1. Information Technology
and Information Science for Tomorrow
By
Tom Adi and Ken Ewell
Management Information Technologies, Inc.
A Seminar and Demonstration
Presented to
The Swedish Council for Job Security
in
Santa Clara, California, USA
October 6, 1987
Unpublished Work, Copyright 1987 MIT, Inc.
All Rights Reserved.
No part of this document may be reproduced in any form
without the written permission of MIT, Inc.
2. 1. Introduction
We have been selected to present this seminar because we have successfully
developed what is being recognized as a new information technology. This new
technology, which we call READWARE(TM) in consideration of the terms
HARDWARE and SOFTWARE, signals a significant change in information
technology. It may have a measurable impact on industry and society in the very
near future.
READWARE(TM) is a new approach to Information Science, whereupon the
faculty for the comprehension of natural language text is added to a computing
machine. One software implementation which you will see later is based on our
discovery of Letter Semantics. Letter Semantics is a theory of natural language
meaning. It deals with understanding natural language texts, human knowledge,
and intelligence, with and without the aid of computers.
Some of the laws and principles of Letter Semantics have been reduced to elegant
mathematical algorithms and implemented in READWARE(TM) software which
can be run on machines as small as personal and desktop computers. In this form,
the technology offers assistance to intelligent people in understanding texts and
analyzing knowledge extracted from those texts, thus, improving both knowledge
and intelligence of its users without ever attempting to replace their human
activity as masters over machines.
Later, we will demonstrate the READWARE(TM) Research Assistant which is
one of the first software applications of our new technology. It can associate the
concepts represented in a string of natural words in Swedish, for instance, to the
concepts contained in a string of words in another language, like Arabic, English,
French, German, Russian, or Spanish. This association, an unrealized dream of
Artificial Intelligence (AI) technology, is based entirely on Letter Semantics, i.e.,
the analysis of words without the aid of dictionaries and merely on the basis of
what letters of alphabets and their combinations mean.
Allow us to digress for a moment, as we would like to stress our independence
from the commercial AI industry which aspires to mechanize intelligence and
replace intelligent humans. We neither use their methods nor share their views.
Our technology provides a cognitive interface between the human and the
machine. In this way, the ability of the machine is extended one step beyond
symbolic recognition, to concept apprehension and comprehension. We did not
fail to understand human intelligence like the leading authorities of Artificial
Intelligence and Computational Linguistics admit to.
Terry Winograd, a Stanford professor and a leading authority of AI, has criticized
the very promise of AI in a recent publication. In it he states: "There is a tacit
3. acceptance of the point we have made in this book--that the techniques of current
AI are not adequate for an understanding of human thought and language."
Winograd "strips" existing AI products and concepts of their "intelligence," one
by one, and in detail, including his own famous system SHRDLU. This includes
products of Computational Linguistics used in natural language front-ends for
databases, as well.
Regarding "robots" he makes an interesting point: ". . . it is important to separate
out the real potential for such devices from the implications that come from
calling them applications of artificial 'intelligence,' and even from the use of the
word 'robot.'"
Winograd suggests with his co-author Flores "a new foundation for design" (the
under title of their book), which is rather a philosophical speculation about what
to do after the failure of AI.
Our company, Management Information Technologies, Inc., has done more than
speculate, we have achieved a new foundation for design as is evidenced in the
READWARE(TM) implementation, which culminates a decade of empirical
research and several years of development and implementation effort.
2. Why AI Failed
We believe that there are two reasons for the failure of AI in addressing language
and cognition. The first is a false linguistic view of natural language as an
evolution of sounds with no relation to meaning. The famous linguist Saussure is
one of countless linguists who firmly believe that words, which are seen as
sounds, relate arbitrarily to their meanings.
This was a double error. Language is not even a matter of sounds. At the
beginning of this decade, "revolting" linguists searched in vain for symbolisms
(meanings) of sounds. We have found out that meanings are in letters rather than
sounds. For example, "seize" and "cease" sound the same, but are different in
meaning. Language change over the centuries makes sounds highly ambiguous.
The linguistic community had to stop all research in that direction.
Linguists still claim, from time to time, that they have a theory of meaning, a
natural language semantics that they operate with. Recently, MIT's famous
linguist, Chomsky, concluded that existing theories of meaning are not semantics
at all, but rather speculations over mental representations. In his view, semantics
must explain the direct relations between language and the world. Linguists
cannot find semantics because of the linguistic dogma that words relate arbitrarily
to their meanings.
4. In contrast to this, we started our research with a belief in the meaning strength of
words in human languages. Further, we based our research on a very special
language, often revered by linguists all over the world, a language where sounds
correspond unambiguously to letters and no language change has taken for over
fourteen centuries: Arabic. The results were verified for many other unrelated
languages. It took less than one month to add Swedish.
The second reason of AI’s fall is its stubborn attempt to derive the science of
cognition from an information science based on physics, believing that humans
are just complex machines. It turns out that AI is not alone in the failure to
advance, but is rather part of a collapsing system of sciences. Physics is assumed
to be the mother of all sciences such as chemistry, biology, and even psychology
and information science, as depicted in Figure 1.:
Fig. 1. Old Hierarchy of Sciences
?? Physics (measurements, study of matter)
Mathematics--+--Information Chemistry Other
| Science | Sciences . . .
+----->-------------------------------------- . . .
This rather materialistic view of the world has produced confusion in science and
technology in the long run. Knowledge is assumed to come from matter by
undefined means, and the human role in knowledge acquisition is obscured by
assuming this hierarchy.
Mathematics seems to come from nowhere but it dictates itself everywhere.
Mathematics has grown into monstrous complexity dragging behind a physics that
is not only complex, but one full of contradictions and unresolved dis-harmonies.
Nowadays, the new buzzword "broken symmetry" is used as an excuse for
insisting contradictions in natural sciences.
The scientific quarrel began with Einstein rejecting the quantum theoretical view
that "God plays dice with the universe." It continues today after many failures to
reach a unified theory of forces in nature. Loopholes in science opened for
different theologies that offer "alternatives" such as "Creationism" instead of
"Evolutionism." Atheists utilize speculations about the physical vacuum to
suggest that "everything comes from the Big Nothing." Last, but not least, the
fallacy of sound symbolism discussed above also stems from this materialistic
view of sciences.
5. We do not believe that religious or anti-religious groups can overtake scientific
research or dictate knowledge; their inherent dogmatic approaches will not allow
it. But what happened to the elegance and harmony of science; where is scientific
objectivity? Where is that fascinating crystalline stuff from which the good old
physics was made? Is confusion the price of progress?
3. Restructuring Sciences: The Intellectual Dimension of READWARE(TM)
Our discovery of Letter Semantics led us to revise the above mentioned
materialistic world view and hierarchy of knowledge. We also found a more
meaningful place for mathematics in the hierarchy of sciences, as depicted in
Figure 2.
Fig. 2. New Hierarchy of Sciences
Language (alphabets, names of things and events, texts)
Math Physics Chemistry Information Other
| | | Science Sciences . . .
+----->--------------------------------------. . .
Letter Semantics (illustrated in Figure 3) means that our knowledge comes from
the names of things. Every letter in the name of a thing is an abstract hint about
that thing; the sequence of the letters hints to the structure of that thing. The name
of a thing is a collection of hints, a puzzle that motivates us to touch and study
that thing. Knowledge Acquisition is the process of following those hints and
making links to natural facts; and Intelligence is the ability to do so. Finally,
human language is natural, i.e., the names of things are supplied by nature and not
invented by humans. Language change may falsify the natural hints and
introduce ambiguity; on the other hand, language change can result from human
ambiguity and then amplify and perpetuate that ambiguity.
6. Fig 3. Letter Semantics & Knowledge Acquisition
by Intelligent Humans
<-----Nature <-------Nature-------->
Thing- ->Name Thing - ->Name Natural Laws
| | | |
| | Abstract hints- - - -> Residual |
| | Concept clusters - - -> hints |
| | | | |
| | | | Closed | Recognition
| | | | links | more
| | | | | certainty
| | |
? Facts---->
Innate, inherited Acquired knowledge
primitive knowledge
All sciences originate and grow from human language in that the names of things
motivate and guide research. We can use pure language to formulate scientific
results. We can also derive from the names of things, abstract concepts (inherent
in letters and their combinations) which crystallize in the form of mathematics.
We can use mathematics to advance any science because mathematics comes from
language, and language deals with everything. This explains the mystery that
mathematics produces far more results than the stuff you squeeze into it. You are
surprised by mathematical results in the same way you jump up when you get a
new idea from the old words in your mind.
This discovery has opened up a whole new world of possible perceptions and
perspectives. Before finding chemical elements we had alchemy. Before the
insights of astronomy people believed only in astrology. We are already beyond
the astrology of artificial intelligence, and we are certain that as this discovery
becomes known and generally accepted, it will cure long standing problems and
lead to many new discoveries. The difference now is that we have a fundamental
discovery in cognitive science (the science of knowledge) that will advance all
sciences and lead to their unification.
We have found out what knowledge acquisition means and what intelligence is.
The question now is: which aspects of knowledge acquisition can be mechanized
and how intelligent can a computer be?
7. 4. Future Impacts on Intellectual Jobs and Activities
Returning to Figure 3, in the transfer from left to right, in the process of
knowledge acquisition, we notice that some abstract hints solidify into links to
natural facts, and some don't. As far as we can see, there will always be residual
hints that are not resolved. This means that there can be no complete knowledge,
neither by humans nor by machines. But the natural feedback, the recognition of
natural laws in the facts attained by solidifying natural hints, increases our
certainty about facts.
We decide whether and how two different hints are related. The most primitive
rule states that the hints from one letter of an alphabet are always related. We
developed powerful and elegant rules regarding this issue that we call our Theory
of Associations. Our READWARE(TM) Research Assistant applies some rules
of this theory to make associations between two expressions in one or two
different languages. This type of application is called automated associative
research.
Another possibility is to analyze existing solid links expressed in free text: to
summarize, generalize, test hypotheses, and otherwise extract information. The
output might surprise us, but there is no hint resolution, there is no core
intelligence in that. This is the subject of ongoing research and development at
MIT, Inc., under the title of automated knowledge extraction.
A future research project, now being conceptualized at MIT, Inc., deals with
studying hint resolution to determine laws of core intelligence, so that hints can be
resolved by guessing and may be tested against the natural laws we already know.
We call this automated knowledge acquisition and it is in fact our most
ambitious research goal and may lead to very powerful computer tools. For
example, with this technology it would be possible to suggest new inventions,
natural laws, or theories, to be checked out by human experts; or to perform
unmanned research on far planets. This type of machine assistance would deserve
to be called artificial intelligence, but we will not use that term any more.
For example, imagine a manager who can supply an idea to a machine that will
read millions of pages of text in minutes and automatically reply with opinions
and arguments supported by extracts from those texts. This is something that
humans are now unable to do. But our existing technology can already achieve a
major part of such a science fiction type analysis. In a short time, this complete
and swift analytical processing will be commercially available. The manager in
our example will still have to judge and decide, but the machine has already saved
him immense amounts of reading and comprehension.
If your job is only to collect literature for your boss and to do some preliminary
analysis, you might be replaced by a program. But you can alternatively increase
8. your capability by mastering the usage of such technology and you may then
replace your boss. You will need less memorizing and a "smaller brain," so to say;
because it's the quality of intelligence that counts, not the quantity of facts that
you can memorize.
Expanding on this prospective, we expect people who are now regarded as less
intelligent because they do not master some stereotype of knowledge, to have new
motivation for learning. The user of this new technology is able to control the
efficiency of, say, knowledge extraction, by using his very personal core
intelligence that his boss does not possess. Use of this technology enhances
human intelligence in ways that schools of today are unaware of.
Information in free text, whether scientific, literary or technological, will be
analyzable by people all over the world, many of whom are regarded as less
intelligent nowadays. Education will change because an intermediate school
student will be able to analyze college literature (with the aid of a computer) on
the basis of the powerful and yet simple universal concepts he or she learned in
elementary school, which are directly derivable from his or her language and
spelling lessons. School time and years should be shorter because if we
understand what intelligence is, we can teach it much faster. Sciences that are
complex and so anti-human can now become simple, elegant, and more
acceptable to the human soul than some of the badly structured and monstrous
scientific complexities of today.
The hostility between machines and nature, between robots and people, should
slowly vanish, because computers will increasingly be used to understand nature
and deal with it. Fitted with our technology, the computer can put a couple of your
ideas together in a natural way instead of forcing you to "think" like a "robot" and
degenerate into mechanical logic like some future visions suggest. It will enhance
human intelligence instead of burning it out. Whatever your IQ might be, it will
help you promote yourself while working with it as a tool. Complex machines will
become abominable and much less useful in an era of clarity of intelligence.
Bureaucratic machines, whether human or mechanical, will be able to receive and
process human information, as well as the fixed relational data and statistics used
exclusively today. Such restricted data presents very narrowed views on a subject.
The capability to process human information in natural (written) discourse will
expand our perspectives and heighten our horizons.
Courts will work much faster because legal analysis will profit from this
technology, but neither judges nor lawyers will be less in number. The judgment
process can be mechanized, but it requires genuine human hint resolution
(personal core intelligence) to validate and check out every case.
Psychology and other social sciences will definitely profit from understanding the
9. relation between what people say and what they conceive, which is a direct
application of Letter Semantics. In this way, those children who now drop out of
society, those who become incarcerated, or become a burden or a danger to
society, can be reached, and saved, through better intellectual understanding.
Further, we know from past experience that language change over the centuries
modifies our world views and that different languages each have their advantages
and disadvantages. The approaches to language processing to date, including
language translation, are simply less productive ways of communicating. We
believe READWARE(TM) is a technology for a Global Community.
Throughout history and today, cross-cultural and cross-lingual sharing of ideas
and experience is seen to be most fruitful. America owes much of her scientific
and technological power to the diversity of lingual backgrounds, to an extent that
we are just beginning to understand. Any technology that is able to aid us share
ideas in such a fashion should be of great benefit to society in general. This
premise became the basis of design for the READWARE(TM) Research
Assistant.
READWARE(TM) technology offers the means of putting together textual
information in different languages to be analyzed interactively.
READWARE(TM) technology utilizes letter-semantic clusters of concepts,
information that is automatically derived from natural texts without the aid of
dictionaries. This universal information can be used in an interactive environment
in many different ways; in ways which we have yet to learn about.
The term READWARE(TM), aside from its kinship to the terms HARDWARE
and SOFTWARE, expresses two facts: first, that knowledge is acquired by
READing and second, that something is there, a -WARE, a computer or a
methodology that helps you understand what you read.
The computerized applications of READWARE(TM) are very efficient
interactive tools. We are also preparing READWARE(TM) COURSES for
enhancing the intelligence and analytical capability of those who have to do a lot
of thinking on their jobs. The first course to be offered will be called
READWARE(TM) PHYSICS and will introduce physicists (and those who need
to think on the basis of physics) to a special branch of Letter Semantics
concentrating on concepts of physics.
Currently, we are ready to prepare a CUSTOMIZED READWARE(TM)
COURSE for any group of people interested in enhancing their cognitive-
analytical capabilities in a special area of research. People teaching English to the
deaf have already shown an interest. You know that the deaf associate a single
object, not a concept, with a certain word. If the deaf were taught Letter
Semantics, they would be able to make concepts out of words and thus benefit by
10. new knowledge derived from their chosen language.
READWARE(TM) helps its user to extract and relate acquired knowledge from
intelligent writings. We as humans cannot hope to attain perfect knowledge, yet
we can hope to achieve a valid definition of the process of intelligence in order to
acquire knowledge. We believe that we have achieved this result. We determined
that the best way to demonstrate machine-aided intelligence was to develop a
system that would reproduce the ability of the comprehending human reader.
Finally, to distinguish our Technology from the Falling AI again, we point out that
AI has limited its endeavor to mechanizing humans and enforcing dry logic onto
their thinking. We, on the contrary, derived an understanding of human thinking
from natural language, the human's very special gift. We found that progress can
be made by associations with nature, by understanding its words and learning
from them, by refreshing our ways from language, rather than paralyzing our
energy into the rules of logic.
We have not created artificial intelligence, but have rather imitated parts of
natural intelligence that we discovered by empirical experiment and analysis.
5. How does READWARE(TM) really work, and what do we mean by Read?
Any machine readable information, i.e., any text in its original form in English,
Spanish, German, French, Russian, Arabic, or Swedish, can be accessed by
READWARE(TM).
READWARE(TM) reads your text word by word and letter by letter and
generates a file (one third its size) that contains the abstract ideas of your text in
compiled clusters of concepts.
You can then enter a free expression (in any of the seven languages) that contains
an idea, a question or any combination of words that expresses what you are
looking for. READWARE(TM) will highlight very quickly the spots of the text
that are meaningfully associated with your input. What you get is found by
"machine thinking" and will often surprise you and add to your knowledge. It
must be pointed out that this is Idea Search, in contrast to Keyword Search. Words
are not marked and indexed; no manual per-processing is required of printable
files. Information is not retrieved by keyword, rather by associated ideas, or more
precisely, primitive concepts. A word like pessimism used in an input expression
will retrieve concepts like gloom, doubt, depression, grief, sadness, sorry, trouble,
etc. . . . if they appear in the text, but the word pessimism need not appear.
Pessimism would have to appear in the text to be retrieved by Keyword Search,
whereas none of the other words would be retrieved.
This is one difference between the applications of Keyword Search and those of
11. Literary Research for which READWARE(TM) has been designed. It is a
decisive difference, but it is not the only one; READWARE(TM) is really
"something else."
READWARE(TM) uses four new and powerful analysis modules:
• nineteen (19) Morphological Analyzers (for seven languages under different
options) which detect and analyze prefixes, infixes and suffixes without
using any dictionaries
• seven Etymological Adapters that take care of such issues as Language
Change and bad Transcription
• seven Letter-Semantic Conceptualizers that construct the compiled clusters of
concepts from the letter combinations in word stems
• the Cross-Lingual Associator that performs a lot of thinking under the control
of your input. It is so powerful that an idea in Arabic will relate properly to an
English idea, both expressed in free language
The READWARE(TM) RESEARCH ASSISTANT has been designed for use as
an intellectual aid to any human performing: Situation Analysis, Literary
Research, and Associative Information Retrieval. Compelling proof to our theory
is that the software can be applied to any information regardless of vocabulary, in
any combination of seven eastern and western languages. The program
understands expressions in full text context and in concordance with the situation
being described. Texts from magazines, books, business reports, business letters
and memos, legal contracts, corporate chargers, operating procedures, litigation
proceedings, regulations, and military specifications have been tested, along with
ancient texts, religious text, theological doctrine, philosophical thought, and
recent scientific documents regarding discoveries in the fields of quantum
mechanics, particle physics, and superstring theory.
The READWARE(TM) RESEARCH ASSISTANT includes multilingual
inquiry editors and simple to use tools that help you to save standard inquiries to
apply over and over again.
Session results can be saved for reporting and manual analysis. A screen based
horizontal and vertical cut and paste mechanism is provided for marking rows or
columns of associated text for subsequent pasting to new (ascii) files, to saved
inquiries (QBE, query by example), or for printing.
A single inquiry or a series of queries about information contained in 1 to 1000
files in the same or different languages and (DOS) disks and directories can be
loaded and processed simultaneously. Inquiries loaded from disk have their results
(session results) and their environments automatically stored; otherwise, session
results can be saved at any time by halting the search and pressing a function key,
or on completion of the session by pressing the same function key.
12. A Zoom feature allows the User to zoom into the text files, mark parts of it to
print or save, or mark an idea and make it the basis of a further inquiry by pasting
and editing.
The READWARE(TM) RESEARCH ASSISTANT is completely screen
oriented with all program operations and User options are fully explained and
accessible from the menu strips on each screen. The program makes use of the
software standards and conventions recommended by Microsoft Corporation for
MS-DOS based IBM-PC type machines and their function keys. The User
interface is straight-forward in both function and design and, in our opinion, is the
epitome of simplistic operation. The Objective of the design was to reduce the
amount of computer (if you know what we mean) between the human, and the
knowledge represented in his information base.
We believe that Letter Semantics is powerful enough to classify universal
language structure. The READWARE(TM) RESEARCH ASSISTANT is only a
first step in this direction. Through our concept validation tools and the
READWARE(TM) RESEARCH ASSISTANT, we can demonstrate how
elegantly we cope with the richness of human expression. The experts would
agree that any theory that is powerful enough to cope with the richness of human
expression should be powerful enough to deal with other aspects of cognition and
cognitive processes as well.
The basic difference between the computer systems of today and those of
tomorrow will be the difference between the highly structured systems that deal
with data in terms of logic and rational, and unstructured systems that deal with
data in terms of concepts, associations, and dynamic relationships like most of us
non-mechanical humans. The AI people cannot deny that their research concerns
robotics. Thinking humans should not be deceived into believing that what is
good for robots is good for humans.
When we understand and accept the concept that perfect knowledge is beyond our
apprehension. When we focus solely upon consistency and certainty. Perhaps
then, only the barrier of progress, and fear of change, will stand between humans
hungry for new knowledge, and intelligent machines helping humans to evolve
intellectually, by free choice of association rather than by per-programmed
mechanics.
Now that you know a little about the underlying technology and the design
features of READWARE(TM), we can proceed with a demonstration of the
READWARE(TM) RESEARCH ASSISTANT.
REFERENCES:
13. Adi, T. (1985). Letter semantics. Unpublished manuscript.
Adi, T., & Ewell, O.K. (1986a). Natural Language is not Artificial. Introductory
paper. Washington, D.C. Management Information Technologies,
Incorporated.
Adi, T., & Ewell, O.K. (1986b). READWARE(TM). Internal report. San Jose,
CA: Management Information Technologies, Inc.
Adi, T., & Ewell, O.K. (1987). Artificial Intelligence versus Machine
Comprehension. INTEC87, Washington, D.C.
Chomsky, N. (1985). Knowledge of language: Its nature, origin, and use (pp. 15-
50). In Anshen, R.N. (ed.). Convergence. New York: Praeger.
Saussure, Ferdinand de (1966). Course in general linguistics, Charles Bally et al.
(eds.). Translated by W. Baskin. New York: McGraw-Hill.
Winograd, T. & Flores. F. (1987). Understanding computers and cognition: A new
foundation for design, (pp. 54-106). Norwood, NJ: Ablex Publishing
Corporation.