Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.sprintlink.net!qns3.qns.com!news.ecn.uoknor.edu!munnari.oz.au!cs.mu.OZ.AU!dnk
From: dnk@cs.mu.OZ.AU (David Kinny)
Subject: Re: Machine Learning Breakthrough
Message-ID: <9516403.24089@mulga.cs.mu.OZ.AU>
Organization: Computer Science, University of Melbourne, Australia
References: <3r3jkv$lnh@newsbf02.news.aol.com>
Date: Mon, 12 Jun 1995 17:16:08 GMT
Lines: 700

marengom@aol.com (MarengoM) writes:

>waynei@csdc02.orl.mmc.com (Wayne Iba) writes:

>>Would you'all _please_ change the subject heading?  I can see that
>>no evidence for AXIS as a breakthrough has been presented and I
>>would hate to miss a REAL breakthrough posted under the
>>appropriate subject heading.
>>
>>What this thread has been doing is discussing the AXIS system.
>>Could we possibly use a little honesty in our subject headings?
>>Please?

>  Somehow we don't think it's the subject heading of this thread that
>really bothers you.  If you don't understand AXIS or if you don't like the
>subject heading, please use your kill file.  Ignoring the thread
>completely would give you peace of mind and would spare us from having to
>read messages that do not contribute to AXIS.

>Best regards,

>Louis Savain

>Marengo Media, Inc.

Hey, he asked nicely and it's a reasonable request too.  But you
(plural) don't think it's the subject heading of this thread that
really bothers him.  That just demonstrates that you're so blinded
by your ridiculous "breakthrough" that you can't even understand why
someone would be bothered by the title.  Let me spell it out ...

Your breakthrough is nothing but a display of your naivete and
ignorance about machine learning and logic, decorated by your
quasi-religious beliefs that one logical connective is somehow
"more fundamental" than another.  Your postings are full of pompous
drivel and obvious errors.  Examples below for anyone who needs
to have their nose rubbed in it.

Axis is not a breakthrough.  Like most cranks, you have come up with
a minor variant of existing ideas, developed some new terminology,
called it a major advance, and totally glossed over the *hard* parts
of the machine learning problem, claiming in passing that you have
solved them, but offering no evidence for this whatsoever.  Seriously,
you're only one or two steps removed from the Abians and Sarfattis.

The really sad thing about this thread is that there are enough
clueless amateur geniuses here to be taken in by it all.  We have you
and lamentably ignorant phd students arguing about whether you can get
OR from AND and XOR, stuff any sophomore logic student would know, but
the occasional sensible question like "How does this relate to adaptive
logic networks?" or "So you're doing unsupervised learning?" go
completely unanswered by you, almost certainly because you haven't
a clue about what is being asked.

Start posting under the subject-line "the AXIS system" or something
similar that more honestly describes this blather, and those of
us who still scan this group for the few grains of intelligent
thought amongst all the babble and ignorance will leave you in peace.
Continue to claim that this is a breakthrough and you take the risk
that this thread will degenerate into a flame-fest.

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
David Kinny
Department of Computer Science
University of Melbourne, AUSTRALIA


Appendix: The annotated AXIS 

Marengom writes:
> The following consists of the first two posts that started the thread
> 'Machine Learning Breakthrough' on comp.ai.
> 
> 
> 
> Machine Learning Breakthrough
> By Louis Savain
> 
> Preface
> 
>   We are in possession of a novel and simple technology that promises to
> revolutionize both computer science and artificial intelligence.  A year
> or so ago, Curtis L. May II and I founded Marengo Media, Inc. to take
> advantage of what we considered to be a remarkable idea having to do with
> the development of extremely reliable parallel/distributed applications. 

Remarkable, above all, for its naivete.

> The technology consists of a simple software mechanism that one can use to
> detect and correct inconsistencies within the logic of a parallel program.
>  We knew then, as we know now, that the ability to detect internal logical
> conflicts in programs had a direct bearing on the field of AI in general
> and learning systems in particular.  One only needs a machine with some
> form of input/output capability, add the capacity to automatically
> generate connections in accordance with a well-defined method, add a way
> to detect and correct errors among the connections and presto, one has a
> general learning machine.

Gee, that simple huh?

>   We are faced with an interesting dilemma:  should we or should we not
> make our findings public?  Since we are a for-profit corporation, the
> temptation, understandably, is to keep our trade secrets to ourselves,
> quietly develop the technology and eventually make a killing in the market
> place.

Dream on ...

>       On the other hand, we realize that no company is an island unto
> itself.  Not only can this technology be of immense benefit to humanity,

Delusions of grandeur here ...

> but its concentration in the hands of a few is not very wise, given its
> enormous potential and the current state of human nature.  We also feel
> that the social, economic, political and military implications are just
> too mind-boggling to even begin to contemplate.  We cannot know the

Perhaps that's how it started, someone's mind got seriously boggled

> consequences of the development of a true AI.  We just know that they are
> far-reaching and that knowledge is probably always better than no
> knowledge.
>   Thus, after prolonged consideration, we decided to go public.  We needed
> a forum where the information could be disseminated throughout the world
> in as fast a manner as possible, so as to render it impractical for any
> individual, organization or government to make any sort of attempt at
> restraining or censoring our efforts.  We concluded that the Internet was
> the ideal forum. 
>   We do not claim any legal rights to the concepts divulged by us on
> Usenet.  You are completely free to use this information in any way you
> wish.  We just claim copyrights to items clearly marked with a copyright
> notice.  Feel free to electronically reproduce the copyrighted items for
> the purpose of discussions on the net.  All other reproductions must be
> approved by Marengo Media, Inc.
>   We tried to be as brief as the nature of the subject would allow. 
> Indeed, our initial discourse is a lot shorter than one might suspect for
> a subject so vast.  This is because the concepts we describe are simple,
> though they carry enormous consequences.  We also use simple language so
> as to appeal to the widest possible audience.  We will expand on our ideas
> in future posts in response to readers' comments.

Translation: We use simple language because we are ignorant of more
precise technical language.  We will expand our ideas in future posts
because we haven't really worked anything substantive out yet.

>   Note: Our goal is to disseminate this information to as many peoples and
> countries as possible.  If it is the breakthrough that we claim it is, may
> all of humanity benefit from it.  If you would like to help us translate
> some of this text into other languages, please let us know via email.  As
> is usually the case with any issue on the net, we anticipate that this one
> will attract as many detractors as partisans.  So be it.  The beauty of
> this technology is that it is not the exclusive domain of experts.  Almost
> anyone with a passable knowledge of programming can implement it on a
> computer.  If we can get at least one person per country to take these
> ideas seriously, we will have accomplished our goals.
>   We organized our expose into a single 4-part article that can be found
> in this thread.  The 4 parts are:
> 
> I.   Introduction.  The Search for a General Learning Mechanism
> II.  The Building Blocks of Knowledge
> III. Putting the Blocks Together
> IV.  Conclusion.  Toward an Autonomous Intelligence
> 
> 
> ***************
> 
> Machine Learning Breakthrough
> By Louis Savain
> 
> 
> I. Introduction.  The Search for a General Learning Mechanism
> 
> 
>   It is generally accepted that the most sensible way to achieve a true
> artificial intelligence is by putting together a self-organizing system
> that is able to learn all sorts of knowledge on its own from a stream of
> sensory data.

There is much argument about how to achieve the goals of AI, whether it
is possible, even what would constitute it.  Nothing is generally accepted.

> The idea that we can program all necessary knowledge into
> an intelligent machine may or may not be valid, but it does not seem that
> such an approach will lead to a true AI in the forseeable future.
>   One of the main problems with current learning systems is that they
> cannot learn on their own.

If you knew anything about current research in ML, you would know
how stupid this statement is.

>                               They need a predetermined, specialized
> "evaluator" to guide them every step of the way.  You could not put a
> scaled version of any one of these systems in a room full of people and
> expect it to eventually pick up the language, or make sense out of the
> objects in the room.  They may be very good at certain types of
> recognition tasks but they lack generality, common sense and autonomy.

Unlike axis, which has been demonstrated to be capable of doing this?

>   I realized very early on in my research that the important thing about a
> general learning machine is that it must be *general*.

That's really deep.

>                                                        It turns out that
> the more general or flexible a system is, the lower the level of
> abstraction at which it must operate.  A low level of abstraction--within
> the confines that the nature of a system admits--means that the basic
> building blocks of the system must be simple.  Knowledge is no exception. 
> If our goal is to find the basic building blocks of knowledge, we should
> keep our minds firmly focused on the notion of simplicity.

Searhing for simple "basic building blocks of knowledge" is like searching
for the philosopher's stone or the holy grail.  Pointless.

>                                                              It follows
> that regardless of the fundamental constituents we use to build a
> versatile intelligence, they must be so simple as to encompass any type of
> knowledge.  An intelligence could not be general otherwise.
>   Once we have chosen the building blocks and we are satisfied that they
> are simple and general, we need to figure out how they can be put together
> in a learning machine.  Learning is exactly that: connecting building
> blocks together in a consistent manner.

Oh, so that's what learning is?  I'm not sure you'd get too many people
agreeing with that definition.  It seems to admit too many "learning"
systems that are very "consistent" but otherwise uninteresting.

> To accomplish this we need several things:
> 
> 1.  A set of simple building blocks.
> 2.  A simple method for connecting the building blocks.
> 3.  A simple method for detecting conflicting connections.
> 4.  A simple method for correcting conflicting connections.
> 
>   Simplicity is conducive to generality.  It is a consistent theme
> throughout this article.
> 

Reflecting the simplicity of the understanding of the author?

> 
> II. The Building Blocks of Knowledge
> 
>   A little over 15 years ago, when I originally set out to find the
> building blocks of knowledge, I decided to look at several systems to
> gather useful analogies.  One of the not-so-surprising things I noticed
> was that all the systems I examined seemed to be composed of fundamental
> building blocks that occurred in *complementary pairs*.  For examples, the
> basic building blocks of living systems, DNA, consist of the famous
> complementary base pairs; mathematics is based on positive and negative
> numbers;  physics talks about particles and anti-particles;  et cetera... 

Another profound insight - opposites exist.

> Whatever the basic building blocks of knowledge are, there are very good
> reasons to suspect that they, too, should manifest themselves as
> complementary pairs.  Why should knowledge be an exception?
>   Instinctively, I immediately thought of the AND and OR connectives.  I
> am sure many others have taken a similar path at one time or another.  And
> why not?  AND and OR are the darlings of logicians and they seem simple
> enough.  However, I soon realized that although AND can be considered to
> be fundamental, i.e., it cannot be divided into smaller components, the
> same cannot be said of OR.  OR can easily be reduced to other constituents.

Crap.  AND and OR are equally fundamental - they aren't even primary.
Considering the 16 functions of two inputs, from NAND alone (or NOR alone)
you can get everything, but you can't get negation from AND or OR alone.
From AND and NOT you can get everything, including OR.  From OR and NOT
you can get everything, including AND.  This is elementary stuff.

> A further problem was that the choice of AND and OR did not
> conform to the criterion of complementarity, i.e., AND and OR are not
> opposite.  There was obviously a need for a different connective that was
> not only fundamental but whose operation and meaning could be understood
> as being the complement of AND.  As I continued my search, the answer
> became more and more obvious: XOR!

More crap.  Firstly, XOR is not the complement of AND any more than it is
the complement of OR.  Secondly, XOR is not fundamental.  From XOR you can
get NOT, but you can't get AND or OR (think about symmetry w.r.t. negation).

>   Please bear with me.  There are many reasons why AND and XOR should be
> considered both fundamental and complementary.  I will just mention two
> here and leave the rest for future articles: a) AND and XOR cannot be
> further decomposed, and b) while AND is used to detect *absolute* Boolean
> values, XOR is concerned with *relative* values.  Note that Reason (b) is
> in direct conformity with the complementarity criterion.

a) is crap.  A AND B = not ((not a) OR (not B))
	     A XOR B = (A AND (not B)) OR ((not A) AND B)

The "absoluteness" of AND and the "relativity" of XOR might equally be
claimed for OR/X-NOR and several others.  The choice is arbitrary.

>   One of the things that finally convinced me that AND and XOR played a
> fundamental role in knowledge and intelligence, was the observation that
> in every day conversation, human beings almost invariably assume the
> exclusive-or meaning when they use the word 'or' or its equivalent in
> other languages.  More often than not, if the inclusive-or meaning is
> intended, it is qualified with something like 'either this or that, or
> both' to remove any ambiguity.

More crap.  See Aaron Sloman's comments.

>   Logical operators need operands and the operands must also be
> fundamental and complementary.  There is no need to look far for
> fundamental operands: they are the familiar 'yes' and 'no' Boolean
> signals.  Notice the telltale pairings: operator/operand, AND/XOR, yes/no,
> true/false.  There is just no escaping complementarity.

Simple things enthrall simple minds.

>                                                         I will argue that
> the fundamental building blocks of knowledge are the AND and XOR operators
> and their operands.  The reasons will become even clearer as I expand on
> the subject in this and future posts.  Note: If figure 1 below looks
> distorted, try using a non-proportional font in your text viewer.
> 
>    _______
> __|       \ ____ yes
>   | AND    |
> __|        |____ no
>   |_______/
>    _______
> __|       \ ____ yes
>   | XOR    \
> __|        /____ no
>   |_______/
>  
> Figure 1.  AND and XOR cells.  The 'yes' and 'no' outputs are
> complementary.
> 

As someone recently pointed out, what you have here are combined AND/NAND
and XOR/X-NOR gates.  But you missed the point, due to your ignorance of
logic and preoccupation with the mystical significance of AND and XOR.

> 
> III. Knowledge Building
> 
>   The choice of a fundamental set of knowledge building blocks (hereafter
> referred to as cells)

Why couldn't we just call them gates or logic functions?  Because we need
some redundant and suggestive terminology to convince us that at last we
have found the true secret of intelligence.  What a joke.

>                       must not only obey the simplicity and
> complementarity criterion I mentioned above, it must also allow us to put
> the cells together in such a way as to make it easy to detect
> inconsistencies or conflicts.  The error detector (evaluator) must conform
> to the requirement for system-wide generality, i.e., regardless of the
> type of knowledge being examined, testing for errors should work the same
> way across the board.  Otherwise, there would be no end to the complexity
> of the evaluator and our search for a general learning mechanism would be
> in vain.

Do you really think it's possible to have a universal, simple way of
detecting inconsistency/errors?  Have you ever heard of Goedel?
The halting problem?

> Let us suppose we have a network of AND and XOR cells connected
> together in some way.  Is there a general mechanism for error detection? 
> Is there a set of simple tests that can be administered to the connections
> to determine their consistency?  Yes indeed!  There are 2 tests, one for
> AND cells and another for XOR cells.
>   AND and XOR are the only logic cells whose input connections can be
> tested for logical conflicts.  A conflict exists if a cell's input
> condition is such that its input connections behave in a manner that is
> contrary to the purpose of the cell.  An input condition is defined as an
> interdependency between 2 inputs.  If no conflict exists, the connections
> are said to be in harmony.  What input conditions will cause a conflict in
> AND and XOR cells?

"Harmony"!  More unnecessary mumbo-jumbo.

> Definitions (Assuming only 2-input cells for now)
> 
> AND Purpose.  The purpose of an AND cell is to detect when its both input
> connections are true.  (absolute test)
> 
> XOR Purpose.  The purpose of an XOR cell is to detect when its input
> connections are opposite.  (relative test)
> 
> AND Conflict.  The input connections of an AND cell are said to be in
> conflict if both remain opposite after changing simultaneously twice in a
> row.
> 
> XOR Conflict.  The input connections of an XOR cell are said to be in
> conflict if both remain the same after changing simultaneously twice in a
> row.
> 
>   Note that the AND conflict is the opposite of the XOR conflict and vice
> versa.  This is exactly what we would expect from true complementary
> entities.  Here are 2 examples for clarification:
> 
> AND Conflict Example.  If A stands for 'the door is open' and B stands for
> 'the door is shut, the statement 'if A and B is true' is illogical because
> the door cannot be both open and shut.
> 

Well, leaving aside the fact that 'if A and B is true' is not a
statement (or grammatical) presumably you are trying to alert us to the
fact that P AND (not P) is unsatisfiable.  This is the law of
non-contradiction, known since ancient times.

> XOR Conflict Example.  If both A and B stand for 'the door is open', the
> statement 'if either A or B is true' is illogical since the door cannot be
> either open or open.

Mmmm, this is even less convincing than the AND example, if that's
possible.  You're saying that A XOR A is illogical, whatever that's
supposed to mean in this context.  A XOR A = false.  Big deal.

> 
>   In the conflict examples above, opening *and* closing the door would
> trigger a conflict.  This illustrates what I mean by "changing
> simultaneously twice in a row."
>   
>   This is all very simple.  Any type of knowledge can be created with AND
> and XOR cells.  The knowledge consists entirely of the learned
> associations between the cells.  It is completely non-symbolic and does
> not suffer from any of the problems associated with symbolic knowledge
> representation schemes.  Also, the logic of any parallel/distributed
> application can be implemented using only AND and XOR cells.  Detecting
> logical conflicts in such programs thus becomes a simple proposition as
> anyone with an elementary knowledge of programming can surmise.

Remarkable, isn't it, we've gone from the most trivial of examples to
the claim that everything is now solved, there are no problems, we can
even implement "the logic of any parallel/distributed application".
Without even having to talk about synchronisation, concurrency, etc.
And somehow, magically, it's also become "completely non-symbolic",
despite the fact that a minute ago A stood for "the door is open".

> The
> simplicity of the theory is such that it can be easily implemented using
> object-oriented programming techniques.  To implement a minimal
> hypothetical system, we would probably need the following software
> objects:
> 
>     Network Objects
> 
> 1.  AND and XOR Cells.
> 2.  Cell Manager/Processor.
> 3.  Sensors and Effectors.
> 4.  Sensor/Effector Manager.
> 5.  Connectors.
> 
>     Searcher/Learner Objects
> 
> 6.  Connection Manager. 
> 7.  Connection Maker. 
> 8.  Conflict Detector.
> 9.  Conflict Resolver.
> 
>     Miscellaneous Objects
> 
> 10. Object Database.
> 11. Cell/Connector Probe.
> 

Probably need? Yet more groundless hypothesizing and speculation.

> List 1.  Software objects needed for a minimal learning machine.
> 
>   I will explain only aspects of the objects listed above that are not too
> obvious.
> 
> -  The cells should be arranged in some sort of array preferably 3-D.
> 

Some sort of array?  Preferably?  Why 3-D?  Why not a hypercube?
Do you know anything at all about network topology?  It appears not.

> -  The Cell Manager/Processor is responsible for 'firing' the cells when
> their inputs are satisfied.
> 

In what order?  How do you handle loops? Race conditions? Do you guarantee
fairness?  Do you have any idea what the concept means?  And what precisely
do you mean by 'firing' and 'satisfied' here? Is this an asynchronous
system?  A clocked system?  Do you know the difference?

> -  Connectors are used for connecting cells together.  The important thing
> to note about connectors is that they have 'connection strengths'.  The
> strength of a connector is given an arbitrarily low integer value on
> creation.  An AND/XOR network can be regarded as a belief system and a
> connection's strength can be viewed as a measure of the degree to which an
> assumption is held.  The strengths are managed by the Connection Manager
> and used by the Conflict Resolver.  Note that, unlike neural networks, the
> imbedded knowledge has nothing to do with the strengths (weights) of the
> connections.  The connection strengths are used exclusively for conflict
> resolution.

"can be regarded as"! How exactly?  "can be viewed as"! How exactly?
As explanations go, this is really pathetic.

> 
> -  The Connection Manager should use a simple method for assigning
> connection strengths:  in general, the older the connection, the stronger
> it is.  There is another slightly more complex method which I will discuss
> in a  future post.

You really don't know anything about machine learning, do you?
There are obvious cases where this naive approach won't converge.

> 
> -  The Connection Maker can be as simple or as sophisticated as the scope
> of a project requires.  Let us assume a 3-D network of mixed AND and XOR
> cells.  Sensors are connected to one side of the network and effectors to
> the opposite side.  We can easily conceive of a simple Connection Maker
> that would traverse the network periodically and make semi-random
> connections between cells.  The connections can be chosen so that the
> direction of the flow of data through the network is from sensors to
> effectors.  Of course, there is no need to connect the inputs of one cell
> to the output of an unconnected or dead cell.  This forces the wiring of
> the network to build up from the sensor side to the effector side.  In
> future posts I will write about more complex forms of searching including
> a process called 'search adaptation' or 'learned/guided search', which has
> to do with attention focusing.
> 

Ah yes, semi-random search amongst the space of possible connections,
that's going to be very efficient ...  But perhaps you'll address
problems of efficiency and convergence in those promised future posts.

> -  The Conflict Detector or evaluator constantly surveys cell inputs for
> conflicts.  If a conflict is found, the culprit is marked as such.  Its
> address might be inserted in a conflict list for use by the Conflict
> Resolver.  In a large project, the Evaluator should work in cooperation
> with the Connection Manager, the Connection Maker and the Conflict
> Resolver so as to focus the intelligence's attention on one particular
> area at a time.  Indeed, it is a good idea to combine items 6, 7, 8 and 9
> in List 1 into a single cooperative object called the Searcher or Learner.

This is getting really clear.  Yet more mumbo-jumbo names for components of 
the system whose purpose you have blurrily described, but whose actual
functioning is still as mysterious as ever.  I don't think you'd know what
an algorithm was if it bit you.

> 
> -  The Conflict Resolver's job is to resolve conflicts by disconnecting
> weak connections first.  The idea is that young, more recent connections
> are more likely to be the offenders.  If disconnecting a particular
> connection causes the conflict to disappear, the conflict is considered
> resolved.  When designing a conflict resolver, it is advantageous to keep
> in mind that a connection located 'upstream' from the conflict location is
> much more likely to be the offender than a 'downstream' connection.

Err, excuse me, what about loops?  Define 'upstream' on a cyclic network,
if you can.  Do you know what a bistable latch looks like?  Or perhaps
you'd like us to restrict ourselves to acyclic feed-forward networks,
but forgot to mention that.

> 
> -  The object database is used to save all objects at various times or on
> termination.
> 
> -  The Cell/Connector Probe is an optional tool, preferably visual, that
> an experimenter can use to dynamically observe the state of one or more
> cell and its connectors.
> 
> 
> IV.  Conclusion.  Toward an Autonomous Intelligence
> 
>   It is safe to bet that this is not what many workers in the field expect
> intelligence to be.  We apologize.  Proponents of genetic algorithms and
> neural networks can take solace in the fact that an AND/XOR learning
> system does incorporate a little bit of both approaches.

It is a safe bet that;

a) you haven't really done anything useful with axis, or even really
constructed a working system, let alone establish any of your fanciful
claims,

b) your ideas are so completely underdeveloped, vague, and muddled that
it will be years, if ever, before this approach can solve even simple
toy problems, let alone useful problems that are *currently* within
the scope of *existing* machine learning techniques.

If I'm wrong, it shouldn't be hard for you to publicly demo an axis
system that learns to control an inverted pendulum by say, thanksgiving.

> Let us have a
> serious but friendly debate.  No flaming and other hate stuff please.

Undoubtedly you will consider this a flame, but don't delude yourself
into thinking it's motivated by hate.  Just by an intolerance for
ignorance, and an unwillingness to see you parade around in the
emperors new clothes without saying something about it.

>   There are many internal and external issues that this short discourse
> cannot even begin to deal with.  I will just mention a few here: 
> 
>     Guided learning through 'reward and punishment'.
>     Learning how to search.
>     Attention focusing.
>     Incomplete or fuzzy knowledge.
>     Recognition tasks.
>     Multiple (more than 2) input cells.
>     The AND/XOR belief network and analogies.
>     Reasoning.
>     Goals.
>     Comparisons to the human mind.
>     The mind-body problem.
>     Consciousness.
>     Asimov's robotic laws.
>     The crazed robot problem.
>     Social implications.
>     Economic implications.
>     Military implications.
>     Legal implications.
>     Medical implications.
>     Moral implications.
>     World wide implications.
> 

You forgot to mention the meaning of life, a cure for aids, world peace.

>   There are many more. Though this is enough to last us a long time, feel
> free to add more items to the list for future discussion.
> 
>   Does this all mean that anyone can go home and create a human-like
> intelligence on a desktop computer?  Of course not.  Human intelligence is
> just too vast.  A comparable intelligence will require lots of computers
> wired into a network and years of work.  This does not mean that we cannot
> use the information given here to create useful applications and
> experiments right now.  Indeed, an immediate benefit of this technology is
> in the design and implementation of safety-critical applications used in
> areas such as air-traffic control systems, power plants and radiation
> therapy machines, to name a few.  This was the basis for the formation of
> Marengo Media, Inc.  In fact, we claim that the reliability of any
> software application based on this technology, will actually increase with
> logical complexity!  In other words, the more complex the application, the
> more reliable it becomes.  This is the exact opposite of what one would
> expect from traditional development environments.

This is exactly the opposite of the reality of complex systems.  There
are no magic bullets for complexity, no simple solutions.  As someone who
has worked on many complex systems, including air-traffic control, I can
assure you your claims to the contrary are just wishful thinking.

> 
> Note.  To encourage research in AND/XOR intelligent systems (AXIS), we are
> currently working on the preliminary rules for a couple of international
> programming contests:  One is for the fastest learning tic-tac-toe program
> that learns to play a correct game using AND/XOR cells.  The tentative
> Prize amount is 1000 US dollars.  We chose tic-tac-toe for the first
> competition because of its simplicity.  The second is for the first chess
> program (based on AND/XOR cells) to achieve expert level in a refereed
> match.  We have not yet decided on a prize amount for the chess contest. 
> We are also thinking about a GO contest.  We will post the preliminary
> rules for the tic-tac-toe competition in the near future in this thread. 
> Stay tuned.  We are open to suggestions as to rules, etc.  Please post
> them in the "Re: Machine Learning Breakthrough" thread.
> 

Ahhh, yes, having contributed these profound insights, you're going
to leave it to others to do the research for you.  And if they can't
quite fulfill the glowing prospects of this breakthrough technology,
it'll be because they haven't understood properly the purity of your
vision.  At least you haven't started asking anyone for money yet.

> Best wishes,
> 
> Marengo Media, Inc.
> 1107 Fair Oaks Ave, Suite 462
> South Pasadena, CA 91030
> USA
> 
> Louis Savain       Curtis L. May II
> Chairman           Vice President 

Go away, try to forget all the mystical profound jargon shit, and try
to learn something about logic, complexity, and existing machine
learning techniques.  Come back when you've got a few clues, a bit
of humility about the significance of your contribution, and you've
actually got some real results.  Meanwhile, don't go anywhere near
any safety-critical systems, and don't claim to be an expert on anything.

>  
> PS. Please realize that we do not have the resources to handle a high
> volume of email or phone inquiries.  Unless otherwise clearly noted, post
> *all* related communications in the "Re: Machine Learning Breakthrough"
> thread in comp.ai.  We will post to comp.ai only.  Thank you.
> 

Perhaps we could create comp.ai.marengo.  If we did, would you promise
to move this thread there?  I suspect not.

> Copyright 1995 Marengo Media, Inc.
> 
> 
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Copyright (c) 1995, David Kinny/Arrogance Unlimited.  All rights reserved

