Basic Aspects of AI
Bug
We are preparing a subject for university... we'd like to teach an
introductory course on AI and we are doing the previous research
before deciding the contents that give a better view of this exciting
topic.
It would be really helpful and a lot more interesting if you could
give me your opinion on which are the basic areas to cover... or, in
other words, what would you consider essential for the best
understanding of the subject.
I have posted this message in different groups. Basically because I
think not everyone may know about everything, but maybe about some
aspects of AI yes. Really interested also!!
So, this is my query. Giving a good understanding of AI and spreading
its contents is basic for getting more minds involved in it. Thank you
very much in advance.
Arthur T. Murray
> we are doing the previous research before deciding the
> contents that give a better view of this exciting topic.
http://www.amazon.com/exec/obidos/ASIN/0595654371/ -- "AI For You"
(AI4U) is an AI textbook that you might consider using as an AI Lab
supplement to pre-Singularity textbooks of artificial intelligence.
> It would be really helpful and a lot more interesting if you could
> give me your opinion on which are the basic areas to cover... or,
> in other words, what would you consider essential for the best
> understanding of the subject.
http://mentifex.virtualentity.com/acm.html shows the essential DIY AI
steps.
http://mentifex.virtualentity.com/lisp.html is a Lisp AI Weblog for Seed
AI.
> I have posted this message in different groups.
> Basically because I think not everyone may know about everything,
> but maybe about some aspects of AI yes. Really interested also!!
http://www.kurzweilai.net/mindx/profile.php?id=26 - Mind-eXchange
is where readers share ideas and support for Open Source AI projects.
Christopher C. Stacy
[the same weird "AI" messages in all kinds of newsgroups daily]
I've just got to ask this: are you, in fact, a program?
Arthur T. Murray
>
>>>>>> On 5 Sep 2003 07:58:20 -0800, Arthur T Murray ("Arthur") writes:
> [the same weird "AI" messages in all kinds of newsgroups daily]
>
> I've just got to ask this: are you, in fact, a program?
http://www.seedai.e-mind.org/lisp.html -- for Seed AI in Lisp.
Homo sum; nil humani mihi alienum puto. (I am in fact a human being.)
http://dev.null.org/psychoceramics/archives/1998.05/msg00018.html
The mentifeX Files for SURFACE TENSION Column by Linton Robinson
For cyberfreeks, being around Seattle in the pre-Y2K era
is a ground zero set: like being into acid rock in the sixties
Bay Area, or steam technology in Victorian England. Or a control
addict in the thirties Berlin, I suppose. Whatever else, it must
have been exciting. And for that matter, we haven't seen how the
computer thing turns out yet. Our futures are being formatted by
people who didn't figure out the calendar was going to roll over
in just forty years. And, really, the difference between
"accelerated obsolescence" and "final solution" is kind of hazy.
But, in a way, it's hard to see from the street that Seattle
homeboys are big kahunas in the cyber-universe. Maybe you didn't hang
around with Cobain and Vetter at Murphy's, but at least you could go
see them play. Where are the local websites that push the envelopes
of the real edgy stuff: virtual, jack-in, artificial intelligence?
There's one: and everybody all over the world interested in AI or
memetics or robotics knows about it. It's just that nobody knows
what the hell it's all about. It's the Mentifex website on the
Seattle Community Network ( http://www.scn.org/~mentifex/ ) -- home
lair of the ubiquitous Mentifex Diagram. It's on all the big forums
and enshrined on the major CD-ROM's that are the basic library of AI.
It looks like a flow-chart full of alphanumeric yada yantra:
it claims to be the design of the mind of the future. Or at least the
language and grammar. To AI fiends, it's all the same thing. Even you,
your own sacred self, are in essence merely a language or grammar,
a standing pattern, a code, a mantra, a hierarchy of representations,
a program of programs, the mother of all boards. Call it what you will;
structure is the only difference between flesh and Spam.
But it's not figures, and certainly not facts, that have pinned
Mentifex up on the world syntelligence map--it's rumors. Which,
don't knock. What would J.D. Salinger be without obscurity and rumors?
A one-book S.E. Hinton, that's what. The big rumble is that Mentifex
is not "about" AI, is not "author of" or "theory about" AI, but actually
IS an AI. It's also widely rumored that more than one AI haunts the
Internet, and that not all of them are benign. A deeper rumor level
says that one of them has been programmed to post in Esperanto
(wouldn't an intelligence prefer an artificial language?) We can
trust the corresponding authorities to get to the bottom of it.
But Mentifex has something other rumors merely envy--an address.
The Area 51 of AI paranoia is the Mentifex website.
Another theory is that Mentifex is a person (or WAS) who created
an AI programmed to post and respond to mail. Also perhaps to crawl the
Web, searching for any reference to itself or areas that sound "right" for
it, then generating posts and ripostes. The only identity the Mentifex
site traces to is the questionable name "Arthur T. Murray" (same name as
the
dance studio that once advertised "anyone will want to be your partner")
or the equally suggestive initials ATM.
Making the search more difficult is the general confusion about
Mentifex and his (her?) (its?) proper place in the AI pantheon. A lot of
people think he is a nutcake spamming the galaxy with an incomprehensible
diagram full of weird diacriticals and signifying nothing. Others, and
they include a lot of the very heavy hitters in AI, robotics, and
memetics, take Mentifex seriously and express continuing interest in his
codes, whether they are or aren't a self portrait. Perhaps even more
tellingly, people like William Gibson "think they've heard of it".
Hovering around the peripheral supraliminals of a cat like the
Neuromancer,
it doesn't matter if you're a person or a program--you're somebody.
The only major clue is the inscrutable diagram, which is the tip
of a Titanic-class iceberg. It's a catchy little hook like E=MC2 that
everybody mentions, but nobody actually mathematically understands. The
glyphs are backed up by line after line of code, which gets dissected by
heavyweight line-by-liners on forums devoted to "Forth" language and such.
But esoteric symbols never do much to start or dispel rumors.
And unlike many conspiracy/hoax rumors that swarm with "proofs" and
arguments, the "Synthifex" theories pretty much rest on allegations of
hearsay (the AI grapevine is small, select and highstrung, vibrating like
worldwide cobweb to the proper intrusion). And on a simple experiment:
"Have you READ this guy?"
Which gets to the only known criterion for whether an intelligence
is real or artificial, the Turing test. If you're cyber-chatting with
either a person of machine and can't tell the difference, then what's
the difference? Of course, Turing set his tests up in a laboratory:
the Internet has turned that experimental design into an environment.
It's like a milieu invented for AI's before they even existed. Creating
a perfect world for a being, then leaving it unpopulated, would be the
kind of vacuum that grosses out Mother Nature--theological malpractice.
So it gets back to language. And there is a certain quality to
Mentifex humor and exposition that is, well, alien. Like a foreigner
trying to "pass". Though the oddness isn't entirely inconsistent with
the identity sketched out by certain local lectroslackers who claim to
have met Mentifex (or possibly Murray)--that he's a seminary bachelor
with a background in classical Greek and Latin who lives in a hermit
burrow and supplements his diet by diving for crawdads in Green Lake.
A sort of urban "UnaBumbler".
Other theorists point out that certain oddball Mentifex speech
ellipses, such as breaking into German or Russian almost at random, and
stringing together clumps of "ten dollar" words in ways even George Bush
couldn't make sense out of, substantiate one detractor who claims to have
traced the Mentifex origins back to a Russian military translation program
developed for disseminating Cold War FAX messages simultaneously to all
Warsaw Bloc nations. Allegedly the Soviet program was called "Mentzifax".
Then again, this certain suave, haywired "Company" type who hangs
at the Honey Bear Bakery with his gorgeous Eurasian girlfriend, Range
Rover, and business card listing voice mail in some mysterious blackbag
enclave on the Microsoft campus might be behind the whole thing, but
declines comment. So. Who'd you rather have controlling the first robot
mind to challenge human intelligence: the Unabomber or Bill Gates? Both
obsolescence specialists, you'll notice.
I vote for aliens, but who really knows where THOSE guys are
coming from? Another rumor, (which you are reading for the first time
right here in VOLTAGE) is that Mentifex essentially invented itself,
spawned by random electronic accidents from some overly clever virus,
evolving into a rudimentary consciousness, then coping with its
environment--lately to include its resident web site and even the ATM
identity which was possibly created to fullfill its own nascent need to
have a creator. User as religion, a la "Tron". Mentifex just COULD be
trouble. Or a messiah, which is about as troubling as an intelligence can
get.
--
http://www.kurzweilai.net/mindx/profile.php?id=26 - Mind-eXchange
Christopher C. Stacy
>>
>>>>>>> On 5 Sep 2003 07:58:20 -0800, Arthur T Murray ("Arthur") writes:
>> [the same weird "AI" messages in all kinds of newsgroups daily]
>>
>> I've just got to ask this: are you, in fact, a program?
>>>>> On 5 Sep 2003 09:42:47 -0800, Arthur T Murray ("Arthur") writes:
>> [a volumenous canned response suggesting that he might be]
Arthur> Another theory is that Mentifex is a person (or WAS) who
Arthur> created an AI programmed to post and respond to mail.
Arthur> Also perhaps to crawl the Web, searching for any reference
Arthur> to itself or areas that sound "right" for it, then generating
Arthur> posts and ripostes.
So, I'll take that as a "Yes" -- "Arthur T Murray" (aka Mentifex)
is a trivial program, similar to ELIZA, designed to spam newsgroups.
Or someone with a lot of time on their hands doing it semi-automated.
Eternal Vigilance
Information interpretation (cognitive... including handling
unknowns/uncertainty)
Randolph M. Jones
> Adaptability (often refered to as learning)
Often mistakenly referred to as learning, in my opinion. It is possible
to be extremely adaptable in the absence of learning.
David Longley
<rjo...@colby.edu> writes
makes sense to look to the literature in learning theory, particularly
the neurobiology of learning (e.g model systems of classical and operant
conditioning in whole, or sub-systems of animals) to see how those
trying to make sense of plasticity approach the problems.
Some useful references and operational definitions were posted in c.a.p
a few months ago.
--
David Longley
Mike
mind. I would group them based on the intention of the proponent:
1) purely medical: they talk about cells, neurons, etc
2) software developing: they want to have a talking encyclopedia able to
read their mind, so that they don't have to explain what they want, and get
the answer; they tend to equate human mind and current computers; seeking
for a better model but loath to use notions outside IT.
3) practitioners: they want to develop their ability to think better; the
problem seems in that we do not use reasoning consciously enough; the
solution is in reflecting on great samples of reasoning and finding
patterns, etc.
4) a lot of other approaches developed based on the proponent's background
Isn't it high time for AI to get its own field of interest, set of notions,
and methodology?
Mike
"Bug" <ophi...@atlas.sk> wrote in message
news:88222545.03090...@posting.google.com...
rick++
> introductory course on AI and we are doing the previous research
> before deciding the contents that give a better view of this exciting
> topic.
MIT has two versions of its A.I. course in its OpenCourseware
selection
at http://ocw.mit.edu/OcwWeb/Electrical-Engineering-and-Computer-Science/6-034Artificial-IntelligenceFall2002/CourseHome/index.htm.
The course presumes a good skills with ObjectLISP taught in course
6.001.
However I think it has narrowed into a core techniques course,
rather then a general synthesis. There are some decent survey
textbooks out there, though a bit dated.
Eternal Vigilance
"Randolph M. Jones" wrote:
What is learning ???? The self-modifying of behavior to effect some
gain....
Adapability ???? in general sounds the same.
Now if learning is defined as being actively instructed -- not just
inherant (programmed) behavior ....
??
David Longley
<wo...@oneeye.com> writes
answer is the "Learning Theory" literature for strictly behavioural
operational definitions, or behavioural neurobiology for operational
definitions in a specific physiological system. When you do this, you'll
come across non-associative plasticity and associative plasticity as
there *are* phenomena which need to be controlled for when one is trying
to confirm that classical conditioning or operant conditioning has in
fact been established.
A lot of folk don't seem to appreciate that talking about "learning",
"memory" etc outside of these contexts (and even possibly within them)
is possibly quite meaningless.
People working in the Experimental Analysis of Behaviour (cf. radical
behaviourism or operant conditioning) are generally uncomfortable with
theories of learning and the notion of "learning" itself, they don't
deny that behaviour is plastic, but they are uncomfortable with the way
in which people glibly talk about it this plasticity.
The operational analysis point really is very strict, and whether we
like this or not, it takes little to see why this must be so. The
alternative tends to get us nowhere, and we can say that in the full
knowledge of decades of research the extent of which most likely covers
most "ideas" which pop into peoples' heads in NGs and beyond. For nearly
30 years or so, psychology *was* learning theory. it literally covered
just about everything - just the words changed e.g.
perception->discrimination learning, and often much more cryptically.
My advice to folk in this NG was to look to the work of some people
researching model systems, such as flexion and tail flick responses in
spinal animals, or the rabbit NMR, or flight in drosophila - which often
graduate to operant analysis. If one wants to work strictly with
behaviour, then the Rescorla-Wagner model and its developments, or the
quite vast technology produced by those in Experimental Analysis of
Behavior.
--
David Longley
Charles Fox
- which would probably form the main two halves of an AI course.
The symbolic half would look mostly at tree-searching algorithms. For
example, how to write a tic-tac-toe or connect-4 program. Languages
like Prolog and Lisp fit in here. (Though you could also write your
tree-searchers with objects in a more modern language.)
The non-symbolic half is more of a mixed bag, and could include things
like neural nets (eg write a character recognition program); genetic
algorithms (try to solve the travelling salesman problem); and more
conventional mathematical optimisation methods (eg matlab optimisation
tools). To be honest I really dont know what the difference between
'AI' and 'optimisation' is these days, if it even exists.
The question of how to represent knowledge can also be part of AI -
though again I couldn't say where this ends and
database/object-oriented design starts.
Finally, I would be dissapoined if I signed up for an AI course and it
didn't have a little bit of philosophy of mind and consciousness at
the end.
Arthur T. Murray
> for an AI course and it didn't have a little bit of
> philosophy of mind and consciousness at the end.
The problem of the obsolete AI textbooks is being dealt with.
http://mentifex.virtualentity.com/cpp.html -- C++ with new AI code;
http://mentifex.virtualentity.com/java.html -- see "Mind.JAVA #001";
http://mentifex.virtualentity.com/lisp.html -- Lisp AI Weblog;
http://mentifex.virtualentity.com/perl.html -- first Perl module;
http://mentifex.virtualentity.com/prolog.html -- Prolog AI Weblog;
http://mentifex.virtualentity.com/python.html -- Python AI Weblog;
http://mentifex.virtualentity.com/ruby.html -- Ruby AI Blog (OO AI);
http://mentifex.virtualentity.com/scheme.html -- "Scheme AI Weblog;
http://mentifex.virtualentity.com/vb.html -- see "Mind.VB #001" link.
Anthony Bucci
> 'optimisation' is these days, if it even exists.
AI would truly be dead if there were no distinction. Folks at MIT seem to
think that way...
But I don't agree. There is, at the very least, a large semantic
difference between optimization and AI. Optimization is about...uh
optimizing stuff. It tends to be mathematical in nature, tends to draw a
lot of insights from the engineering disciplines in which it's used, etc.
AI, by contrast, is more of a guideline to a field. Meaning, we have this
wish to create smart machines, and for now we just don't know how to do
that. However, the dream is there and a lot of people put energy into it.
It's the dream which defines the field. Sure, there is substantial
overlap in terms of algorithms, in terms of mathematics used, in terms of
problems studied. However, where the two fields draw their insights and
drives differs substantially.
> The question of how to represent knowledge can also be part of AI -
> though again I couldn't say where this ends and
> database/object-oriented design starts.
Knowledge representation is one of the pillars of classic AI. It's
a critical part of the field today as well.
Here, there is substantial difference between databases, OO systems, and
AI knowledge representation. AI applications use a variety of wild
datatypes to represent knowledge, whereas databases and OO systems are
substantially more constrained. Advances in AI knowledge representation
can lead to niftier databases.
> Finally, I would be dissapoined if I signed up for an AI course and it
> didn't have a little bit of philosophy of mind and consciousness at
> the end.
Me too. That's the cool stuff. On the other hand, you often sign up for
math classes and get only the technical details, not the cool intuitions
that led to the development of the techniques. It's the same in computer
science and in many fields, for that matter.
One of my favorite discoveries in that respect was when I finally got my
hands on Stephen Kleene's original paper on regular expressions (including
what later came to be called Kleene's theorem). It's in a 1956 volume
called Automata Studies editted by Claude Shannon which included papers by
Minsky and von Neumann, among others. Lots of good stuff in there.
Anyway, Kleene's paper was about the behavior of "nerve nets," which today
we'd think of as particularly simple neural networks. He worked out how
such networks would behave and showed how they were limited in what they
could do. He developed a syntax, regular expressions, for expressing the
behavior. That syntax is basically all that survives now.
So, how many NLP, information retrieval, or theoretical computer science
courses nowadays tell you about these early connections between neural
networks, finite state automata, and regular expressions? None that I
took did.
Anthony
Randolph M. Jones
>
> "Randolph M. Jones" wrote:
>
>
>>Eternal Vigilance wrote:
>>
>>>Adaptability (often refered to as learning)
>>
>>Often mistakenly referred to as learning, in my opinion. It is possible
>>to be extremely adaptable in the absence of learning.
>
>
>
>
> What is learning ???? The self-modifying of behavior to effect some
> gain....
>
> Adapability ???? in general sounds the same.
I'm arguing that only a subset of "self-modifying of behavior" counts as
learning. If I program a robot so that it always heads towards the
brightest light source it detects, it will "adapt" and "self-modify" its
behavior as the environment changes (the light moves around). That's a
very limited form of adaptability, but I certainly wouldn't call it
learning.
JGCasey
"Randolph M. Jones" <rjo...@colby.edu> wrote in message
news:3F65FFB8...@colby.edu...
Learning I think implies an internal change that improves
the systems ability to achieve some goal.
Adaptability is relearning in a new situation where the
old behaviours do not return the best results.
John Casey
>
Randolph M. Jones
> "Randolph M. Jones" <rjo...@colby.edu> wrote in message
> news:3F65FFB8...@colby.edu...
>
>>Eternal Vigilance wrote:
>>
>>>"Randolph M. Jones" wrote:
>>>
>>>
>>>
>>>>Eternal Vigilance wrote:
>>>>
>>>>
>>>>>Adaptability (often refered to as learning)
>>>>
>>>>Often mistakenly referred to as learning, in my opinion. It is possible
>>>>to be extremely adaptable in the absence of learning.
>>>
>>>
>>>
>>>
>>>What is learning ???? The self-modifying of behavior to effect some
>>>gain....
>>>
>>>Adapability ???? in general sounds the same.
>>
>>I'm arguing that only a subset of "self-modifying of behavior" counts as
>>learning. If I program a robot so that it always heads towards the
>>brightest light source it detects, it will "adapt" and "self-modify" its
>>behavior as the environment changes (the light moves around). That's a
>>very limited form of adaptability, but I certainly wouldn't call it
>>learning.
>
>
> Learning I think implies an internal change that improves
> the systems ability to achieve some goal.
I'd say that's a pretty reasonable definition of "adaptability". For
me, a definition of "learning" would include a "long-term" modifier,
such as "...implies a long-term internal change...".
David Longley
<kjc...@hotkey.net.au> writes
With all due respect, this sort of speculation is *almost* always
fruitless. For instance, in the above you say that learning "implies an
internal change that improves the systems ability to achieve some goal"
- so what about all of your "bad habits"? What about clinical behaviour
disorders which are "learned behaviours"?
Surely the best way to find out how to use words like "learning" and
"adaptation" is to look to the scientific literature where people have
spent decades working on precisely this. You can then talk productively
about the current state of this human endeavour and be of some real use
to each other.
All you do by doing what you are doing in this thread is sharing your
relative ignorance, and I mean no offence by saying that - we all tend
to do this in one area of another, but when there is an opportunity to
do otherwise, you should take it.
Throughout the 1930s, 40s and 50s, psychology was almost defined as the
study of "learning"..
--
David Longley
Eternal Vigilance
David Longley wrote:
Sorry to disagree, understanding how other people define/understand these
subjects
is more important than trying to pigeonhole them into some
psycho-babel limitations. If you didnt notice, this is a 'computer' newsgroup,
not a 'psycho' group.
> For instance, in the above you say that learning "implies an
> internal change that improves the systems ability to achieve some goal"
> - so what about all of your "bad habits"? What about clinical behaviour
> disorders which are "learned behaviours"?
We cant help old labeling problems from the Head Shrinking industry.
'Bad habits' are a social judgement and may or may not be improvements
depending on context. And 'disorders' is a similar term. You are dumping
evidence that is out of context of the discussion.
>
>
> Surely the best way to find out how to use words like "learning" and
> "adaptation" is to look to the scientific literature where people have
> spent decades working on precisely this. You can then talk productively
> about the current state of this human endeavour and be of some real use
> to each other.
One tiny view of it maybe. Productivness of our discussion is not for you to
say
or define.
Again, try to comprehend, this is a computer oriented newsgroup and we tend
to talk along that context. Human processes may not even apply in many cases.
We are interested in making mechanisms that simulate behavior not just
in observing/diagnosing them.
>
>
> All you do by doing what you are doing in this thread is sharing your
> relative ignorance, and I mean no offence by saying that - we all tend
So you wont take offense when I say that you should be more informative than
judgemental. If you have some links to appropriate information, provide it.
>
> to do this in one area of another, but when there is an opportunity to
> do otherwise, you should take it.
>
> Throughout the 1930s, 40s and 50s, psychology was almost defined as the
> study of "learning"..
And the significance of invoking 50+ year old understanding is???
>
> --
> David Longley
David Longley
<wo...@oneeye.com> writes
>
>
The fact that you have to even ask just makes you look very silly
--
David Longley
David Longley
michaels <d...@oricomtech.com> writes
>David Longley <Da...@longley.demon.co.uk> wrote in message
>news:<feyso$HleXa$Ew...@longley.demon.co.uk>...
>> >> fruitless.
>> >
>> >Sorry to disagree, understanding how other people define/understand these
>> >subjects
>> > is more important than trying to pigeonhole them into some
>> >psycho-babel limitations. If you didnt notice, this is a 'computer'
>> >newsgroup, not a 'psycho' group.
>> >
>
>> >
>> >to talk along that context. Human processes may not even apply in
>> >many cases.
>> >We are interested in making mechanisms that simulate behavior not just
>> >in observing/diagnosing them.
>> >
>
>> >> study of "learning"..
>> >
>> >And the significance of invoking 50+ year old understanding is???
>> >
>> >
>>
>> The fact that you have to even ask just makes you look very silly
>
>
>he's not a behaviorist. And psychology defined again for the benefit
>of the technos. Who cares.
>
>As Vigilance points out, the word psycho doesn't appear in
>comp.ai.philosophy. Who is silly!
You are (as we already know).
It is simply stupid to ignore the relevant research. The post I am
commenting upon shows as poor a grasp of the relevant disciplines as you
do.
--
David Longley
Ray Dillinger
>
> It is simply stupid to ignore the relevant research. The post I am
> commenting upon shows as poor a grasp of the relevant disciplines as you
> do.
It's open to question how relevant it is, at this point.
If we were medical students trying to make a functional model of
a human brain, for medical modeling purposes or to increase our
understanding of human psychology, then studying human brains
would be what we needed to do.
But we're not. The biological model isn't always the only
appropriate model for carrying out a task, and this has proven
particularly true with information-related tasks. We still
don't understand, really, how humans do fundamental operations
like adding two numbers. Oh, sure, we can talk about visual
memory and process memory and integration and motor memory
and so on... but do we really understand how any of that
works? The answer is no. But that didn't stop us from figuring
out how to build silicon circuits that add numbers.
When we wanted to get from place to place really fast, we could
have spent time studying the muscles and bones and circulatory
systems of cheetahs and ocelots and other animals that run
really fast. Would that have helped us build the internal-
combustion motors we're using in our cars? What part of
biological legs rotate around an axle? Our Engineering isn't
up to the amazing capabilities of biological forms, such as
self-repair and self-replication and so on, but within our
limitations, we were able to come up with a pretty good (if
drastically different) solution to the problem of getting from
place to place really fast.
Right now, people pursuing biology-based models of AI have
gotten as far as, maybe, fairly smart insects. The best of
those models may have the sophistication of an ant, or a bee.
That's actually pretty awe-inspiring work, when you fully
realize how complicated ants and bees are. Neurobiology is
relevant to those models; but human psychology, as yet, is
not.
Relatively few people in AI are even thinking yet about building
a "synthetic person" who dreams and desires things and hopes and
doubts and has joys and sorrows and pleasures and sufferings in
addition to language, memory, and knowledge representation. If
we did so it would raise ethical questions regarding such things
as enslavement and whom to charge with murder after a reboot.
Some are, but most of us would be really happy just to get
something capable of language, memory, and knowledge
representation functioning, like a sort of natural-language
operated database. Imitating the fallible human memory is
not a particularly high-priority goal here, nor is providing
our device with the cognitive equipment humans use to enable
them to decieve themselves or the ulterior motives to lie
about stuff, nor, especially, the cognitive power to suffer.
IOW, we want a tool with capabilities we don't have, and we
want it to be reliable in working for us, and we don't want
something that will suffer pain and hardship and angst over
its state of servitude. If we wanted human memory and human
reliability and human personality, we'd just hire somebody.
So this device we're trying to build isn't a human mind. It
might wind up resembling a human mind to about the same degree
that the engine and drivetrain of a Stanley Steamer resemble the
heart and lungs and legs of a cheetah. In fact, I don't know if
any of the engineers who designed the Stanley cars ever studied
cheetahs, or for that matter studied the kinematics of humans
running. They'd probably have judged, correctly, that it was
only marginally relevant to the somewhat different problem they
were trying to solve.
Now, it will please us to use the terms "memory" and "learning"
to describe operations whose effects are analogous to human
memory and learning, whether or not accomplished in anything
resembling the same way. Since this is in the comp.* hierarchy,
we don't expect anyone to be confused by the context.
Bear
JGCasey
SNIP
> >
> >Learning I think implies an internal change that improves
> >the systems ability to achieve some goal.
> >
> >Adaptability is relearning in a new situation where the
> >old behaviours do not return the best results.
> >
> >John Casey
> >
============================================
>
> With all due respect, this sort of speculation is *almost* always
> fruitless. For instance, in the above you say that learning "implies an
> internal change that improves the systems ability to achieve some goal"
> - so what about all of your "bad habits"? What about clinical behaviour
> disorders which are "learned behaviours"?
Point taken, the definition is a bit sloppy. However the definition
doesn't make subjective assumptions about good and bad only
that some "goal" (or reward) is controlling what internal changes
are taking place.
Although maladaptive behaviours are indeed learnt they are learnt
as a result of the person "improving their ability to achieve some
short term goal". This may not be a desirable goal from a social
point of view or from a long term point of view but it is still a goal.
You feel bad, you have a smoke, now you feel better. Feeling
better is the goal and you have improved it.
> Surely the best way to find out how to use words like "learning" and
> "adaptation" is to look to the scientific literature where people have
> spent decades working on precisely this. You can then talk productively
> about the current state of this human endeavour and be of some real use
> to each other.
Unfortunately there are only so many hours in a day and I have
to select my reading to my goals and abilities.
What I have read about learning comes from AI books and some
basic psychological books and books about the brain.They have
been useful in giving me ideas on how to write programs that "learn"
and "adapt". (For example "learn" to recognize a hand written
character and "adapt" to different handwriting).
For me the program is *a* definition of a word.
> All you do by doing what you are doing in this thread is sharing your
> relative ignorance, and I mean no offence by saying that - we all tend
> to do this in one area of another, but when there is an opportunity to
> do otherwise, you should take it.
Well my programs work so I cannot be a complete nong :-)
However I appreciate any practical suggestions on how to reduce
my ignorance in areas that have some direct effect on my goals.
John Casey
David Longley
writes
>David Longley wrote:
>>
>> It is simply stupid to ignore the relevant research. The post I am
>> commenting upon shows as poor a grasp of the relevant disciplines as you
>> do.
>
>It's open to question how relevant it is, at this point.
It is *not* open to question - though it might be open to
"enlightenment"!
A rational person does not ignore a whole science which has been, at one
time or another, almost exclusively dedicated to researching learning
and intelligence. It has nothing to do with whether it is discussed in
the context of a psychology or computer NG, and a moments reflection on
this would make this clear after looking at both what is being discussed
and how it is being discussed in these cross-linked NGs!
--
David Longley
Ray Dillinger
>
> >It's open to question how relevant it is, at this point.
>
> It is *not* open to question - though it might be open to
> "enlightenment"!
>
> A rational person does not ignore a whole science which has been, at one
> time or another, almost exclusively dedicated to researching learning
> and intelligence.
Was it irrational, then, of the designers of the stanley steamer
automobile to ignore the sciences of physiology and anatomy (and
of course the complexities of balance and control of running
animals)? After all, that was where the "secret" of how biological
organisms run fast had been studied. Was it irrational of the
Montgolfier brothers to ignore the aerodynamics of birds?
Between us, I'm rather glad they did. We're *still* trying to
figure out the control mechanisms we need for balance while running,
and we have yet to build a useful passenger-carrying ornithopter.
But cars have been pretty darn useful for at least a century now
and lighter-than-air craft were useful for about a century before
we figured out some flight aerodynamics and how to build airplanes.
Besides, with wheels (or hot-air ballons) you get a less bumpy
ride. :-)
I say that I for one am trying to do something different in kind
than emulate biological intelligence. What I work on is inspired
by biological intelligence, but I don't seek to imitate it.
Future generations will build machines capable of joy and desire
and suffering and lying and madness; I for the moment am trying
to do something much simpler. I'm like the Montgolfiers, trying
to get a hot-air balloon to work. Once there are balloons, there
will be dirigibles. Once there are dirigibles, there will be a
serious study of aerodynamic engineering in terms of the kind of
inflexible, brittle parts we can actually make rather than the
fluid biological forms we have no hope of imitating this century.
Once that study has been done, somebody like the Wright brothers
will come along and relate that mechanistic engineering knowledge
to the aerodynamics of birds and come up with a better flying
machine. And whether the generation after that builds jet passenger
aircraft or manned rockets or birdlike ornithopters, I can't even
see yet.
Bear
David Longley
<kjc...@hotkey.net.au> writes
>
>"David Longley" <Da...@longley.demon.co.uk> wrote in message
>news:VLX+9CL+bxZ$Ew...@longley.demon.co.uk...
>> In article <3f6672db$1...@news.iprimus.com.au>, JGCasey
>> <kjc...@hotkey.net.au> writes
>
> SNIP
>
>> >
>> >Learning I think implies an internal change that improves
>> >the systems ability to achieve some goal.
>> >
>> >Adaptability is relearning in a new situation where the
>> >old behaviours do not return the best results.
>> >
>> >John Casey
>> >
>============================================
>>
>> With all due respect, this sort of speculation is *almost* always
>> fruitless. For instance, in the above you say that learning "implies an
>> internal change that improves the systems ability to achieve some goal"
>> - so what about all of your "bad habits"? What about clinical behaviour
>> disorders which are "learned behaviours"?
>
>Point taken, the definition is a bit sloppy. However the definition
>doesn't make subjective assumptions about good and bad only
>that some "goal" (or reward) is controlling what internal changes
>are taking place.
It's not just a *bit sloppy*.
And what you go on to say is *not* true (and what you say is right there
on the screen). The reason I challenged it was precisely because you
said that it was an "internal change that improves the systems ability
to achieve some goal".
Leaving aside the "internal" bit, the key word was "improves".
>
>Although maladaptive behaviours are indeed learnt they are learnt
>as a result of the person "improving their ability to achieve some
>short term goal". This may not be a desirable goal from a social
>point of view or from a long term point of view but it is still a goal.
>You feel bad, you have a smoke, now you feel better. Feeling
>better is the goal and you have improved it.
>
How on earth do you relate this to phobias and other neurotic disorders?
>
>> Surely the best way to find out how to use words like "learning" and
>> "adaptation" is to look to the scientific literature where people have
>> spent decades working on precisely this. You can then talk productively
>> about the current state of this human endeavour and be of some real use
>> to each other.
>
>Unfortunately there are only so many hours in a day and I have
>to select my reading to my goals and abilities.
That's fair enough - but you don't have to say *anything* and if you do,
and surely you would prefer to have something which might be
fundamentally wrong challenged rather than just ignored?
>
>What I have read about learning comes from AI books and some
>basic psychological books and books about the brain.They have
>been useful in giving me ideas on how to write programs that "learn"
>and "adapt". (For example "learn" to recognize a hand written
>character and "adapt" to different handwriting).
Fair enough - and how people end up producing novel behaviour through
superstitious conditioning or other machinations (Quinean analogical
synthesis) of their experience is a hot topic in itself, but getting
some things right about what learning is in its own right is just as
important.
>
>For me the program is *a* definition of a word.
>
>> All you do by doing what you are doing in this thread is sharing your
>> relative ignorance, and I mean no offence by saying that - we all tend
>> to do this in one area of another, but when there is an opportunity to
>> do otherwise, you should take it.
>
>Well my programs work so I cannot be a complete nong :-)
>
Of course not - and as you weren't being arrogant about it perhaps the
word *ignorance* was a little unfair - unfamiliarity might have been
more apposite. But you get my point, and sometimes, if these points
aren't made forcefully they aren't received at all.
>However I appreciate any practical suggestions on how to reduce
>my ignorance in areas that have some direct effect on my goals.
>
>John Casey
>
Fair enough.
--
David Longley
JGCasey
After I sent the post I did think that perhaps the word "improves"
was an issue. Note the definition does not assume the goal is
reasonable or rational only that certain changes can result in the
system improving its chances of reaching that goal state more often.
Perhaps I should have added the caveat that learning may also
be neutral in relationship to any goals. It may simply change the
potential set of useful behaviours.
The word "improve" is somewhat subjective in that change for
the better depends on what you mean by "better". You may
become a "better" cheat or killer but is that a change for the
"better" according to your morals and ethics.
In the above definition it simply meant better at achieving
some goal.
It is the "internal" mechanisms that ultimately interest me as I would
like to program them into simple learning machines.
> >Although maladaptive behaviours are indeed learnt they are learnt
> >as a result of the person "improving their ability to achieve some
> >short term goal". This may not be a desirable goal from a social
> >point of view or from a long term point of view but it is still a goal.
> >You feel bad, you have a smoke, now you feel better. Feeling
> >better is the goal and you have improved it.
> >
> How on earth do you relate this to phobias and other neurotic disorders?
Your goal is to avoid stress. An irrational fear (phobia) causes stress.
We get closer to our goal, low stress, by learning how to avoid those
situations that cause the stress. These learnt behaviours may not be
rational but they do "improve" the subjects stress level.
Fear of flying is irrational but it certainly improves the phobics chances
of achieving their goal not to die in a plane crash. Their goal is to reduce
their anxiety although their behaviour may actually increase the chances
of dying by taking a car trip instead.
> >> Surely the best way to find out how to use words like "learning" and
> >> "adaptation" is to look to the scientific literature where people have
> >> spent decades working on precisely this. You can then talk productively
> >> about the current state of this human endeavour and be of some real use
> >> to each other.
> >
> >Unfortunately there are only so many hours in a day and I have
> >to select my reading to my goals and abilities.
>
> That's fair enough - but you don't have to say *anything* and if you do,
> and surely you would prefer to have something which might be
> fundamentally wrong challenged rather than just ignored?
You can choose not say anything and people may think you do so
out of ignorance. You may say something and prove yourself ignorant.
But then at least you can be corrected. So many social problems
result from people making false assumptions about other peoples
motivations and never having them corrected because they fail
to verbalize those assumptions or beliefs. But then again some
things are best left unsaid :-)
> >What I have read about learning comes from AI books and some
> >basic psychological books and books about the brain.They have
> >been useful in giving me ideas on how to write programs that "learn"
> >and "adapt". (For example "learn" to recognize a hand written
> >character and "adapt" to different handwriting).
>
> Fair enough - and how people end up producing novel behaviour through
> superstitious conditioning or other machinations (Quinean analogical
> synthesis) of their experience is a hot topic in itself, but getting
> some things right about what learning is in its own right is just as
> important.
Get it wrong and a "learning" program will fail to learn.
> >For me the program is *a* definition of a word.
> >
> >> All you do by doing what you are doing in this thread is sharing your
> >> relative ignorance, and I mean no offence by saying that - we all tend
> >> to do this in one area of another, but when there is an opportunity to
> >> do otherwise, you should take it.
> >
> >Well my programs work so I cannot be a complete nong :-)
> >
> Of course not - and as you weren't being arrogant about it perhaps the
> word *ignorance* was a little unfair - unfamiliarity might have been
> more apposite. But you get my point, and sometimes, if these points
> aren't made forcefully they aren't received at all.
And you can dish out your comments as forcefully as you like.
I will not take offence. But I cannot guarantee I will know enough
in your field of expertise to understand or make use of those
comments.
John Casey
Eray Ozkural exa
> Your goal is to avoid stress. An irrational fear (phobia) causes stress.
> We get closer to our goal, low stress, by learning how to avoid those
> situations that cause the stress. These learnt behaviours may not be
> rational but they do "improve" the subjects stress level.
>
> Fear of flying is irrational but it certainly improves the phobics chances
> of achieving their goal not to die in a plane crash. Their goal is to reduce
> their anxiety although their behaviour may actually increase the chances
> of dying by taking a car trip instead.
This is a very nice analysis :)
David Longley
>>
>> It is simply stupid to ignore the relevant research. The post I am
>> commenting upon shows as poor a grasp of the relevant disciplines as you
>> do.
>
>It's open to question how relevant it is, at this point.
>
>If we were medical students trying to make a functional model of
>a human brain, for medical modeling purposes or to increase our
>understanding of human psychology, then studying human brains
>would be what we needed to do.
This is largely self-deception - and it ends up with people talking all
sorts of metaphysical nonsense about "intelligent" machines having
consciousness and/or other mental states. You can see that happening in
your newsgroups, and denying it is just more self-deception. All of this
happens because you have cavalierly appropriated terms used in folk
psychology for operations in computers. I agree that we don't know how
these terms are used in folk psychology, but that's a problem and it's a
problem which you are importing into Computer Science and AI.
At the end of this post you say that you can just go ahead and "use the
terms "memory" and "learning" to describe operations whose effects are
analogous to human memory and learning, whether or not accomplished in
anything resembling the same way" and that nobody will be confused
because the discussion occurs in "the comp.* hierarchy".
You start by using the terms as metaphors - but you end up being seduced
by those metaphors.
My point had nothing to do with building "synthetic persons" or any of
the above.
>Some are, but most of us would be really happy just to get
>something capable of language, memory, and knowledge
>representation functioning, like a sort of natural-language
>operated database.
And here we see the metaphor and it's mentalistic philosophical baggage
insidiously exerting its muddling effects.
> Imitating the fallible human memory is
>not a particularly high-priority goal here, nor is providing
>our device with the cognitive equipment humans use to enable
>them to decieve themselves or the ulterior motives to lie
>about stuff, nor, especially, the cognitive power to suffer.
I have said nothing about imitating human memory - quite the contrary in
fact.
>
>IOW, we want a tool with capabilities we don't have, and we
>want it to be reliable in working for us, and we don't want
>something that will suffer pain and hardship and angst over
>its state of servitude. If we wanted human memory and human
>reliability and human personality, we'd just hire somebody.
>
You clearly haven't read anything I've been saying elsewhere - that's to
be expected as you probably don't take comp.ai.philosophy. But what you
say here is a long non sequitur.
>So this device we're trying to build isn't a human mind. It
>might wind up resembling a human mind to about the same degree
>that the engine and drivetrain of a Stanley Steamer resemble the
>heart and lungs and legs of a cheetah. In fact, I don't know if
>any of the engineers who designed the Stanley cars ever studied
>cheetahs, or for that matter studied the kinematics of humans
>running. They'd probably have judged, correctly, that it was
>only marginally relevant to the somewhat different problem they
>were trying to solve.
Well, as a point of fact, if you look at what people *were* trying to do
when they built the first computers, or if you just look back at how
Turing came up with his original model (leaving aside Post, Church and
others for the time being), he was clearly driven a human behavioural
model - a computer in fact - but that's just an aside.
>
>Now, it will please us to use the terms "memory" and "learning"
>to describe operations whose effects are analogous to human
>memory and learning, whether or not accomplished in anything
>resembling the same way. Since this is in the comp.* hierarchy,
>we don't expect anyone to be confused by the context.
>
> Bear
Sadly this is NOT true as you will see if you look a bit further around
you.
I'm prepared to accept that *you* may not be confused by the context
(perhaps), but there are plenty of others who are.
--
David Longley
Wayne Rasmussen
David Longley wrote:
Change is the key. Improves is subjective and what may seem like an
improvement today might be a drawback later...
JGCasey
"Wayne Rasmussen" <XvirtualD...@gomonarch.com> wrote in message
news:3F70726A...@gomonarch.com...
If something changes for the better (improves) we say it has learnt.
If something changes for the worse (doesn't improve) we say it has
failed to learn.
John Casey
David Longley
<kjc...@hotkey.net.au> writes
This may be the way that *you* and some other thoughtful people think
about learning, but it is *not* the way those who have been researching
"learning" over the past century or so have defined learning. I agree
that what you say has some intelligent, practical uses, and in many ways
it would be nice if your definition was adequate, but it's always a good
idea (in my view) to look to the experts to see how *they* have dealt
with these matters, and even then, to be extremely careful about
*speculating* even if you find what they have come up with is
unsatisfactory.
--
David Longley
Tim McCormick
I have to agree with DL on the point that improvement is subjective.
Stealing an example from earlier in the thread...
Mr X has been driving around in cars his whole life and never seen an
accident. He has learnt that driving cars is safe.
Mr X takes his first trip on a plane. It crashes and he's lucky to be alive.
He has learnt that flying in planes is very dangerous.
Mr X now drives everywhere instead of taking flights thus increasing his
chances of being involved in a fatal travel accident.
Mr X's learning has changed his internal reasoning for the worse - He is
more likely to die and spends more time travelling.
Maybe then a slight rewording of JCs definition would improve things...
If an agents internal reasoning changes in response to an external stimulae,
it is said to have learnt.
I must say though, for all DL's reduction of everyone elses attempts to
define learning, I can't seem to find him quoting a definition that does
satisfy him. I did miss the first few mails on this thread, so apologies if
its there, but I'd be interested to hear it.
Cheers,
Tim
JGCasey
"Tim McCormick" <tim.mc...@openwave.com> wrote in message
news:3f717ec1$0$15782$ed9e...@reading.news.pipex.net...
> Maybe then a slight rewording of JCs definition would improve things...
> If an agents internal reasoning changes in response to an external
stimulae,
> it is said to have learnt.
Lets take the definition suggested by one of my books on
learning and memory and the reasons for that definition.
"Learning is a relatively permanent change in behaviour potentiality
that occurs as a result of reinforced practice." Gregory Kimble.
permanent - because changes in behaviour may be due to mood,
fatigue, hunger ect.
practice - because to we want to exclude changes due to maturation
or sensory adaption.
behaviour potentiality - because a learnt behaviour may not be
translation into behaviour (performance).
reinforced practice - learning will not take place unless it leads to or
is followed by some reward
Actually I think learning in itself can be rewarding. Learning in the
sense of reducing your uncertaintity about things. What is over the
other side of the hill? What is in the box? What will happen if I
press this button? The reward may be in the knowing.
John Casey
David Longley
<kjc...@hotkey.net.au> writes
>
>"Tim McCormick" <tim.mc...@openwave.com> wrote in message
>news:3f717ec1$0$15782$ed9e...@reading.news.pipex.net...
>
>> Maybe then a slight rewording of JCs definition would improve things...
>> If an agents internal reasoning changes in response to an external
>stimulae,
>> it is said to have learnt.
>
>Lets take the definition suggested by one of my books on
>learning and memory and the reasons for that definition.
That seems like a positive step.
>
>"Learning is a relatively permanent change in behaviour potentiality
>that occurs as a result of reinforced practice." Gregory Kimble.
And it's a good book if it's "Hilgard and Marquis' Conditioning and
Learning" (1961) - a little old, but a classic. Have a look at models by
Rescorla & Wagner (1972) [see Rescorla 1988 - "Pavlovian Conditioning -
It isn't what You Think It Is"], and Skinner's work on Operant
Conditioning (NB it's not S-R conditioning) as well.
>
>permanent - because changes in behaviour may be due to mood,
>fatigue, hunger ect.
>
OK.
>practice - because to we want to exclude changes due to maturation
>or sensory adaption.
>
OK.
>behaviour potentiality - because a learnt behaviour may not be
>translation into behaviour (performance).
>
OK.
>reinforced practice - learning will not take place unless it leads to or
>is followed by some reward
>
Tricky area - it has occasionally been suggested that learning *may*
occur in the absence of reinforcement - but this is contentious.
>Actually I think learning in itself can be rewarding. Learning in the
>sense of reducing your uncertaintity about things. What is over the
>other side of the hill? What is in the box? What will happen if I
>press this button? The reward may be in the knowing.
>
>John Casey
>
There is a danger of speculating here - I'd advise you to let your
thoughts on learning be *shaped by* the terms and conditions which have
come to be used in the empirical literature - do not think of reward in
hedonic terms, and whilst "uncertainty reduction" (information) *is* a
term which has been used by some, my advice, again, is to be driven by
the empirical literature and models. read about learning and
non-associative plasticity in very simple animals such as Aplysia or
Drosophilia or higher animals such as pigeons and rats.
Finally, if you can find the paper, see what Skinner says about
"theories" of learning.
--
David Longley
JGCasey
> In article <3f721...@news.iprimus.com.au>, JGCasey
> <kjc...@hotkey.net.au> writes
> >
> >"Tim McCormick" <tim.mc...@openwave.com> wrote in message
> >news:3f717ec1$0$15782$ed9e...@reading.news.pipex.net...
> >
> >> Maybe then a slight rewording of JCs definition would improve things...
> >> If an agents internal reasoning changes in response to an external
> >stimulae,
> >> it is said to have learnt.
> >
> >Lets take the definition suggested by one of my books on
> >learning and memory and the reasons for that definition.
>
> That seems like a positive step.
>
> >
> >"Learning is a relatively permanent change in behaviour potentiality
> >that occurs as a result of reinforced practice." Gregory Kimble.
>
> And it's a good book if it's "Hilgard and Marquis' Conditioning and
> Learning" (1961) - a little old, but a classic. Have a look at models by
> Rescorla & Wagner (1972) [see Rescorla 1988 - "Pavlovian Conditioning -
> It isn't what You Think It Is"], and Skinner's work on Operant
> Conditioning (NB it's not S-R conditioning) as well.
The definition was in an old 1977 text book "Fundamentals of Learning
and Memory" by John P. Houston that I picked up for a few dollars from
the "brain food" section of a used book sale.
SNIP
> >Actually I think learning in itself can be rewarding. Learning in the
> >sense of reducing your uncertaintity about things. What is over the
> >other side of the hill? What is in the box? What will happen if I
> >press this button? The reward may be in the knowing.
> >
> >John Casey
> >
>
> There is a danger of speculating here - I'd advise you to let your
> thoughts on learning be *shaped by* the terms and conditions which have
> come to be used in the empirical literature - do not think of reward in
> hedonic terms, and whilst "uncertainty reduction" (information) *is* a
> term which has been used by some, my advice, again, is to be driven by
> the empirical literature and models. read about learning and
> non-associative plasticity in very simple animals such as Aplysia or
> Drosophilia or higher animals such as pigeons and rats.
>
> Finally, if you can find the paper, see what Skinner says about
> "theories" of learning.
I found "Are Theories of Learning Necessary?" B.F. Skinner 1950
Observation and experiment may be the requirement of hard
science but trying to figure out what mechanisms may produce
such behaviours is required if you want to build machines that
can act "intelligently". This involves creative thinking or theory
weaving. And as BF Skinner says in the conclusion in the above
article, theories are fun.
Regardless at what level of description is used in explaining
how a particular program works it can always be reduced
to a sequence of objective statements at the hardware level.
If I wanted to explain how one of my programs worked I
would not give someone a print out of the code. First I would
explain it in higher level (folk psychology?) terms and then
explain how I embodied these ideas in actual code.
In truth I have never written a program that learns in the
biological sense. Their "learning" was restricted to reweighting
associations between features and "objects".
It is much simpler to hard code fixed processing steps and
fine tune the code or add extra processes to get it to work.
Thus if the program makes a mistake you figure out why
and change the code to take the unpredicted condition
into account. These are "instinctive" behaviours not learnt
behaviours evolved in human-machine system rather than a
environment-machine system. The programmer is an essential
part of the process.
But I am interested in "learning" mechanisms. Intelligence
at a higher level implies the ability to learn by itself without
the need for a programmer to feed it everything it needs
to know.
John Casey
David Longley
<kjc...@hotkey.net.au> writes
>
>"David Longley" <Da...@longley.demon.co.uk> wrote in message
>news:7T01LKEmHqc$Ew...@longley.demon.co.uk...
>> In article <3f721...@news.iprimus.com.au>, JGCasey
>> <kjc...@hotkey.net.au> writes
>> >
>> >"Tim McCormick" <tim.mc...@openwave.com> wrote in message
>> >news:3f717ec1$0$15782$ed9e...@reading.news.pipex.net...
>> >
>> >> Maybe then a slight rewording of JCs definition would improve things...
>> >> If an agents internal reasoning changes in response to an external
>> >stimulae,
>> >> it is said to have learnt.
>> >
>> >Lets take the definition suggested by one of my books on
>> >learning and memory and the reasons for that definition.
>>
>> That seems like a positive step.
>>
>> >
>> >"Learning is a relatively permanent change in behaviour potentiality
>> >that occurs as a result of reinforced practice." Gregory Kimble.
>>
>> And it's a good book if it's "Hilgard and Marquis' Conditioning and
>> Learning" (1961) - a little old, but a classic. Have a look at models by
>> Rescorla & Wagner (1972) [see Rescorla 1988 - "Pavlovian Conditioning -
>> It isn't what You Think It Is"], and Skinner's work on Operant
>> Conditioning (NB it's not S-R conditioning) as well.
>
>The definition was in an old 1977 text book "Fundamentals of Learning
>and Memory" by John P. Houston that I picked up for a few dollars from
>the "brain food" section of a used book sale.
>
Still, it's sounds like a reliable source - the Hilgard & Marquis book
is by Kimble and the definitions are in fact from there.
>
> SNIP
>
>> >Actually I think learning in itself can be rewarding. Learning in the
>> >sense of reducing your uncertaintity about things. What is over the
>> >other side of the hill? What is in the box? What will happen if I
>> >press this button? The reward may be in the knowing.
>> >
>> >John Casey
>> >
>>
>> There is a danger of speculating here - I'd advise you to let your
>> thoughts on learning be *shaped by* the terms and conditions which have
>> come to be used in the empirical literature - do not think of reward in
>> hedonic terms, and whilst "uncertainty reduction" (information) *is* a
>> term which has been used by some, my advice, again, is to be driven by
>> the empirical literature and models. read about learning and
>> non-associative plasticity in very simple animals such as Aplysia or
>> Drosophilia or higher animals such as pigeons and rats.
>>
>> Finally, if you can find the paper, see what Skinner says about
>> "theories" of learning.
>
>I found "Are Theories of Learning Necessary?" B.F. Skinner 1950
>
>Observation and experiment may be the requirement of hard
>science but trying to figure out what mechanisms may produce
>such behaviours is required if you want to build machines that
>can act "intelligently". This involves creative thinking or theory
>weaving. And as BF Skinner says in the conclusion in the above
>article, theories are fun.
I'm not sure about that - it's been the subject of some discussion in
c.a.p over recent months. The problem is that we are embedded in an
implicit theory which we all take for granted ('folk psychology') and we
have good reasons to believe that it is radically flawed. One can easily
find social endorsement of one's folk psychological preconceptions,
especially in forums such as these - none of which really helps anyone
in the long run.
>
>Regardless at what level of description is used in explaining
>how a particular program works it can always be reduced
>to a sequence of objective statements at the hardware level.
>
>If I wanted to explain how one of my programs worked I
>would not give someone a print out of the code. First I would
>explain it in higher level (folk psychology?) terms and then
>explain how I embodied these ideas in actual code.
I'm not sure why - especially if the person you're explaining it to is a
programmer and knows the language.
>
>In truth I have never written a program that learns in the
>biological sense. Their "learning" was restricted to reweighting
>associations between features and "objects".
>
>It is much simpler to hard code fixed processing steps and
>fine tune the code or add extra processes to get it to work.
>Thus if the program makes a mistake you figure out why
>and change the code to take the unpredicted condition
>into account. These are "instinctive" behaviours not learnt
>behaviours evolved in human-machine system rather than a
>environment-machine system. The programmer is an essential
>part of the process.
Surely that's what any scientist does though?
>
>But I am interested in "learning" mechanisms. Intelligence
>at a higher level implies the ability to learn by itself without
>the need for a programmer to feed it everything it needs
>to know.
>
>John Casey
>
You might find what Bill Modlin is trying to do quite
interesting/challenging - have a look at the problem he has outlined in
past posts to c.a.p
--
David Longley
Eray Ozkural exa
> >Regardless at what level of description is used in explaining
> >how a particular program works it can always be reduced
> >to a sequence of objective statements at the hardware level.
> >
> >If I wanted to explain how one of my programs worked I
> >would not give someone a print out of the code. First I would
> >explain it in higher level (folk psychology?) terms and then
> >explain how I embodied these ideas in actual code.
>
> I'm not sure why - especially if the person you're explaining it to is a
> programmer and knows the language.
It's obvious that you don't have a clue about programming. Giving a
print out is the least efficient way of communicating the abstract
design of a program.
And if the code isn't commented well or documented in some other
fashion, it's going to be pretty hard to understand it: if the code is
large enough it's not going to be feasable. Been there and done that,
dealing with bit junk is no fun.
It's no surprise that you are not qualified to talk about this subject
not having written any code above the dump-db-app level.
cap's resident real-world and academic programmer,
--
examachine
David Longley
Ozkural exa <er...@bilkent.edu.tr> writes
>David Longley <Da...@longley.demon.co.uk> wrote in message
>news:<o1MRCICF09c$Ew...@longley.demon.co.uk>...
>> >Regardless at what level of description is used in explaining
>> >how a particular program works it can always be reduced
>> >to a sequence of objective statements at the hardware level.
>> >
>> >If I wanted to explain how one of my programs worked I
>> >would not give someone a print out of the code. First I would
>> >explain it in higher level (folk psychology?) terms and then
>> >explain how I embodied these ideas in actual code.
>>
>> I'm not sure why - especially if the person you're explaining it to is a
>> programmer and knows the language.
>
>It's obvious that you don't have a clue about programming. Giving a
>print out is the least efficient way of communicating the abstract
>design of a program.
You're a little too fond of deciding what's obvious. In this case, the
really important point is what language one uses to communicate
scientific knowledge. The reason why there's so much argument (rather
than useful explicative exchange) in newsgroups and elsewhere is because
people use the wrong language.
I stand by what I said - and there is certainly no reason why anyone
should need to explain their programs in folk psychological terms.
Structuring and documenting a program is standard professional practice,
and even where this is not done, a competent programmer can follow the
logic in 3GLs or assembler.
>
>And if the code isn't commented well or documented in some other
>fashion, it's going to be pretty hard to understand it: if the code is
>large enough it's not going to be feasable. Been there and done that,
>dealing with bit junk is no fun.
As happens too frequently here - pointless and irrelevant remarks.
>
>It's no surprise that you are not qualified to talk about this subject
>not having written any code above the dump-db-app level.
You have no reason to believe any of the above - and such remarks serve
no purpose than to foment pointless conflict.
>
>cap's resident real-world and academic programmer,
>
>--
>examachine
Grow up.
--
David Longley
Bill Modlin
<snip>
>>If I wanted to explain how one of my programs worked I
>>would not give someone a print out of the code. First I would
>>explain it in higher level (folk psychology?) terms and then
>>explain how I embodied these ideas in actual code.
>I'm not sure why - especially if the person you're explaining it to is
>a programmer and knows the language.
For the same reason that all good programs are liberally annotated with
comments. Programming languages are for controlling computers, not for
communicating with people. It is laborious and error prone to attempt
to discover the purpose of and algorithm implemented by a bit of code
just by inspecting that code.
This is true even for the original author of the code. Code that seemed
crystal clear and obvious at the time you wrote it is often
indecipherable when a year later you are called upon to adjust or fix it
in some fashion.
Of course, the extensional function of each line of code is obvious to
anyone who knows the language. There is no mystery in a line of code
that says to increment the value of some variable by two: it says
exactly what is to happen.
What is missing is the intensional context: why, at this particular
point in the program one should want to do just this and not increment
by one or three, or perhaps do some other entirely different and
unrelated function. That may not be at all obvious, and one can spend
days or weeks trying to figure it out. In the context of fixing old
code it may easily appear that you have found a problem, so you change
that instruction and it fixes the bug you were working on, but since you
did not understand why it was supposed to be an increment by two, this
breaks some other aspect of the program function and things become even
worse.
Literal quoting of code is sometimes a useful adjunct to understanding
of function. We all generally like to have actual source to work with.
But it is only an adjunct: the necessary ingredient for understanding of
non-trivial function is explanation of the principles of operation in
human-interpretable language. Given the code one may never understand
what it is for. Given an understandable explanation of just what is to
be done and how to do it, any programmer can produce new code for a
function readily enough: having an actual worked out version to go by
is a dispensable luxury.
<snip>
>>But I am interested in "learning" mechanisms. Intelligence
>>at a higher level implies the ability to learn by itself without
>>the need for a programmer to feed it everything it needs
>>to know.
>>
>>John Casey
>
>You might find what Bill Modlin is trying to do quite
>interesting/challenging - have a look at the problem he has outlined in
>past posts to c.a.p
>
>--
>David Longley
Thanks for the plug, David. :-) In fact John has written to me, and I
am in the process of trying to compose an appropriate response to his
reasonable request for a "simple, not too wordy" explanation of my
ideas. It's not easy: I'm tempted to quote my previous writings in
which I have tried to provide just that. But instead I'm working on yet
another explanation, a paraphrase, in my ongoing attempt to convey an
intended meaning rather than preserve a literal set of words which have
proven inadequate to the task of inducing understanding in my hearers.
Bill Modlin
Acme Posting
> In article <fa69ae35.03092...@posting.google.com>, Eray
> Ozkural exa <er...@bilkent.edu.tr> writes
> >
> >not having written any code above the dump-db-app level.
>
> You have no reason to believe any of the above - and such remarks serve
> no purpose than to foment pointless conflict.
>
> >
> >cap's resident real-world and academic programmer,
> >
> >--
> >examachine
>
> Grow up.
Eray grew up. Eray listens. Eray absorbs information like a sponge.
Eray towers over you intellectually. Eray absorbed you almost
instantly and spit you out. Eray grows. You diminish.
Larry
Acme Posting
> "David Longley" <Da...@longley.demon.co.uk> wrote in message
> news:o1MRCICF09c$Ew...@longley.demon.co.uk...
>
> >>If I wanted to explain how one of my programs worked I
> >>would not give someone a print out of the code. First I would
> >>explain it in higher level (folk psychology?) terms and then
Discussing program design is "folk psychology." Funny none of my
program design manuals have "Psychology" in the title.
<snip>
>
> which I have tried to provide just that. But instead I'm working on yet
> another explanation, a paraphrase, in my ongoing attempt to convey an
> intended meaning rather than preserve a literal set of words which have
> proven inadequate to the task of inducing understanding in my hearers.
:-)
Larry
Arthur T. Murray
<snip>
>>> If I wanted to explain how one of my programs worked
>>> I would not give someone a print out of the code.
>>> First I would explain it in higher level (folk psychology?)
>>> terms and then explain how I embodied these ideas in actual code.
>> I'm not sure why - especially if the person you're explaining it to
>> is a programmer and knows the language.
> For the same reason that all good programs are liberally annotated
> with comments. Programming languages are for controlling computers,
> not for communicating with people. It is laborious and error prone
> to attempt to discover the purpose of and algorithm implemented by
> a bit of code just by inspecting that code.
Nevertheless the original source code has to be made available, e.g.
http://www.scn.org/~mentifex/jsaimind.html -- AI Mind in JavaScript;
http://mentifex.virtualentity.com/mind4th.html -- Mind.Forth PD AI.
BM:
> This is true even for the original author of the code.
> Code that seemed crystal clear and obvious at the time
> you wrote it is often indecipherable when a year later
> you are called upon to adjust or fix it in some fashion.
ATM:
The factoring of Forth and the hierarchy of objects help here,
plus liberal comments and statements at the end of each Mind-
Module explaining to what other Module program-flow returns:
http://mentifex.virtualentity.com/progman.html -- Modules.
BM:
> Of course, the extensional function of each line of code is obvious
> to anyone who knows the language. There is no mystery in a line of
> code that says to increment the value of some variable by two:
> it says exactly what is to happen. [...]
> Literal quoting of code is sometimes a useful adjunct to understanding
> of function. We all generally like to have actual source to work with.
Yes, and so the AI4U (AI For Your) textbook of artificial intelligence at
http://search.barnesandnoble.com/bookSearch/isbnInquiry.asp?ISBN=0595259227
contains a full source-code listing of the AI Mind-1.1 in MSIE JavaScript.
On September 24, 2003, a rather mean-spirited reviewer at B&N, hiding
behind anonymity, complained that "This book is not really a textbook
because it explains nothing. It rather is a collection of 'modules'
which the author expects the reader to implement" -- without pointing out
that the AI Mind modules are already implemented at the end of the book.
Now this author is indeed trying to get Minds coded in XYZ languages:
http://mentifex.virtualentity.com/cpp.html -- C++ with new AI code;
http://mentifex.virtualentity.com/java.html -- see Mind.JAVA #1 & #2;
http://mentifex.virtualentity.com/lisp.html -- Lisp AI Weblog;
http://mentifex.virtualentity.com/perl.html -- first Perl module;
http://mentifex.virtualentity.com/prolog.html -- Prolog AI Weblog;
http://mentifex.virtualentity.com/python.html -- Python AI Weblog;
http://mentifex.virtualentity.com/ruby.html -- Ruby AI Blog (OO AI);
http://mentifex.virtualentity.com/scheme.html -- Scheme AI Weblog;
http://mentifex.virtualentity.com/vb.html -- see "Mind.VB #001" link.
BM:
> But it is only an adjunct: the necessary ingredient for understanding
> of non-trivial function is explanation of the principles of operation
> in human-interpretable language.
ATM:
I have created every conceivable instrument of explaining the AI Mind.
http://mentifex.virtualentity.com/acm.html is Steps to DIY AI.
http://mentifex.virtualentity.com/standard.html -- AI Standards.
http://mentifex.virtualentity.com/variable.html -- AI Variables.
BM:
> Given the code one may never understand what it is for.
ATM:
Recently I have been meeting in person with an incognito engineer
who has volunteered to attempt to translate (port) Mind.Forth into
http://mentifex.virtualentity.com/aicppsrc.html AI C++ Source Code.
BM:
> Given an understandable explanation of just what is to be done
ATM:
Nothing beats having a one-on-one interactive discussion of
what is to be done to code AI and how to do it. That optimal
option is why I have a dreamworld fantasy of hiring out to CS
departments and AI labs to have my primitive but AI-coding
brain picked clean and dry of any contribution I make to AI.
> and how to do it, any programmer can produce new code
> for a function readily enough: having an actual worked out
> version to go by is a dispensable luxury. [...]
A.T. Murray
--
http://www.bloggingnetwork.com/Blogs/Blog.aspx/ai/ has been solved.
http://www.amazon.com/exec/obidos/ASIN/0595654371/ -- AI Textbook;
http://www.sl4.org/archive/0205/3829.html -- Goertzel on Mentifex;
http://doi.acm.org/10.1145/307824.307853 -- ACM SIGPLAN Mind.Forth
David B. Held
news:3f74...@news.victoria.tc.ca...
> [...]
> The factoring of Forth and the hierarchy of objects help here,
> plus liberal comments and statements at the end of each Mind-
> Module explaining to what other Module program-flow returns:
The PROBLEM is that the only implementation of your "AI Mind"
that I've seen run does nothing but produce a bunch of gibberish
that would make Eliza blush and Parry scream in a fit. To foist
such garbage on the world under the guise of AI makes you a first-
class charlatan. At least Weizenbaum never claimed that Eliza was
truly intelligent.
> [...]
> Yes, and so the AI4U (AI For Your) textbook of artificial intelligence
You're below 500,000 on B&N's ranking, and well below 2 million
on Amazon's ranking. You better pump up those references, Art!
> [...]
> On September 24, 2003, a rather mean-spirited reviewer at B&N,
> hiding behind anonymity,
LOL!!! As opposed to the "honest" reviewer who also implied a
knowledge of Latin and sounds suspiciously like you? I wouldn't
waste my time putting in a fake review for a book I haven't read,
but if, for some reason, I had bought the book, I would have given
a much longer review than that, believe you me.
> complained that "This book is not really a textbook because it
> explains nothing.
If it's anything like your web site (and there's no reason to believe
that it's substantially different), I think that's a pretty justifiable
claim.
> It rather is a collection of 'modules' which the author expects the
> reader to implement" -- without pointing out that the AI Mind
> modules are already implemented at the end of the book.
Well, Forth isn't exactly a popular language (at least, not compared
to Java, C++, etc.). And your "implementation" might produce some
output that is interesting to someone studying randomness, but it
hardly produces the level of AI one would expect from a book
with the audacious subtitle "Mind 1.1 Programmer's Manual".
> [...]
> I have created every conceivable instrument of explaining the AI
> Mind.
> [...] is Steps to DIY AI.
> [...] -- AI Standards.
> [...] -- AI Variables.
You're right. I couldn't possibly conceive of any other way of
explaining an AI. For instance, I couldn't conceive of creating a test
problem and showing how the AI solves it. I couldn't conceive of
referencing existing AI architectures and explaining how it is better
or worse than some other model. I couldn't conceive of writing a
language-neutral pseudo-algorithm that gives a precise description
of function without the distractions of a well-formed language-
specific program.
> [...]
> Recently I have been meeting in person with an incognito engineer
> who has volunteered to attempt to translate (port) Mind.Forth into
> [...] AI C++ Source Code.
So you found someone who doesn't know C++ very well, doesn't
have a career ahead of him, and has nothing to lose by associating
himself with you. Congratulations. You have a lackey. If the
engineer is so proud to be a part of the AI4U team, why is he
incognito?
> [...]
> Nothing beats having a one-on-one interactive discussion of
> what is to be done to code AI and how to do it. That optimal
> option is why I have a dreamworld fantasy of hiring out to CS
> departments and AI labs to have my primitive but AI-coding
> brain picked clean and dry of any contribution I make to AI.
I think you've picked your own brain clean of AI, wiped the
contents on a web site, and left it to fester in the sun. Needless
to say, there is no CS department in the country (and probably
not the world) that would spend one minute talking to you about
your ideas after seeing your work. Even fresh-faced undergrads
can see that you have nothing but smoke and mirrors...without
the smoke...or the mirrors.
Dave
E. Robert Tisdale
when you respond to this thread.
JGCasey
SNIP
> >You might find what Bill Modlin is trying to do quite
> >interesting/challenging - have a look at the problem he has outlined in
> >past posts to c.a.p
> >
> >--
> >David Longley
>
> Thanks for the plug, David. :-) In fact John has written to me, and I
> am in the process of trying to compose an appropriate response to his
> reasonable request for a "simple, not too wordy" explanation of my
> ideas. It's not easy: I'm tempted to quote my previous writings in
> which I have tried to provide just that. But instead I'm working on yet
> another explanation, a paraphrase, in my ongoing attempt to convey an
> intended meaning rather than preserve a literal set of words which have
> proven inadequate to the task of inducing understanding in my hearers.
Remember *not* to use any folk psychological terms otherwise
David will not be able to translate it into anything meaningful :-)
Sorry I have been too lazy to go back over your previous writings.
This time I will try an understand your viewpoint.
John Casey
JGCasey
SNIP
> >Observation and experiment may be the requirement of hard
> >science but trying to figure out what mechanisms may produce
> >such behaviours is required if you want to build machines that
> >can act "intelligently". This involves creative thinking or theory
> >weaving. And as BF Skinner says in the conclusion in the above
> >article, theories are fun.
>
> I'm not sure about that - it's been the subject of some discussion in
> c.a.p over recent months. The problem is that we are embedded in an
> implicit theory which we all take for granted ('folk psychology') and we
> have good reasons to believe that it is radically flawed. One can easily
> find social endorsement of one's folk psychological preconceptions,
> especially in forums such as these - none of which really helps anyone
> in the long run.
At this stage I do not have sufficient grasp of the work done,
or the concepts involved, to have an opinion or any arguments
with regards behavioural versus cognitive viewpoints.
If this 'folk psychology' cannot be translated into anything useful
surely time will reveal its limitations?
Perhaps everything can ultimately be reduced to "behaviours"
just as everything can be reduced to "atoms".
If the methodology of the behaviourist is the appropriate way
to untangle (decode) biological systems it will reveal this by
on going success while other methods will be left in its wake.
Or perhaps I am showing my naivety on the subject?
> >Regardless at what level of description is used in explaining
> >how a particular program works it can always be reduced
> >to a sequence of objective statements at the hardware level.
> >
> >If I wanted to explain how one of my programs worked I
> >would not give someone a print out of the code. First I would
> >explain it in higher level (folk psychology?) terms and then
> >explain how I embodied these ideas in actual code.
>
> I'm not sure why - especially if the person you're explaining it
> to is a programmer and knows the language.
Without documentation (including self documentation in the case
of HLL) it can be very difficult. Explain it in English and maybe
some mathematics and within no time at all a programmer can sit
down and tap out their own version of the algorithm.
Also I would like to explain the code to non programmers.
They might have ideas, expressed in English or sketches etc,
which I can then translate into code.
Code is for computers not people.
SNIP
> > . . .The programmer is an essential part of the process.
>
> Surely that's what any scientist does though?
The idea in AI is to remove the programmer and have a
system that does it by itself :-)
John Casey
David Longley
<kjc...@hotkey.net.au> writes
cases something interesting *does* ensue as a consequence of such novel
'creations', and that's been discussed elsewhere in the context of
"discovery". However, where one is relying on rationality, &
"justification", one must understand how and why logical inference
actually *breaks down* in such contexts.
--
David Longley
David Longley
>news:o1MRCICF09c$Ew...@longley.demon.co.uk...
>> In article <3f73a...@news.iprimus.com.au>, JGCasey
>> <kjc...@hotkey.net.au> writes
>
> SNIP
>
>> >Observation and experiment may be the requirement of hard
>> >science but trying to figure out what mechanisms may produce
>> >such behaviours is required if you want to build machines that
>> >can act "intelligently". This involves creative thinking or theory
>> >weaving. And as BF Skinner says in the conclusion in the above
>> >article, theories are fun.
>>
>> I'm not sure about that - it's been the subject of some discussion in
>> c.a.p over recent months. The problem is that we are embedded in an
>> implicit theory which we all take for granted ('folk psychology') and we
>> have good reasons to believe that it is radically flawed. One can easily
>> find social endorsement of one's folk psychological preconceptions,
>> especially in forums such as these - none of which really helps anyone
>> in the long run.
>
>At this stage I do not have sufficient grasp of the work done,
>or the concepts involved, to have an opinion or any arguments
>with regards behavioural versus cognitive viewpoints.
Fair enough.
>
>If this 'folk psychology' cannot be translated into anything useful
>surely time will reveal its limitations?
>
It isn't as extreme as that, and that's in part why it is so difficult
to discuss. "Folk Psychology" refers to a natural *theory* of "mind",
(or "common-sense") which we naturally learn, teach our children, and
socially reinforce. In my view, it is also what professional research
psychologists study *as scientists*. However, it is not the method
whereby they study - or at least, it shouldn't be. As a tacit "theory"
It's not entirely flawed (which is in part why it is so hard to change),
and parts of it have been refined into science itself. But there are
some facets of it which have known biases (some well known, others not)
and whilst these have been explicated by both philosophers and
experimental psychologists, and whilst such biases tend to be "designed
out" of many professional practices these biases and distortions are
extremely resistant to change even in such professionals. Professional
skills tend to be applied context specifically - which is perhaps
revealing about the nature of learning itself.
There's a lot to this, so don't take the above paragraph to be any more
than a pointer to what is effectively a *vast* literature. The status of
"folk psychology" is a central issue in contemporary philosophy of mind,
and some of the issues are central to AI (although, from many posts in
c.a.p you'd be forgiven for not seeing this).
If you are interested, (http://www.longley.demon.co.uk/Frag.htm or the
.pdf file) - the first part of the former is an application, the
theoretical part is a little way in. It's*not* an easy read
unfortunately, but then, it wasn't easy to write either.
>Perhaps everything can ultimately be reduced to "behaviours"
>just as everything can be reduced to "atoms".
I don't think it's a matter of "reduction" - there are just some areas
of behaviour where we are still rather ignorant - how one deals with
this ignorance is critical - does one fill in the gaps with magical non-
explanations (something "Cognitive Science" tends to do, albeit
inadvertently) or does one keep those gaps open and highlight them as
important areas for research?
I've put a slant on this - but may don't see the problem at all.
>
>If the methodology of the behaviourist is the appropriate way
>to untangle (decode) biological systems it will reveal this by
>on going success while other methods will be left in its wake.
Which is, in fact what happens - although you would think otherwise
through reading some of the more popular literature.
>
>
>Or perhaps I am showing my naivety on the subject?
We are all naive on this sadly. It's a difficult issue - psychology is a
popular university subject. Many do it just as a minor subject and
there's just too much to study to get an accurate understanding of what
a lot of it is all about. This is true of any science or profession of
course, but everyone considers themselves a psychologist to some extent.
>
>> >Regardless at what level of description is used in explaining
>> >how a particular program works it can always be reduced
>> >to a sequence of objective statements at the hardware level.
>> >
>> >If I wanted to explain how one of my programs worked I
>> >would not give someone a print out of the code. First I would
>> >explain it in higher level (folk psychology?) terms and then
>> >explain how I embodied these ideas in actual code.
>>
>> I'm not sure why - especially if the person you're explaining it
>> to is a programmer and knows the language.
>
>Without documentation (including self documentation in the case
>of HLL) it can be very difficult. Explain it in English and maybe
>some mathematics and within no time at all a programmer can sit
>down and tap out their own version of the algorithm.
English is fine, my challenge was "folk psychology" - they're not the
same thing.
>
>Also I would like to explain the code to non programmers.
>They might have ideas, expressed in English or sketches etc,
>which I can then translate into code.
>
>Code is for computers not people.
>
It's a mater of languages and behaviour.
>SNIP
>
>> > . . .The programmer is an essential part of the process.
>>
>> Surely that's what any scientist does though?
>
>The idea in AI is to remove the programmer and have a
>system that does it by itself :-)
>
>John Casey
>
Possibly, and this would appear to be a significant problem which Bill
Modlin more than anyone else here (apart from N.R *perhaps*) is deeply
immersed in.
>
>
>
--
David Longley
Acme Posting
> In article <3f74b...@news.iprimus.com.au>, JGCasey
> <kjc...@hotkey.net.au> writes
> >
> >"Bill Modlin" <mod...@metrocast.net> wrote in message
> >news:3LOdnYcq9oV...@metrocast.net...
>>> Thanks for the plug, David. :-) In fact John has written to me,
and I
>>> am in the process of trying to compose an appropriate response to
his
>>> reasonable request for a "simple, not too wordy" explanation of my
>>> ideas. It's not easy: I'm tempted to quote my previous writings
in
>>> which I have tried to provide just that. But instead I'm working
on yet
>>> another explanation, a paraphrase, in my ongoing attempt to convey
an
>>> intended meaning rather than preserve a literal set of words which
have
>>> proven inadequate to the task of inducing understanding in my
hearers.
>>
>>Remember *not* to use any folk psychological terms otherwise
>>David will not be able to translate it into anything meaningful :-)
<snip>
>"discovery". However, where one is relying on rationality, &
>"justification", one must understand how and why logical inference
>actually *breaks down* in such contexts.
The only place where logical inference "breaks down" is inside
Longley's head.
"Hmmm, I'm in my early 40's. What should I do with the rest of my
life? Maybe I should write up my opinions on a web site and then
repeat my 8 favorite words thousands of times in a newsgroup? Duh,
that sounds logical! Then everybody will know that it wasn't my fault,
it was those idiot folk psychologist's fault!
"Hmmm, now I'm about 50 and nobody has agreed with my opinions despite
thousands of repetitions. Duh, logic must have broken down! Quine was
right! Everybody else's language is worse than I deduced! Everybody in
newsgroups is folk psychological too! How to solve this problem?"
"Let me analyze more extensionally! Well, I could spend my 50's
posting another 8,000 repetitions. Duh, that sounds like a good
decision! And if that doesn't work, as a backup I still have my 60's!"
Larry
David Longley
Posting <lf....@lycos.co.uk> writes
http://www.homeoffice.gov.uk/rds/pdfs2/r206.pdf
and for background:
What Works & What Can be Effectively Managed: A Close Look at the Data
(1997a) http://www.longley.demon.co.uk/Sm-97apr.pdf
A Further Comment on Recent Claims in the 'Rehabilitation of
Rehabilitation Literature..(1997b)
http://www.longley.demon.co.uk/Workj97.pdf
and:
http://www.longley.demon.co.uk/Frag.htm
or
http://www.longley.demon.co.uk/Fragjn97.pdf
--
David Longley
Acme Posting
Well in light of all these crossposts I suppose someone might infer
that I know something about Longley's activities besides posting
11,000+ newsgroup messages. I don't. I'm sure he did other things.
Assuming it's possible to be unfair to someone who has posted his
handful of ritual words and promoted his web site 8,000+ times, I
apologize.
I would also like to point out that the original instigation for the
antagonism between Longley and I was an attack by him on logical
argumentation more-or-less calling it "folk psychology." He knew then
and knows now that I am highly invested in it. There was little doubt
in my mind that his comments at the top were at least partly intended
as a further provocation though of course I don't know anything for
sure.
If anyone finds one of my posts ambiguous, just ask and I will be
happy to provide a straightforward explanation.
Larry
David Longley
I find your post more than *ambiguous* - it is disingenuous to say the
least. Not only do you not understand some of the most important
concepts in modern philosophy, psychology and AI, but you falsely
ascribe statements to me which are in fact pure fabrication. You do
*not* argue validly you argue invalidly (e.g your frequent recourse to
ad hominem and intensional idiom), which I have referred to as
"nefarious rhetoric". When I explain this to you and reference it by
quoting Quine (one of most accomplished philosophers of mathematical
logic in recent history), you don't learn from thus, instead, you
dismiss *him* as verbose, or worse!. This is not a good way of arguing
to put it mildly, and the fact that you choose to indulge in this, and
other kinds of rhetoric, and that I choose not to, does not mean that
you have skills in "arguing" or "reasoning" which I do not - it means
precisely the opposite, ie that you behave irrationally and that you
don't see what's wrong it.
As I see it, what you tend to do in this newsgroup (in your
correspondence with me, about me or about what I post) is confuse your
own vague and muddled "thoughts" about what I have written with what I
have actually written itself. You may well be "highly invested" in this
irrational behaviour, but in this context I am only really interested in
your difficulties to the extent that I can refer to them as
*illustratively* obstructive to rational discussion of the philosophy of
AI and effective behaviour management. These difficulties are, in my
view, characteristic of a widely held and deeply flawed folk psychology.
In brief you share the rather characteristically human difficulty of not
being very skilled at discriminating what is your own private behaviour
(thoughts/interpretations/translations) from what is actually said
(publicly). This has been highlighted as a characteristic "feature" of
private behaviour (the subject of at least one seminal paper by Skinner
written back in 1945 "The Operational Analysis of Psychological Terms").
This is of course quite central to what I have been drawing attention to
more widely. Instead of *counting* fragments of my posts to Usenet,
perhaps your time would be better spent actually reading some of them!
--
David Longley
Ray Dillinger
<Clip -- doesn't really matter>
You know, I really like ice cream.
Bear
Ray Dillinger
>
<Clip -- Doesn't really matter>
Chocolate syrup is really good.
Ray Dillinger
<clip -- doesn't really matter>
Y'know, strawberries are really nice.
Especially when preserved with sugar
and served over ice cream. Yum!
Bear
Ray Dillinger
<clip -- really doesn't matter>
It's good to have a cat curled up in your lap and purring.
Well, unless cat hair makes you sneeze.
Bear
Ray Dillinger
<clip -- doesn't really matter>
Sunshine feels great.
Too much of it can give you cancer, but what the hey.
Bear
Ray Dillinger
<clip -- doesn't really matter>
Y'know, vegetables are very important to a balanced diet.
Without them food would never get big enough to slaughter.
Bear
Ray Dillinger
<clip -- doesn't really matter>
When I was a kid, I used to love sledding on winter mornings.
Now I live where there's no snow.
They say the climate is better here.
But the sledding is a lot worse.
Bear
Ray Dillinger
<clip -- doesn't really matter>
It's really great, in the fall when the leaves have turned
and fallen, to go out and rake leaves. When you're done you
can jump in the big pile of leaves and roll around. If you
get too cold, you can come in and drink hot spiced apple cider.
Things like that make me a lot happier than studying
what philosophers say about happiness.
Bear
Ray Dillinger
<clip -- doesn't really matter>
You know, you may consider many of my replies to David
today to have been offtopic.
You may be right.
But consider them in context. :-)
Bear
Alan Grimes
> Isn't it high time for AI to get its own field of interest, set of notions,
> and methodology?
> Mike
That's premature.
Right now AI is an "art".
It can't become a science until someone has "perfected" it to some
limited extent.
You are free to develop your own notations, etc...
They will either catch on or not depending on how smart your code proves
itself to be. ;)
JXStern
<alang...@starpower.net> wrote:
>> Isn't it high time for AI to get its own field of interest, set of notions,
>> and methodology?
>
Say what?
If anything, it's been too busy with its own notations to draw on
other domains, and has been for forty years.
Which is not to say that it doesn't desparately need even more.
J.
Steven M. Haflich
> Right now AI is an "art".
> It can't become a science until someone has "perfected" it to some
> limited extent.
This position seems to ignore 40+ years of history in the field.
There is a (possibly-apocryphal) statement by Minsky that AI is
only what we don't _yet_ know how to make a computer do. There was
a time within my living memory when computer speech of English
text was considered AI. Then there were successive advances that
basically solved the problem, evidenced for example by the
appearance in the early 1980's of products like DECTalk which
was a $600 device that took RS-232 ASCII in one end and emitted
human speech audio on the other. Today you can get even better
software that rnus on the $0.50 sound chip on the generic sound
card in your $250 generic PC.
If AI is only what we don't yet know how to make a computer do,
it follows that it is silly to expect someone to pay for AI.
It is quite reasonable to pay an engineer to do something he knows
how to do; it is not often reasonable to pay an engineer to
do something he does not know how to do (except sometimes when
it is explicitly identified as "research").