COMMENT ⊗   VALID 00003 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.<<work[w86,jmc]		Computers and intellectual work
C00018 00003	Expanding the right of initiative
C00027 ENDMK
C⊗;
.<<work[w86,jmc]		Computers and intellectual work
.
.Aids to intellectual work
.
.1. what is beyond word processing
.
.2. keeping track
.
.3. cbcl
.
.4. increasing the efficiency of bureaucracy
.
.5. improving the right of initiative
.>>
.require "memo.pub[let,jmc]" source;
.cb "Computers, artificial intelligence, and intellectual work"
	
.cb "by John McCarthy, Stanford University"

	Computer technology has already affected intellectual work, but
mostly in peripheral ways.  More fundamental changes are in store
for us.

	At present we have word processing, we have electronic mail and
we have databases.  All of these are unevenly available.

Word processing and document preparation

	I am preparing this paper at home using a terminal connected
by private telephone line to a computer at Stanford University.
It is one of several thousand files in the computer that I can
look at and modify at any time.
When I get done writing it, I will put in type-setting commands
and print it.  I have had a terminal at home since 1965, but only
in 1986 did I get a good quality printer at home capable of many
fonts and sizes of printing.  I suppose most people who write
professionally in the U.S. have some form of word processor now.

	While word processing facilities can be improved, greater effects
will come from making them universally available than from the
improvements.

Electronic mail

	Our laboratory has used electronic mail since about 1970
and has been connected to the ARPAnet since 1971.  The ARPAnet
and the other networks connected to it allow me to exchange
electronic mail in minutes with people all over the U.S. and
as far away as Japan, Sweden and Israel.  However, the coverage
is spotty.  There are many people at my own university with
whom I still can't communicate electronically.

	The requirement is to allow all systems to communicate
with each other so that electronic mail becomes as universal
and simple as the telephone.  The key to this, in my opinion,
is to use the telephone.  For the great majority of messages
telephone connections maintained for short times are cheaper
than the fixed networks generally in use.  These fixed networks -
ARPAnet, etc. - are basically a mistake.  Using direct telephone
connections would have been cheaper and more general - and it
still is.


Databases and libraries

	Thousands of databases all over the world are available,
but most of the potential users find them awkward to use.  I have
used the on-line catalog of our university library, the Mathfile
database of reviews of mathematical papers, the Books-in-Print
database, a pharmaceutical database to determine the properties
of a medicine prescribed for me and a few others occasionally.

	What will revolutionize many kinds of intellectual work
is an electronic public library - e.g. the U.S. Library of Congress.
Even counting the cost of conversion, it is now cheaper to keep
books on magnetic disk than on paper, and they can be accessible
all over the world.  A pilot project for an electronic public
library is being carried out in Marseilles under the direction
of Michael Griffiths.

	It will be such a pleasure to get rid of ninety percent
of the paper in my home and office and be able to follow any
reference to a book or article just by typing its name on
my terminal.


Electronic publishing

	When there is an electronic public library, there will
be a revolution in publishing, because an author or even a would-be
author can publish a book by putting it in the database and paying
a few dollars a month disk rental for keeping it there.  Authors
will then be independent of publishers.  However, there is another
important advantage.

	When something controversial is published, those who disagree
will be able to publish their rebuttals immediately, and a reader
of the original who asks the library about a rebuttal will be able
to find it immediately.  This will have a powerful effect on public
debate, because anyone who writes something controversial will have
to count on his readers looking for an answer and will have to
write it so as to anticipate possible answers.  At present there is
too much hit-and-run controversial writing, because the author
is not kept honest by the prompt appearance of answers.


Computerization of bureaucracy

	I don't know whether the organizers of this meeting consider
bureaucratic and administrative work to be intellectual.  I shall.

	While the productivity of industrial work has greatly increased
in the last few centuries, the efficiency of bureaucratic work has
improved much less, and that mainly in its mechanical aspects via
typewriters, printing presses, xerox machines, word processors and
electronic mail.  As a result the number of lawyers, accountants,
purchasing agents, and large and small clerks of all kinds has
multiplied.  Some people regard this proliferation as inevitable
and even desirable, referring to the "information revolution".
In my view what is referred to as the information revolution is
often just bureaucratic inefficiency.
Further increases in standard of living will require improvements
in bureaucratic productivity.  Since governments (and universities)
have little motivation to improve their administrative productivity,
as new tools become available, they will first be adopted by
companies in very competitive businesses where those who adopt them
promptly will survive and others won't.

	In my view there are three keys to increasing bureaucratic
productivity - automation of bureaucratic operations, expert systems
and substantive communication between computer programs belonging
to different organizations.  Each of these is a larger subject
than there is space or time for in this article.


Automation of bureaucratic operations

	Some airlines, e.g. PSA in California, have introduced machines in
their terminals that communicate directly with their reservation systems.
I can put my credit card in the machine, see a display of available flights
and choose one.  I can check in or by a ticket or both according to my
requirements.  The next step should be to allow me to make the reservation
from my home computer or terminal and to eliminate the ticket or at least
make it machine readable.  When I get to the door of the airplane, I could
move my credit card or machine readable ticket through the slot, and it
would let me through.

	Most large organizations have purchasing departments.  Someone in
the organization who needs something fills out a "purchase request" or has
his secretary do it, and sends it to the purchasing department where
various requirements are checked, and in the fullness of time it is turned
into a purchase order and sent to a seller - often with several mistakes.
With an automated purchasing department, the person who needs the purchase
would interact with the purchasing program directly from his office
terminal.  That program would make sure the requirements were met via
dialog and then, while the user sits ther, would interact with the
programs belonging to suppliers and settle on the purchase.

	A major component of automated bureaucracy will be what I
have called a "Common Business Communication Language".  This is
a language used by a computer program belonging one organization
to interact with computer programs belonging to others.  CBCL would
standardize inquiries about price and delivery and the actual
placement and acceptance of orders, for example.  Many other
kinds of dialogs would also be standardized.


Artificial Intelligence

	Many ``expert systems'' have been and are being developed
for a variety of purposes.  The present expert system technology
is good at keeping maintaining large numbers of simple rules
such as ``IF the Gram stain of the organism is Gram-negative,
the morphology of the organism is rod, and the aerobicity of the
organism is anaerobic, THEN here is suggestive evidence (0.7) that
the identity of the organism is bacteriodes''.  Notice that
having hundreds of such rules does not require that the expert
system know that bacteria are organisms that grow and reproduce
and cause disease.  Indeed it is not even necessary for such an
expert system to even reason about processes occurring in time.

	Reasoning about processes occurring in time is possible,
and a few present program do it, though not with the flexibility
of a human being.  To make the long story of the present state
of artificial intelligence short, we humans haven't yet been able
to understand enough about humans reason about processes or about
the mathematical structure of such reasoning to make computer
programs that do it very well.  Progress is being made, but
fundamental new concepts are almost certainly required.

	Many kinds of intellectual work do involve following
large numbers of rules, and there expert systems can help.


Keeping track and personal optimization

	I want to mention an application that is on the borderline
of feasibility at present.  Consider a program, available to a
person in his home and workplace, that would help him optimize
personal decisions and remind him of important considerations.
For example, if a trip was necessary, it would remember goals
and obligations that it had been told about and suggest ways
of achieving several with the same trip and how this might be done.
It would also help its user avoid contracting incompatible obligations.
Such a program requires common sense information about the effects
and side-effects of many kinds of actions.  It also requires the
ability to reason about goals and to use its knowledge of actions
to plan their achievement.  It needs to be able to reason about
the effects of events concurrent with the actions it might suggest
to its advisee.  Achieving these abilities, required for a good
adviser, is probably possible with the present understanding of
artificial intelligence, but it hasn't been done in any generally
useful way.

Expanding the right of initiative

	I cannot resist mentioning one grand vision, even though it is
probably still far off.

	Many people have ideas of how something might be done better.
Most ideas are both bad and old, but a few are worthwhile.  While good
new ideas are rare, they are what moves society forward - yes forward.
Unfortunately, most good new ideas lie undeveloped, because our
society is not very good at recognizing the good ones, developing
them to a practical level and putting them into practice.  Indeed
the ability to and resources required to put an invention into practice
is much greater than that required to make
one.  Perhaps that's why entrepreneurs are often richer than inventors.

	The possibilities for getting a good new idea adopted vary
according to the field.  The best situation is in mathematics and
the theoretical parts of other sciences.  There one proposes ideas
by writing papers and submitting them to scientific journals.  Most
journals send all submitted papers to referees (experts in the field)
except those the editor regards as obviously inferior.  The editor
and the referees consider the submitted articles independently of
the status of the author, and a complete unknown can get an article
published if it seems interesting.  As an extreme example, Einstein
was an unknown employee of the Swiss patent office but got four papers
published in the world's most famous physics journal Annalen der Physik
in the one year 1905.  These papers revolutionized physics.

	Close to the other extreme are ideas about government policy.
Many people have ideas, but only a few have power to implement anything -
and then not much.  Whether an idea is considered depends a great deal
on who proposes it, and unknowns usually have no chance.  Even worse
off are ideas about changes in social customs.  There is no authority
at all that can implement them and usually no place to publish them.

	The idea for improving this situation goes back to Leibniz, who
said, that when people faced a controversial subject, eventually they
would say, "Let us calculate" instead of just arguing.  Leibniz wanted to
invent a mathematical logic that would enable people to calculate what
consequences would follow from given axioms and assumptions.  He didn't
get very far - not even as far as the propositional calculus invented by
Boole 150 years later.

	A modern implementation of Leibniz's idea would involve a
computer program for checking proofs in mathematical logic and
a database of accepted facts about the domain under discussion.
The user of the system would enter the steps of his reasoning
on a terminal, and the computer program would check them.  If
the program found an error, he could try different reasoning
or he might discover that what he was trying to prove wasn't
true.  Such programs exist and do somewhat more.  Namely they
help their user construct proofs, and can prove some things
all by themselves.  As artificial intelligence develops, the
programs will become more powerful.  However, the present
programs are usable only in mathematics.

	Using them for human affairs, as Leibniz intended, is
much more difficult.  First of all, there has to be a database
of agreed facts about the common sense world, and it has
proved very difficult to formulate even the beginning of such
a database.

	However, although mathematical logic is highly developed today, it
still isn't usable for settling questions about human affairs.  It lacks
something.

	Many people would guess how this essay might continue.  They would
say logic lacks emotion or maybe it lacks soul.  That's not the direction
I'm going, and that idea has been a distraction from the real problem.
What logic so far lacks is
something rather technical.  Among possible other things, it lacks
non-monotonic reasoning, and it lacks a formal notion of context.  These
are being worked on in artificial intelligence, but we're not far enough
along.  When we are, it may be possible to have a vast expansion of what
we may call the ``right of initiative''.

	Suppose a person not connected with the government has a new idea
about (say) the foreign policy of his country.  Suppose it is a really
good idea, and would be accepted by the authorities if they only paid
enough attention.  Its chances are very slight, because these authorities
are continually bombarded by proposals of all kinds from people in
relatively high positions.  Even if they are entirely open-minded, they
don't know what proposals to pay attention to.  Moreover, the initiator of
the idea cannot know what facts are being taken into account.

	In the future Leibnizian world, our hero will be able to develop
his idea in interaction with a database of foreign policy facts and
assumptions.  It won't be easy for him to demonstrate to the
reasoning-checker program that his idea is good.  It will take him a lot
of work, and very few people will succeed.  When he does succeed it will
be a major event and rare enough so that the authorities will be able to
take the time to consider it.  There was not time or space to do more
than sketch this notion of an ``idea filter'' that would get attention
to the ideas that passed its test.

	The Leibnizian world will put a premium on hard thought as
compared to salesmanship or the skills of a courtier.