at them one
by one:
you are being watched!!! You are all fucking paranoid scums!
By
the time you graduate, you will be either dead or lucky
being locked
up in a nuthouse. Now is the time to quit and find a descent
job,
maybe become a respected chef in McDonald's, and in 20
years,
if you work hard enough, you might make it to becoming a
manager.
A lifetime supply of free Big Macs, or become an AI dead
freak.
Your choice. You have 10 seconds to make up your minds......
Ok, too late, prepare to die! Now let's begin the class,
shall we...
Bwahahahaha!
George
GOGAR
person when say 3 people are talking at the same time..
it looks kinda impossible to me.. well from a signal processing point of
view..
"George" <george...@hotmail.com> wrote in message
news:3CF40F2D...@hotmail.com...
Helmut Leitner
I hardly know anything about AI but I wonder why the word "pattern"
doesn't turn up in your list.
Recently I stumbled over an linguistic classic "The Meaning of Meaning" (about 1930)
by Charles K. Ogden and he analyzes (a long time before AI) how tightly integrated
the recognition of symbols (patterns), our way of thinking and our perception of
the reality really are. A few week later I reread a summary of Karl Popper's
philosophy which expanded on his similar view of the scientific method and the
biological evolution as the creation and testing/falsification of hypothesis
(mutations). His "life is solving problems" seems very near to a "life is the
recognition of patterns in our perception of reality" which would put the
term "pattern" at the very center of AI.
So I would start with the line:
Patterns recognition, representation, transformation
--
Helmut Leitner lei...@hls.via.at
Graz, Austria www.hls-software.com
George
IMO knowledge base indexing/searching is mapping, which is about
dealing with patterns, including pattern recognition, representation
(mapping) and should include object, knowledge or pattern
transformation issues. I guess it is an open terminology at this
point.
> Recently I stumbled over an linguistic classic "The Meaning of Meaning" (about 1930)
> by Charles K. Ogden and he analyzes (a long time before AI) how tightly integrated
> the recognition of symbols (patterns), our way of thinking and our perception of
> the reality really are. A few week later I reread a summary of Karl Popper's
> philosophy which expanded on his similar view of the scientific method and the
> biological evolution as the creation and testing/falsification of hypothesis
> (mutations). His "life is solving problems" seems very near to a "life is the
> recognition of patterns in our perception of reality" which would put the
> term "pattern" at the very center of AI.
I am not so much into definitions but more in explanations. What life is
in terms of patterns does not mean much to me, although art comes
to mind which is heavily based on the study of patterns, completeness
and harmony of nature.
George
George
> Hey George how do you think we can "listen" to and understand a single
> person when say 3 people are talking at the same time..
> it looks kinda impossible to me.. well from a signal processing point of
> view..
I like to see big bouncing breasts too. Once humanity moves
to space, bras and their gravity-support will become redundant
features, though I've heard that some women nipples become
irritated when rubbed against T-shirt fabric for too long, so a
tight T-shirt will still be desirable options. Don't want to see
women with irritated nipples in space, if you know what I mean..
George
George
I think that before every man marries, he should be sent for a
marriage test. He should be sent up to space for a month naked
at all times, with 3 hot horny lesbians with huge boobs, all locked
up in one cabin. See if he can resist cheating. Getting massaged by
the 3 women would not count as cheating IMO.
George
Davin C. Enigl
wrote:
>Helmut Leitner wrote:
>
>> George wrote:
>> >
>> > I think AI is mainly about:
>> >
>> > Knowledge base indexing/searching
. . .
>> > Etc.
>>
>> I hardly know anything about AI but I wonder why the word "pattern"
>> doesn't turn up in your list.
>
>IMO knowledge base indexing/searching is mapping, which is about
>dealing with patterns, including pattern recognition, representation
>(mapping) and should include object, knowledge or pattern
>transformation issues. I guess it is an open terminology at this
>point.
Patterns are important in an indirect way, but they are
over-emphasized IMO.
At the first level: the data-base of patterns are stored (memorized in
memory), then memory search mechanisms act, mapping the memory.
At the next level down: mapping the power set of N-symbol strings in
memory occurs in biological systems (e.g., human brains). The patterns
in memory are important (indirectly) because the memory inputs and
outputs will tend to act upon themselves (patterns acting upon
patters) which create novel patterns and novel memories.
At the next further down level: These novel N-power set -memories can
be acted upon by selection as in evolution (say biological evolution).
However, in the biological brain, this apparently happens at a
subsymbolic level, so we are *not* memorizing completed patterns.
This is far different that (most) current strong AI philosophies -- an
exception is a reconstruction of Minsky and Arbib, made by Clark (in
his Mindware: an introduction to the Philosophy of Cognitive Science,
p. 33). This is however, common in weak AI philosophies (Searle).
So, computer simulation of a biological brain system is *not*
currently being done correctly (IMO), because the computer AIs are
only acting at the symbolic level and not the subsymbolic level.
Patter is being *over* emphasized (symbol is syntax, not semantics)
and in some cases pattern is being incorrectly taken as the end point
of AI simulation.
[Top-Down, AI has not worked very well. That is why I do my research
in the Bottom-Up AI following biological evolution's history as far as
we conjecture it as possible. Of course, being a microbiologist that
is what I know best anyway.]
>> Recently I stumbled over an linguistic classic "The Meaning of Meaning" (about 1930)
>> by Charles K. Ogden and he analyzes (a long time before AI) how tightly integrated
>> the recognition of symbols (patterns), our way of thinking and our perception of
>> the reality really are.
Ogden and Richards was ahead of its time, but I think now, that they
found only the surface without explanation of the deeper levels (e.g.
the "emotive" term what we are "really" doing and what we "think" we
are doing). Popper pointed out the problems with *meaning* and Popper
instead, introduced explanation without definitions of meaning.
Popper did this because he considered definitions and meaning infinite
regressions and therefore, meanings could not important.
>> A few week later I reread a summary of Karl Popper's
>> philosophy which expanded on his similar view
Actually Popper's view was quite the opposite, or I should say
different, from what I think you are trying to say below. You will
find many summaries of Popper that are wrong, especially if the
summary was written by one of the Logical Positivists of the Vienna
Circle. I think the primary sources (Popper) are safer to read.
>> of the scientific method
You should also read Feyerabend and Alan Chalmers for counter-views to
universal scientific method meta-theory.
>and the
>> biological evolution as the creation and testing/falsification of hypothesis
>> (mutations).
Testing of conjectures by refutations vs. corroborations, in evolution
is the difference between death and life for a life-form.
>His "life is solving problems" seems very near to a "life is the
>> recognition of patterns in our perception of reality" which would put the
>> term "pattern" at the very center of AI.
This just seems completely off from Popper. I have no idea where you
are going with this. Popper has a great theory of evolution, but it
has nothing to do with patterns as far as I know.
Also of note for AI is Popper's Psub1 -> TTsubn -> EEsubn -> Psub2n ->
CED this is exosomatic mutation and is found in Popper's tetradic
schema for his theory of *emergent* evolution through problem solving
(nothing to do with patterns). This is in his (1994) Knowledge and
the Body-Mind Problem: In Defense of Interaction, esp. p. 12-13 and p.
62-63.
>I am not so much into definitions but more in explanations.
Popper would agree: definitions are *not* important, and explanations
*are* important. Understanding is semantics and required, but *not*
definition of meaning which is syntactical symbol manipulation.
> What life is
>in terms of patterns does not mean much to me,
Correct. In fact an understanding of life based on patterns would
stop efforts to discover a universal theory of life that could be
applicable to astrobiology (my field) and possibly stop a universal
theory of life from helping develop (strong or weak) AI.
>although art comes
>to mind which is heavily based on the study of patterns, completeness
>and harmony of nature.
Correct, patterns seem more applicable to aesthetics. Some of these
concept can be found in the pre-Socratics and early Greek philosophy.
In more modern times, I would say Herschel Chipp's (1968) Theories of
Modern Art and Benjamin Gal-Or's (1981/1983/1987) book are
interesting, esp. in Gal-Or, Popper's preface, the introduction and
the last chapters.
--Davin C. Enigl
Helmut Leitner
I noticed this, of course.
But patterns are usually applied in a problem specific way.
Perhaps our brain uses generalized pattern algorithms - which seem
pretty hard to understand and to implement.
Perhaps it would help to solve some problems if we developed algorithms
designed for image patterns, accustic patterns and language patterns
at the same time? Common representation, common transformation?
> > Recently I stumbled over an linguistic classic "The Meaning of Meaning" (about 1930)
> > by Charles K. Ogden and he analyzes (a long time before AI) how tightly integrated
> > the recognition of symbols (patterns), our way of thinking and our perception of
> > the reality really are. A few week later I reread a summary of Karl Popper's
> > philosophy which expanded on his similar view of the scientific method and the
> > biological evolution as the creation and testing/falsification of hypothesis
> > (mutations). His "life is solving problems" seems very near to a "life is the
> > recognition of patterns in our perception of reality" which would put the
> > term "pattern" at the very center of AI.
>
> I am not so much into definitions but more in explanations. What life is
> in terms of patterns does not mean much to me, although art comes
> to mind which is heavily based on the study of patterns, completeness
> and harmony of nature.
Definition don't mean a thing to me, either.
Sometimes if I'm stuck I try to change me viewpoint. AI seems to be stuck.
Looking at everything in AI from a *generalized* pattern viewpoint just might
be such a promising change.
marika
> Hey George how do you think we can "listen" to and understand a single
> person when say 3 people are talking at the same time..
> it looks kinda impossible to me.. well from a signal processing point of
> view..
what about if they were singing instead?
Helmut Leitner
"Davin C. Enigl" wrote:
> >Helmut Leitner wrote:
> >> George wrote:
> >> > I think AI is mainly about:
> . . .
> >> > Etc.
> >>
> >> I hardly know anything about AI but I wonder why the word "pattern"
> >> doesn't turn up in your list.
> >
> >IMO knowledge base indexing/searching is mapping, which is about
> >dealing with patterns, including pattern recognition, representation
> >(mapping) and should include object, knowledge or pattern
> >transformation issues. I guess it is an open terminology at this
> >point.
>
> Patterns are important in an indirect way, but they are
> over-emphasized IMO.
>
> ...
Perhaps it's impossible to summarize two philosphical viewpoints
and my own conclusions within a few lines of text.
> You will
> find many summaries of Popper that are wrong, especially if the
> summary was written by one of the Logical Positivists of the Vienna
> Circle. I think the primary sources (Popper) are safer to read.
I'm happy to find someone competent to talk about Popper, although
this isn't my primary concern. I'm Austrian and live only 200 km
from Vienna where he grew up. In fact I'm very fond of Popper and
most of Popper's books are on my bookshelf (whatever this means).
> >> of the scientific method
>
> You should also read Feyerabend and Alan Chalmers for counter-views to
> universal scientific method meta-theory.
My point was only that Popper saw a strong similarity between
biological evolution and the way science works.
> >and the
> >> biological evolution as the creation and testing/falsification of hypothesis
> >> (mutations).
>
> Testing of conjectures by refutations vs. corroborations, in evolution
> is the difference between death and life for a life-form.
This is the point where my conclusions start (you may note the word "seems"):
> >His "life is solving problems" seems very near to a "life is the
> >> recognition of patterns in our perception of reality" which would put the
> >> term "pattern" at the very center of AI.
>
> This just seems completely off from Popper. I have no idea where you
> are going with this. Popper has a great theory of evolution, but it
> has nothing to do with patterns as far as I know.
I didn't intend to argue that it has.
IMO problem solving has much to do with the recognition of patterns,
think of how chess grandmasters "think" or how we solve mathematical
problems. And my use of the word "pattern" has nothing to do with
the usual usage. It aims more at the way a biological system may see,
remember or transform regular structures. This should be pretty near
what you call "subsymbolic".
> Also of note for AI is Popper's Psub1 -> TTsubn -> EEsubn -> Psub2n ->
> CED this is exosomatic mutation and is found in Popper's tetradic
> schema for his theory of *emergent* evolution through problem solving
> (nothing to do with patterns). This is in his (1994) Knowledge and
> the Body-Mind Problem: In Defense of Interaction, esp. p. 12-13 and p.
> 62-63.
>
> >I am not so much into definitions but more in explanations.
>
> Popper would agree: definitions are *not* important, and explanations
> *are* important. Understanding is semantics and required, but *not*
> definition of meaning which is syntactical symbol manipulation.
Are we in a Popper discussion?
> > What life is
> >in terms of patterns does not mean much to me,
>
> Correct. In fact an understanding of life based on patterns would
> stop efforts to discover a universal theory of life that could be
> applicable to astrobiology (my field) and possibly stop a universal
> theory of life from helping develop (strong or weak) AI.
Could you explain why the term "pattern" (in the vage form I use)
is such a threat to your thinking?
P.S. Please also note the work of Christopher Alexander who bases
his whole theories on his interpretation of the word "pattern"
(he directly connects "pattern" and "life" althought I havn't read
his final books about "The Order of Nature" published this year).
George
They can be reused.
>
> Perhaps our brain uses generalized pattern algorithms - which seem
> pretty hard to understand and to implement.
> Perhaps it would help to solve some problems if we developed algorithms
> designed for image patterns, accustic patterns and language patterns
> at the same time? Common representation, common transformation?
It has to be started by collecting information from the environment.
It is just there is aughfully a lot of useful information that needs to
be considered and represented in a useful form. It is hard to write
software to parse information out of a picture.
The quick solution may lay in what Curt Welch suggested, trying
to find something that adapts objectively to the environment
with extremely general pattern recognition rules, though I think all
levels need equal consideration.
>
> > > Recently I stumbled over an linguistic classic "The Meaning of Meaning" (about 1930)
> > > by Charles K. Ogden and he analyzes (a long time before AI) how tightly integrated
> > > the recognition of symbols (patterns), our way of thinking and our perception of
> > > the reality really are. A few week later I reread a summary of Karl Popper's
> > > philosophy which expanded on his similar view of the scientific method and the
> > > biological evolution as the creation and testing/falsification of hypothesis
> > > (mutations). His "life is solving problems" seems very near to a "life is the
> > > recognition of patterns in our perception of reality" which would put the
> > > term "pattern" at the very center of AI.
> >
> > I am not so much into definitions but more in explanations. What life is
> > in terms of patterns does not mean much to me, although art comes
> > to mind which is heavily based on the study of patterns, completeness
> > and harmony of nature.
>
> Definition don't mean a thing to me, either.
>
> Sometimes if I'm stuck I try to change me viewpoint. AI seems to be stuck.
Not to me.
> Looking at everything in AI from a *generalized* pattern viewpoint just might
> be such a promising change.
I listed 17 general areas to consider for AI, I felt it was a fairly complete
list of topics for covering AI.
George
Davin C. Enigl
<helmut....@chello.at> wrote:
[I think I understand your points better now. Thank you, for the
clarifications. --DCE]
>"Davin C. Enigl" wrote:
>> You will
>> find many summaries of Popper that are wrong, especially if the
>> summary was written by one of the Logical Positivists of the Vienna
>> Circle. I think the primary sources (Popper) are safer to read.
>
>I'm happy to find someone competent to talk about Popper, although
>this isn't my primary concern. I'm Austrian and live only 200 km
>from Vienna where he grew up. In fact I'm very fond of Popper and
>most of Popper's books are on my bookshelf (whatever this means).
My (Enigl) family is from Vienna and is also from between Emmersdorf
and Martinsberg down the road west of Krems. I really like Popper but
not just because he is also Austrian :-) of course, and I've read
almost all of his books -- I think I own all of his books in-print,
except his two volumes of the Living Philosophers series by Schilpp.
. . .
>My point was only that Popper saw a strong similarity between
>biological evolution and the way science works.
Oh, OK. Yes, that make sense. What you mean is the conjectured
theory (mutation) and the refutation by experiment (death of wrong
mutations) or corroboration of the theory (survival of the "fit"
mutations by survival of the organism).
. . .
>IMO problem solving has much to do with the recognition of patterns,
Yes, if you read Gearald Edelman's research or Searle's summary (this
summary seems safe so far) in The Mystery of Consciousness, I think
you will find a lot to corroborate your ideas. You may want to read
Edelman's new book (2000) A Universe of Consciousness, or his own
summary of his research in Bright Air, Brilliant Fire, or if you want
more academic detailed writing, try Neural Darwinism or Topology or
Remembered Past. Crick also has his Astonishing Hypotheses (but the
philosophy is wrong unfortunately, the rest of the book is great).
Edelman in particular has a theory about visual patter re-recognition
(via a process named reentry interactions of the different neuronal
maps) and he hypothesizes that the brain re-creates from map parts,
rather than stores, the memory upon seeing an object for the second
(another time after the first time) time.
The first seeing performs an elimination of wrong patterns that do not
fit the object -- so his theory is very much like a Popperian
right-to-left logical transmissibility system (falsification of a
conjecture leaves the "corroborated" (you might say) pattern of neuron
maps.
The unification (or a possible binding problem solution) is theorized
to be an interaction of partial representations via reentry. Edelman
might not say that exactly, but I have also read Minsky and Arbib and
Clark and that is what I say Edelman's research and theories imply.
Each specialized neuronal group is a map of part of the object's
patter (size, shape, color, edges, etc.).
>think of how chess grandmasters "think" or how we solve mathematical
>problems. And my use of the word "pattern" has nothing to do with
>the usual usage.
I think you should have explained that, hence my reaction was
initially one of misunderstanding (sorry).
>It aims more at the way a biological system may see,
>remember or transform regular structures. This should be pretty near
>what you call "subsymbolic".
It is not subsymbolic, but more on that below. Still, Edleman says
many simultaneous "patterns" (in the normal sense of the word, but
split into neuronal sheets specialized for each part of the problem
task) of a object are then transformed into partial representations
(ala Arbib) in the neurons. Societies are formed (ala Minsky) by
these neuronal "pattern" maps when they interconnect to form a unified
representation of the object as a whole.
. . .
>Are we in a Popper discussion?
Sure, why not? Popper wrote one of the most interesting books (with
Sir John C. Eccles) I've read on the subject. It is dated now, but
still interesting (1977/1983) The Self and Its Brain: An Argument for
Interaction. It is an anti-eliminative reductionist philosophy,
similar to property or substance dualism, except it is a pluralism of
three world, not two.
. . .
>Could you explain why the term "pattern" (in the vage form I use)
>is such a threat to your thinking?
Because pattern is so often misused as a justification for "language
first" philosophers. They use pattern as leverage to say that we
cannot *think* (of a pattern) without a descriptive language, words,
word-concepts (e.g., Dennett's (1983) Intentional Systems in Cognitive
Ethology: the 'panglossian paradigm' Defended, Behavioral and Brain
Sciences, 6, pp 343-90, esp. p. 384 ).
The problem is they are not using pattern in the same way I ( and you
are using still a different understanding from them and me)
understanding of the word. The understanding of pattern by them is
different, in that, pattern for them, can only be "thought" via
linguistics, even while in the brain's neurons -- and I do not agree
with that. So, I am touchy about the word. I am not so touchy about
the way you use it, because I was a (am a) programmer.
>P.S. Please also note the work of Christopher Alexander who bases
>his whole theories on his interpretation of the word "pattern"
>(he directly connects "pattern" and "life" although I haven't read
>his final books about "The Order of Nature" published this year).
Yes, I like architecture too and I am also an artist (my father is a
professional artist). I find Alexander's application to object
oriented programming similar to the way I programmed in machine code
by using "canned" sections of code to do small (problem solutions in
context) tasks, then building the program was far easier and more
natural, like bricks of a house and the programs had a higher order of
organization -- but unfortunately I don't think they were subsymbolic
in the way I am looking at the partial representations. But they
*are* close to that in several ways. The difference is sequential vs.
parallel programming fro one thing.
There is a problem with merely looking at Alexander's pattern language
as "A pattern is a solution to a problem in a context," (which was a
very poor statement made by Alexander himself) because that cuts out
several important ideas Alexander's pattern language should (IMO) be
composed of.
For instance, I think Alexander attempts to look past (look over the
top of) syntax and goes straight to semantics (which is very good to
try and do), but he uses a higher-order of organization than symbolic
manipulation, which is also good, . . . but not at the subsymbolic
level (lower level) than I am interested in.
I will give a generalized example of a problem situation (as Popper
would say) that Alexander's "A pattern is a solution to a problem in
a context" does not cover: Problem: 1) I have a reoccurring event
that is not normally a problem, but 2) that is "made" into a problem
by a non-reoccurring-event context, and 3) the solution is a one-time,
one-shot solution, because the context will never happen again and is
not applicable to the next problem. The solution *will,* say, solve
that unusual problem, but only in a non-reoccurring context and it is
not an "Alexander pattern" (IMO).
Contrast that to what an Alexander pattern *should* be (IMO) for AIs:
The bio-based AI must 1) detect a situation more than one time, 2)
the situation (or a discoverable, mappable, variation of the
situation) must be applicable to the next problem, and 3) the
solution must have a unified or "bound" reference to a specific course
of action. This *would* be an Alexander pattern. But it is a
high-level pattern, one that is a solution to the so-called "Binding
Problem," yet, not a subsymbolic explanation of how consciousness
works to get that Alexander pattern in a biological lifeform.
Have you read the troubling debates on the "pattern-discussion mailing
list"? I have heard about them second hand, but as I see it: An
Alexander pattern must contain 1) the problem plus 2) the context,
plus 2) the solution, plus 4) the high-level structure of recurrence
of the whole situation problem and solution, plus 5) high-level of a
way of "teaching" to the organism, of how to solve similar problems in
similar context situations (very much like Popper), plus 6) perhaps
even "high-level" way of syntactical naming of the whole (semantic)
pattern, problem situation plus solution, (like "encoding" the problem
situation).
Is not that really what Alexander is trying to get at?
-- Davin C. Enigl
Helmut Leitner
I think you are right. Alexander seems to be interested in reusable
and actually reused patterns alone in an empirical way.
> Contrast that to what an Alexander pattern *should* be (IMO) for AIs:
> The bio-based AI must 1) detect a situation more than one time, 2)
> the situation (or a discoverable, mappable, variation of the
> situation) must be applicable to the next problem, and 3) the
> solution must have a unified or "bound" reference to a specific course
> of action. This *would* be an Alexander pattern. But it is a
> high-level pattern, one that is a solution to the so-called "Binding
> Problem," yet, not a subsymbolic explanation of how consciousness
> works to get that Alexander pattern in a biological lifeform.
Again I agree with you. But it depends on the level on which you decide
to a apply this "pattern view". When he looks at "architectural patterns"
he need not think about "building material patterns" although one could
change the viewpoint anytime.
> Have you read the troubling debates on the "pattern-discussion mailing
> list"? I have heard about them second hand, but as I see it: An
> Alexander pattern must contain 1) the problem plus 2) the context,
> plus 2) the solution, plus 4) the high-level structure of recurrence
> of the whole situation problem and solution, plus 5) high-level of a
> way of "teaching" to the organism, of how to solve similar problems in
> similar context situations (very much like Popper), plus 6) perhaps
> even "high-level" way of syntactical naming of the whole (semantic)
> pattern, problem situation plus solution, (like "encoding" the problem
> situation).
>
> Is not that really what Alexander is trying to get at?
I don't know. I read a lot of the discussions in Ward Cunningham's
PortlandPatternRepository (bettern known as WikiWikiWeb) but I was
too late to take part. ( <http://c2.com/cgi/wiki?FrontPage> )
Basically the strict formalism is less about patterns and what
patterns are, but about what information should be gathered and
described if one thinks to have found a pattern. So it is about
how to cooperate and agree on an empirical level.
I think that Alexeander's definition is incomplete in one important
way (although I don't know the consequences). The definition should
at least read: "A pattern is a solution to a problem in a context
that takes the form of an object". This rules out all types of
strategies (what to do when...) that are often suggested as
patterns.
I think that Alexandrian patterns could play a role (though perhaps
not important) in understanding the human brain. Is is probable that
a lot of medium level brain functions are reused for different
purposes until at some stage the complex human consciousness comes
into existance (but this is really an area where I know nothing about).
I assume you can read German, so maybe the work of Dieter Dörner
(his PSI-project and his book "Bauplan einer Seele") is of interest
to you. He goes a long way to build neuronal explanations for
human abilities and emotions (and to program bots to reproduce human
decision processes and reasoning).
Anyway I would like to invite you into the wiki world. Especially
into the German DseWiki (Deutsches Software Entwickler Wiki) which
tries to build a high level collaborative (open and uncommercial)
community for open knowledge in the area of software development,
see
<http://www.wikiservice.at/dse/wiki.cgi?RecentChanges>
<http://www.wikiservice.at/dse/wiki.cgi?OffenesWissen>
<http://www.wikiservice.at/dse/wiki.cgi?StartSeite>
<http://www.wikiservice.at/dse/wiki.cgi?KarlPopper>
to name just a few starting points.
George
In other words patterns in computers treat data in bits and bytes,
integers, longs, floats, doubles, long doubles and can be extended
to custom defined types held in structures and objects. Patterns are
a highly general concept.
> At the next further down level: These novel N-power set -memories can
> be acted upon by selection as in evolution (say biological evolution).
> However, in the biological brain, this apparently happens at a
> subsymbolic level, so we are *not* memorizing completed patterns.
> This is far different that (most) current strong AI philosophies -- an
> exception is a reconstruction of Minsky and Arbib, made by Clark (in
> his Mindware: an introduction to the Philosophy of Cognitive Science,
> p. 33). This is however, common in weak AI philosophies (Searle).
>
> So, computer simulation of a biological brain system is *not*
> currently being done correctly (IMO), because the computer AIs are
> only acting at the symbolic level and not the subsymbolic level.
> Patter is being *over* emphasized (symbol is syntax, not semantics)
> and in some cases pattern is being incorrectly taken as the end point
> of AI simulation.
>
> [Top-Down, AI has not worked very well. That is why I do my research
> in the Bottom-Up AI following biological evolution's history as far as
> we conjecture it as possible. Of course, being a microbiologist that
> is what I know best anyway.]
It starts with neurons reacting to recognized objects and triggering
appropriate events that are known to be applicable from past cases
or are justified by current reasoning.
> >> Recently I stumbled over an linguistic classic "The Meaning of Meaning" (about 1930)
> >> by Charles K. Ogden and he analyzes (a long time before AI) how tightly integrated
> >> the recognition of symbols (patterns), our way of thinking and our perception of
> >> the reality really are.
>
> Ogden and Richards was ahead of its time, but I think now, that they
> found only the surface without explanation of the deeper levels (e.g.
> the "emotive" term what we are "really" doing and what we "think" we
> are doing). Popper pointed out the problems with *meaning* and Popper
> instead, introduced explanation without definitions of meaning.
> Popper did this because he considered definitions and meaning infinite
> regressions and therefore, meanings could not important.
Meanings are a product of descriptions. A person reaches for an
apple because he intends to eat it, and meaning in this case is
both in the intention and the description of the corresponding
actions.
George
George
> doesn't turn up in your list.
Patterns are too generic concepts that tend to appeal more to
intuition than something that can be broadly defined, as the
case with polymorphism.
George
Helmut Leitner
In solving dp problems over more than 20 years I never found that
a concept was too abstract or too generic. Usually it is the other
way round: people are too immersed in their specific problems and
implementations to see the simple and general solutions.
And yes, I do appeal to intuition.
George
> George wrote:
> >
> > > I hardly know anything about AI but I wonder why the word "pattern"
> > > doesn't turn up in your list.
> >
> > Patterns are too generic concepts that tend to appeal more to
> > intuition than something that can be broadly defined, as the
> > case with polymorphism.
>
> In solving dp problems over more than 20 years I never found that
> a concept was too abstract or too generic. Usually it is the other
> way round: people are too immersed in their specific problems and
> implementations to see the simple and general solutions.
>
> And yes, I do appeal to intuition.
Ok, we have computers and can define structures and pattern
maps (through designing generalized classes), but can you simply
build a class that dynamically generates just the right set of
needed classes and objects, and is called CGeneralIntelligence()?
What does it need to do? Map sensory data, capable of learning
anything new and formulate intelligent actions. Internally map sensory
data as efficiently as possible for recognition and reuse (recall), and
associate all content to the measure of usefulness, balance action
priorities for the formulation of action sequences (also for attention
management), etc., I believe exploring list above covers the main
areas of consideration.
George
Helmut Leitner
George wrote:
>
> Helmut Leitner wrote:
>
> > George wrote:
> > >
> > > > I hardly know anything about AI but I wonder why the word "pattern"
> > > > doesn't turn up in your list.
> > >
> > > Patterns are too generic concepts that tend to appeal more to
> > > intuition than something that can be broadly defined, as the
> > > case with polymorphism.
> >
> > In solving dp problems over more than 20 years I never found that
> > a concept was too abstract or too generic. Usually it is the other
> > way round: people are too immersed in their specific problems and
> > implementations to see the simple and general solutions.
> >
> > And yes, I do appeal to intuition.
>
> Ok, we have computers and can define structures and pattern
> maps (through designing generalized classes), but can you simply
> build a class that dynamically generates just the right set of
> needed classes and objects, and is called CGeneralIntelligence()?
I can't, but if I would really go after AI think I would start that way and
keep as many parts as possible within this system of general intelligence.
E.g. I think that certain representations of "space" could be
located there. Why is a game board usually a specialized array
or data structure (faster? easier ot program? does that really
matter? ...). Why isn't it represented by a generalized pattern?
> What does it need to do? Map sensory data, capable of learning
> anything new and formulate intelligent actions. Internally map sensory
> data as efficiently as possible for recognition and reuse (recall), and
> associate all content to the measure of usefulness, balance action
> priorities for the formulation of action sequences (also for attention
> management), etc., I believe exploring list above covers the main
> areas of consideration.
I fully agree. You gave a valid directory of the specific details.
I tried to supply a unifying view to glue them together.
george
> George wrote:
> >
> > Helmut Leitner wrote:
> >
> > > George wrote:
> > > >
> > > > > I hardly know anything about AI but I wonder why the word "pattern"
> > > > > doesn't turn up in your list.
> > > >
> > > > Patterns are too generic concepts that tend to appeal more to
> > > > intuition than something that can be broadly defined, as the
> > > > case with polymorphism.
> > >
> > > In solving dp problems over more than 20 years I never found that
> > > a concept was too abstract or too generic. Usually it is the other
> > > way round: people are too immersed in their specific problems and
> > > implementations to see the simple and general solutions.
> > >
> > > And yes, I do appeal to intuition.
> >
> > Ok, we have computers and can define structures and pattern
> > maps (through designing generalized classes), but can you simply
> > build a class that dynamically generates just the right set of
> > needed classes and objects, and is called CGeneralIntelligence()?
>
> I can't, but if I would really go after AI think I would start that way and
> keep as many parts as possible within this system of general intelligence.
>
> E.g. I think that certain representations of "space" could be
> located there. Why is a game board usually a specialized array
> or data structure (faster? easier ot program? does that really
> matter? ...). Why isn't it represented by a generalized pattern?
While I think patterns are very general concepts, knowledge
represented by software objects needs to be applied
specifically to specific problems, if that is what you mean.
George
Helmut Leitner
...that is what I doubt.
George
I think knowledge needs to be represented specifically, meaning, that
if you need to bring a solution to a problem consisting of the selection
of many possible actions, you can only solve it if you have specific
knowledge on how to approach and solve the problem and why
exactly so. Although humans solve many such complex problems
intuitively where we have built up the necessary routines and habits
allowing us to approach the real-world problems, where often the
memories of the original reasoning and learning curve may fade
away while the successful habit may remain "implicit".
A computer program cannot be generated without the programmer's
specific knowledge on how to convert a real-world problem into the
computer. Another well known fact is that knowledge is acquired
based on the necessity to relate to it, where relating is a logical process,
and logic is based on "consistency", in other words on being specific
(be it implicitly or explicitly).
George
Helmut Leitner
This may be true, but it depends on the definition of the problem.
IMO as long as you "solve specific problems by specific knowledge"
you will never arrive at AI. That's just plain conventional programming.
Rick Craik
news:3D06E831...@chello.at...
> George wrote:
> >
[snip]
> > Another well known fact is that knowledge is acquired [...]
>
> This may be true, but it depends on the definition of the problem.
>
> IMO as long as you "solve specific problems by specific knowledge"
> you will never arrive at AI. That's just plain conventional programming.
>
I would think it should still be called an AI, where it acquired the
knowledge artificially. If we construct an AI that acquires knowledge
more naturally, I would still think that it should be called an AI.
In the first case the knowledge is artificial, in the second case the
intelligence system is artificial.
My definition of "artificial" means man-made.
My definition of "intelligence" means an ability to
acquire and apply knowledge.
My definition of "knowledge" is irrelevant here.
We may use "human level knowledge" if we wish to define
"true AI" as such. In any case, the intelligence (ability of a system)
is always real, but the source is artificial as in artificial light, and not
as in artificial flower.
I think I understand your meaning though. I also have a problem
with: What is "not GOFAI"? At least a "Good Old Fashioned AI" is
still an AI.
Regards,
Rick
Guido
I've not read all the stuff of this NG post but some ideas I developped are
worth to inject here:
- neuronal networks will not work for A.I. They are too natural and not
enough artificial.
- patterns appear in neuronal nets and elsewhere too, but meaningful
patterns out of our small planet history is IMO not what we need. We need a
different kind of intelligence to face the emerging challenges.
- in very complex situations triggering all the appropriate events will
result in nothing more than a deep crash of the system with a poor,
unpredictable and probably useless result.
- more relevant results emerge from simulations; thinking, in my humble AI,
is just a complex simulation of a given environment - our world with all the
stuff in the books, the databases and in the news (to learn new stuff)
back to neuronal nets:
How to tell them to run a simulation of the world?
Do these nets (our brains) understand language or are only events triggered
when patterns occur?
And my favorite in AI
Once we have created a working AI with almost unlimited (planet size)
resources... what will we do of it?
Guido
"George" <george...@hotmail.com> schrieb im Newsbeitrag
news:3CFFA2ED...@hotmail.com...
George
>
>
>> > > Another well known fact is that knowledge is acquired [...]
>> >
>> > This may be true, but it depends on the definition of the problem.
>> >
>> > IMO as long as you "solve specific problems by specific knowledge"
>> > you will never arrive at AI. That's just plain conventional programming.
>> >
>
> Helmut,
>
> I did not have access to your latest reply to me, so I am replying to you
> through Rick's post which I had access to.
>
> I personally prefer sticking to the object oriented programming world,
> why, perhaps because that is what I am used to in terms of high-level
> programming mentality and comfort. My original idea began 12 years
> ago or so was the recognition that programming can become far more
> powerful if a program can be written that writes programs. I am
> personally not fund of neural networks and pattern directed thinking
> as they are assembly level to me and not intuitive in terms of high-level
> understanding of problems at large. I enjoy (certainly used to enjoy)
> working with object oriented programming and design issues, distributed
> objects, etc.
>
> To write a program that writes programs, some manual containing the
> project specifics needs to be maintained and communicated to the
> system efficiently, and the source code is then expected to be generated
> instead of programmed specifically by a programmer. I am basing all this
> idea on the realization that it takes me an hour to study and understand
> a project, and I can tell usually if it can or cannot be implemented within
> a certain time frame. Typical commercial software projects take a year
> or more at least to implement but may take as little as an hour to
> comprehend. I am focusing on the idea that a computer could possibly
> perform this implementation work much faster in the future than
> programmers could, given computers can work with far greater
> memory and attention span.
>
> Having looked deeper into this subject in the past 2 years in this NG
> I found out how difficult such an idea is to implement in the real
> world. I am still at the same level basically that here, a C++ programming
> environment, now go and write AI.
>
> If a new specifics are required, the program-generating program needs
> to be modified to allow supporting them. The program generating
> program needs to focus on issues of general familiarity with project
> development and produce the specific programs as close to satisfying
> expectations as possible.
>
> As for another thought on the specific/generic example, a stupid
> thought just came to my mind. Consider this example: "Wednesday
> was a long time ago".
>
> Some people might agree with this statement if they had a rough,
> eventful or painful week. Usually last Wednesday is not considered
> as something that happened a long time ago. The program would
> need to know where exactly that statement is applicable.
> Without specific applications there would be no meaningful
> communication possible. This sort of logic in itself is too generic,
> maybe this is the kind of stuff that is needed for constant head
> hammering if AI is to be explored deeper.
>
> IMO AI needs to be high-level language and communication based,
> and it is expected to be as educated as possible in it's responsiveness
> toward human requests.
>
> George
Sorry, the above post made it after all...
Consider this: A function is something specific as it has a name and
a specific utility. A function can unite many sub-functions into one
"shortcut", or into sub-functions into one distinct object or entity.
The only difference between AI and conventional programs is that
the AI agent has abilities to decide on it''s own which functions
execute, when to execute them and keeping track for what purpose
they are executed. As an agent navigates, a task or plan is constructed,
and later (or subsequently) executed. A plan/task should be
constructed based on understanding a problem at hand, and based
on the ability to distinguish between correct and incorrect solutions.
Define understanding: Understanding can be implicit or explicit
as well perhaps. Explicit is tied to the ability to carry out logical
reasoning through proper relation management (i.e., such and such
tasks are accomplished to satisfy such and such needs, or such and
such states are present to serve such and such purposes, something
along that line). I think, an AI agent is about ability to navigate while
detecting problematic outcomes and successful solutions in the
process. That is what it is all about in general I believe. Just a step up
from Rick's definition perhaps. And of course there is always so
much more that needs to be considered.
George
George
Here is another one: How do you coordinate many agents to work
together in a multi-agent system? (The human brain might have
zero connection to this concept as the brain is mainly massively
response/association based, and multiagent modeling is just one
thing we can learn to do, IMO.)
George
Iscando
---------
Hello friends:
I am new to sharing my information here.
I think that just a small bit of information is needed here.
There is 2 types of programming.
#1 procedural programming, as supported by standard programming languages
like C or Basic. The programming is a step by step procedure where the
problem is solved exactly as as it is programmed by the programmer.
#2 declarative programming, as supported by languages like Prolog.
The programming is the input of facts and rules that describe the
problem. The problem is solved by finding all the possable solutions
to rules that can be proved as a result of the facts contained in the
knowledge base. In this case the programmer has no control over the
procedure that the computer uses to solve the problem. The facts and
rules and the order that they are implimented determine the procedure
that the machine uses to solve the problem. In this case the procedure
is not important as the solution is determined completely by the facts and
rules.
A good AI language must be able to do both kinds of programming. The tools
that the program uses are normally procedures that access the machines
resources. Then the declarative programming handles the real world facts
and rules. The rules can also indicate a need to access the system
resources.
About a program that writes programs. In Prolog that is a simple thing as
the language is also a database language. In the database you might have a
predicate that looks like this EXAMPLE DATABASE PREDICATE:
xprogram(Name,Sequence,Command) Where the Name is the program name. The
Sequence is the Sequence number of the command that is in the last
argument Command. Then the programmer generated prolog predicate liKe.
EXAMPLE PROLOG PREDICATE:
globaltool(Number,Command) if
system("dir >look.txt"),file_str("look.txt",LOOK),
this command issues a command to the system to put the result of the dos
dir command into the file look.txt. Then it puts the contents of the file
look.txt into the variable LOOK.
This is an example of a generic command the can be called by a database
program. These database programs can be written by the system or run by
the system during runtime. This is an example of an imbedded interperter.
The best of all worlds is a system like Visual Prolog where you can create
a tool in C++ and compile it to an object and then put it into the prolog
library so that it can be called as part of the existing system.
There is an example of this kind of programming at iscando.com
in the form of a communication learning machine / expert system shell.
Hope this helps.
Writing in prolog sence Turbo Prolog 1986 Iscando
Helmut Leitner
George, I was on the road for a few days and I'll try to answer
all your postings here. But again remember that I'm not an AI expert
and that my background is pragmatic programming and philosophical only.
With respect to programming languages I share most of your opinions.
I also prefer the OO paradigm, although I think that it is sometimes
too restrictive. I also think that any programming (not just AI) is
about building new languages (not just using the programming language
at hand). Typically this is a difficult art, that even most advanced
programmers do not master. But here we are not yet talking about AI.
It is also clear that powerful programming languages help any
programming (not just AI). So once maybe LISP was a language of
choice for AI, but now this seems to be more a tradition and the
effect of existing codebase.
The ability of "a program creating programs" seems clearly related
to AI, but I do not believe it is. Recently I looked around and even
in LISP I found only rudimentary abilities and coding samples.
Even apart from the technical basics I do not think that true
program generators are feasible for a long time, because they
need very high-level AI.
Programs are not completely described by their specifications. They
are built on a lot of assumptions and back-ground-knowledge that
programmers have about the user, his working environment and his
requirements. So AI for this task would have to reproduce very much
of the human understanding of the whole world. IMO this is like
Mount everest: you can look there, but you can't hope to reach
its peak in one jump. Even if we want to go step by step, it may
be better to target smaller montains nearby first (and not book
the flight to Nepal).
I also agree that to model the brain some kind of multi-agent
system must be built. Especially if we want to model the brains
real time abilities. But again I am not sure that this is the
right way to go. NASA lost centuries in the exploration of space
because they targeted to set foot on the moon instead of going
step by step.
George
I truly appreciate your comments Helmut. You put it all in proper
perspective, I have no comments to add, perhaps that you deserve
a kit kat or something. Either way, finding AI remains but a game,
a fantasy. Perhaps I should stick to writing virtual reality games
where such fantasies can become reality. Who knows???
George
--
Game over. Perhaps.
Helmut Leitner
George wrote:
> I truly appreciate your comments Helmut. You put it all in proper
> perspective, I have no comments to add, perhaps that you deserve
> a kit kat or something.
I like virtual kit kats. :-)
> Either way, finding AI remains but a game,
> a fantasy. Perhaps I should stick to writing virtual reality games
> where such fantasies can become reality. Who knows???
If you start a project, I'd like to contribute.
cu.