.require "memo.pub[let,jmc]" source
.double space
.begin center
THE REAL COMPUTER REVOLUTION

Prof. John McCarthy
Artificial Intelligence Laboratory
Computer Science Department
Stanford University, California
.end

	I'll begin my talk  about the  computer revolution by saying
that  there has not been any; it is still in  the future.
There is a literary  convention  that technology is accelerating
faster and faster  and people are having difficulties adapting to  it,
but in terms  of  technology affecting daily life,
there has been a slow down since about 1930.
In terms of benefits of technology and also in terms of
stressing people's capacity to adapt,
less has
happened  since  World War II than happened say between 1890 and 1920.
We have had nothing that demands as  much adaptation as the automobile
did,  as  the telephone  did, as  having electricity in the home
did,  as  having
domestic refrigeration  did.  This may have the paradoxical
effect of promoting anti-technological attitudes, because  the  public
can quite  reasonably ask, "What have they  done  for  us lately"?
Technology has had large recent effects in  increasing prosperity and
increasing  health,  but both  prosperity and health  are  things that
people quite reasonably  take for granted.

      Perhaps  I was expected to speak about  artificial  intelligence
and the social problems it might produce, but I  would like to explain
why I won't.  I think that artificial intelligence,  and by
this I mean intelligence on the human level, is achievable and when it
comes it  will be  more  revolutionary than  any previous  scientific
development, but I also think that I do not know enough  about what it
will really be  like  to say anything sensible about what its  social
effects will  be.  I  thought about it quite a bit, and  I  think  almost
everyone else knows even less than I about it.

	Some people, for example Ed Fredkin of M.I.T and Donald Michie
of Edinburgh University, beat the
drum about the social importance of AI
and hold conferences on "what if" and so
forth. I deplore this and  ask you  to imagine
that someone  had managed in the  last
American  presidential election  or  the  forthcoming one or  in  some
Israeli election to  ask  the leading  candidates  their  positions on
artificial intelligence.  After they finish  saying  "huh?"  and decided
that  there  was  some  possible  advantage in having  a
position, they would turn to some  underling  who  would write
whatever came to his mind and whatever would suit better the  doctrine
of that  party and we would be no better off; it probably would not do
any harm but it might, because who  knows, there might be some new law,
stupefying in its silliness.  Computer scientists
would at best find themselves in the position of the biologists
regarding recombinant DNA - spending full time explaining why proposed
new bureaucracies and regulations would be harmful.

     I shall talk about is home  computer  terminals  and
unreconstructed time-sharing and compare  them with home  computers.
I don't  object  to home  computers,  that  is to  say  having a
terminal  with a  computer in it, but I think  what is happening  with
home computers is what  could  have been expected; most people, even
most scientifically educated people have little
use for  computing per se except perhaps as a hobby.
I have  had  a terminal in
my home  for more  than ten years, and  I use it a lot for
writing books, papers and letters and sending and receiving messages and keeping
up with the news
and occasionally for writing programs. However,
I have  never  written  a  program  that has  anything  to do with  my
domestic life though  I do keep a  few files on hand that are relevant
to that, and sometimes they are helpful if I actually remember to look
at them.  These applications don't require that the terminal have
independent computing capability; mine has, and I haven't used it.
They do require communication capability and the availability of substantial
private and public files on disk.

       Let me  begin by describing some of the things we  have done at
Stanford and  that  have been done at some other places that might  be
relevant to the question of home use of computers.
When we  started our  laboratory  in 1966, we had already  decided
that  documentation was  going  to be  an important use of our computer.
We arranged to
have  a line printer that would print  upper and lower case and a large
character set, which was much more of an  expense  at that time than it
is today.  When the opportunity came around  1970 we planned and built
a display system  with a  terminal in each office rather than
terminals in terminal  rooms, and  we arranged  that  the  secretaries
should also have terminals.  I became interested in  home computing
about that time and foresaw a very substantial chicken-and-egg problem,
because  the applications  of home computing really
are  communication and data  base applications, and without
the data bases, you can't demonstrate the applications.

	We found one application where the necessary data bases are already
electronically available, and this is a news service.
We have  coming into our computer the news wires of
the Associated Press and  the %2New York Times%1.  Our computer is
connected to the wire just as though it were a teletype in a newspaper
office.
  As each story comes into
the computer, a concordance is made according to
all non-trivial words, i.e. all but words like "the"
and "he".
This concordance is continuously updated as the stories come in.
Each midnight the file of
the day's stories is closed, and we keep two weeks; if we had
more  disk  file we would  keep more.  When the  program user types
NS for "news
service", he can give the program a "key word expression".  Thus
if he were interested in Israel but felt he had heard  enough
about Dayan's resignation then he might type  the key word  expression
"israel-dayan" which would give  all stories that
had the word "Israel"  in them  but  did not have the word  "Dayan" in
them.  You  can  make  arbitrarily  complex  Boolean  expressions.
You get some false hits, but
the data  base is small  enough  so that this isn't harmful.

	The system is quite crude, but in some respects it compares
favorably with the much more sophisticated %2New York Times%1 Data Bank.
The latter employs human abstractors who make  abstracts of
everything that appears in the %2Times%1 and many other publications
and files them under key words that seem
important to the abstractors
at the time.  If something becomes  important in retrospect,
for example  the building Watergate,  then you cannot go back  in time
and  say  what stories  mention  Watergate  before the time  it became
important  because  the abstractor didn't then consider  it  important.

	One
can ask NS to be notified when stories
that fit specified key word expression come in.  You get
an announcement if your are logged in and
electronic mail  if  you  are not.
You can also move stories to files
or print them.   This  was one of the few experiments aimed at home
computer use that could done in 1972.

	Our  Dialnet  project is also relevant.
It is a  project for communication  between
computers giving services like those of the ARPANET  to any computer that will
equip itself with a computer operated
telephone dialer and a 1200 baud Bell System
compatible modem and implement the Dialnet protocols.
Given this, a user can transfer a
file between  his computer  and any other computer.  He need only give
the local file name and the foreign file name and then get through
the other computer's password fence.

	It does  require something that one might take  as obvious -
that every computer system have a systemwide naming of files.  However
curiously enough, and this will be my first nibble on a hand that have
been very  generous.    Few if any IBM operating
systems  have  this  elementary
characteristic of having systemwide file names.  What we actually need
and can get by suitable prefexing is world wide file names so that you
can  refer by name  to any  file in the  world -  not necessarily  that
everybody would let you have any file or even admit that it exists.

	Dialnet  is also used to send  computer  mail.  Thus a
computer may  receive a telephone call from another computer in which
it says "I have mail for  your user  named so-and-so", and  of course the
receiving
computer can  reply or  just  hang  up or it can reply that it has no such
user or it can accept the mail.  Perhaps this is the way computer mail
should have  been developed in
the  first place instead of basing an ARPAnet on
leased dedicated facilities.  The standard dial-up telephone network much
more  capability than has been exercised.  For example, it is already
international.

	Another  sort of protoproduct project
is the Common  Business Communication Language.  When I first thought of
it, it didn't seem to have much scientific interest and certainly no
connection with artificial intelligence research.  It seemed to be
just another idea for using computers to increase human productivity.
It  was set  off by
reading a 1965 article about the world of the future - a 1965
model world of the  future in which  businesses
would be quite computerised.

	There was  a  scenario in which a clerk in
the  purchasing  department of a company A  hears a beep and
turns to  a  display screen that says, "We need 5000 pencils.   Order
them from company B".  She (Sorry Pat, this is a 1965 model
world of the future.)  turns from her terminal
to her  typewriter  and types  out a purchase order  which
is duly sent  to company B where another clerk  turns from her
incoming mail basket to a
terminals and types "Send IBM 5000 pencils".

	We would like to let both
clerks  do something  more  worthy  of  their  talents  and  have  the
computers  communicate  with each other directly.  In order to  do this we
need a standard  business communication language wherein one computer
can call up another and  can say "What is your price  and delivery for
5000 pencils?" and "I hereby  order 5000 pencils" and things like that.
This may  seem like a grubby bit of commercial standardization
of no interest to computer  scientists,
but  in  fact it  raises  all sorts  of interesting
questions.

	One question is how such a system can ever change.
I began to think about this while
visiting a San Diego naval training facility.
There is something called  the fleet data system
whereby every  ship  in  a  fleet would  know about all  of  the
airplanes and ships that any of the  ships  in the fleet had  detected
by exchanging messages. At  the time NATO was
going to join this system and suitable committees had determined  that
the system should be updated or improved at that time and there was to
be a cutover day.   There was  the hope that the Russians would not do
any thing on this cutover day  because great confusion was expected.
Presumably this has happened  by  now,
since the visit was 7 or 8 years ago.  Massive changes on cutover dates are
quite unsatisfactory for large communication and data systems and would
be almost impossible for loosely coupled systems involving tens of
thousands of computers of differing ages, sizes and manufacturers under
no central management.

	Growth requires that when a new kind of message is introduced
message sources begin transmitting both kinds of message.  When enough
new style messages are being transmitted, receivers begin using them.
When no-one requires old style messages, their transmission may stop.

	Another design idea is that each message should be a
list, and every item should be a either an
atom or a list whose first item
designates  what  kind of a thing the list represents.
You may have heard of  such
lists; they begin and end with parentheses.   Remember that
only computers get to see them.

	Another characteristic was suggested
in N.  Chomsky's book %2Reflections on Language%1; he claims that human
languages have it, other linguists dispute it, but anyway we want it for
our Common Business Communication Language.  It is that the  syntax
never requires an atom at
any place.  For  example, if a  price is to be designated then of course
α$3.12  is a  reasonable  thing to  appear as a list item,  but any
expression designating a  price should also  be allowed,  such as
translations of %2"the same price as last week"%1 or %2"in accordance with
our contract"%1 or %2"the price granted to the most favored customer"%1
or whatever.

	When I tried to specify a simple version of a CBCL that would
be expandable, I found substantial
difficulties which led to some interesting scientific problems.  Consider
designing how delivery is specified.  First, the conditions might be
unspecified, meaning  that the delivery conditions are
standard, or it might  say %2"the best way"%1.  However, if you want
to elaborate what is really to  be allowed in delivery specifications,
there  are  all
kinds of  conditional specifications such as %2"to  be sent  by 
 United Parcel Service
unless they are  on strike,  in which case it goes  by post  office"%1 or
vice versa.  It turns out that a substantial and interesting part of
the ⊗semantics of natural language is involved in formalizing what one
business computer might say to another.

	This is interesting from a scientific point
of  view, because  interest  in computational  linguistics  has  mainly
focused on the syntax of language.  Here is a case where we are going
to use  a perfectly rigid syntax decided in advance, but we have all
the semantic  problems of deciding exactly  what so you want one computer
 to be able to say to another.
This has led me to the belief that putting English language "front ends"
on programs that  basically  speak  in  "computerese"
may miss the point scientifically, because more interesting than
how to say in  natural language what we already say in
various computer input-output  jargons,  is how to make the computer say
things said in
natural language that  we do  not presently know how to make computers
say at all.

     This was by  way of an advertisement  for  the scientific
interest of some of these problems.  Another important feature
of our environment is
the mail system that permits sending electronic  mail all over
the  ARPANET  and to other places.  I  certainly have gotten  into the
state where I divide the world into two classes of  people -
those to whom I can send ARPANET mail and those with whom I have to rely
upon more ordinary kinds of communication.  I think I will not sing my
song about  why  it  is  often better than telephoning and things like
that. I will say that I suspect that IBM may have  gone into some kind
of blind  alley if  they  suppose that  the speech  filing system will
obviate the need for written electronic mail facilities for everybody -
even executives - who  can type as well as engineers
when  they have the motivation - and
under certain circumstances, I believe they will.

     I would like to do as  the other speakers have done and talk about social
effects,  but  I want to  express the opinion that much of the talk
about the effects of technology has been harmful and even immoral.
I believe  that people  should be cautious
about  talking about  social  effects, especially about  talking about
planning desirable social effects and planning that undesirable social
effects  should be avoided.  The danger is that  this  will
come down to planning other people's  lives for them,  and it seems
to me  that this disease afflicts many of those
Americans who refer to themselves as the  "%2public policy
community"%1. 

	In
particular, I doubt the virtue of the Office  of Technology  Assessment
which Congress created to advise about the consequences of new technology.
Its attitude seems to be %2"Don't do anything new without consulting us, and our
new director will take  office in a few months, and we will be able to
consider  these things when we have decided in what  order to start to
consider them"%1.

	Let me  put  it this  way: hardly any technology has
specific
 social effects, good  or bad; what technology  does  is  provide
opportunities.  Some  people can  use these opportunities to do things
that will  turn out to be good  and other  people  can  use them to do
things that turn  out to be harmful.  Which  you emphasize
depends  less on specific technologies than on  how you
look at people  and institutions and the  ways they can be
expected to use any opportunity. 

	You may say that  if we give them  the  chance
they are sure to abuse it, and you can have two kinds of worry: one is
what they will do to each other and the other is what they will  do to
themselves.

	Some people  worry that if people  have  too much  access  to
computer files they will waste their time or if they have time-sharing
they  will try  out  programs  without  adequately thinking.  They are
really all out to help each other by disciplining each other.  Dijkstra's
book %2"A  Discipline of Programming"%1 deserves
a book jacket picture of  someone
standing on a dais with  a whip menacing a crowd of half-naked programmers
at computer terminals.  So if
you want discpline, put an ad in the %2Berkeley Barb%1.

	My idea of social consciousness is different, and instead of
inventing ways in which things might go wrong, I would like to discuss
some useful opportunities.  After all, the probability that an opportunity
will be missed is far greater than the probability that an unfortunate
aspect of an application will go unnoticed.

	For some years now, disk storage has been cheap enough so that
it has been economical to have a national library in which everything
that has been written is stored on disk and available from a terminal
anywhere in a country or even in the world.  Such a file would be the
effective equivalent of a very large number of copies of the book.
Recently it has become cheaper to store even a single copy on disk
than on a library shelf - counting the cost of the library, staffing
it, cataloging the book and the cost of the book itself.
Our society has suffered from missing the 1970 opportunity to create
such a library, and a good part of
the blame for this belongs to the technology
assessors, whose imaginations extend only to what may go wrong
with new technology.

	The simple convenience
of having everything in the Library of Congress immediately available
is important in itself; much misinformation would be avoided if such
facts as are available could be easily looked up.  However, its effect
on the future of publication is sociologically even more interesting.
If all publication is available via terminals from disk file,
then publication consists merely of declaring public a file that
you  have put into the system.  In order to get anybody to
read it you  have  to get it to  his  attention, and since we can't suppose
that people will read more than now, this will continue
to be complicated and uncertain.  As now, some people  will  read
files constituting  book reviews, and  the reviewers will  review what
books  they  would consider  worthwhile,  and  so forth.
People who
write something  eccentric or not  of much  interest to other
people will have as much trouble getting their stuff read under a new
system as they do now.

	The important fact is that such a  system will  make the writers,
especially the well known  ones, substantially independent of  publishers. The
cost of getting a book out for a reader is now something like  between
10α% and  20α% to the writer and 80α% printing and distribution costs, and
these numbers  can easily be reversed.  I think  this would make for
fewer fads  and it would also have the important social effect  that a
smaller public would  be  needed in order   to support someone  who is
writing because he could get his income with  roughly one-fifth of the
number of readers that he presently needs.  Another aspect is that much
more detailed information would be available for those who want it.

      When we first put the  Associated Press into our computer I  had
this  hope that there must be  all these great stories that the AP put
out that  the newspapers  don't  print, but of course that  was a dumb
idea. AP is a co-op of newspapers and it  is not going to put  out  any
stories that the newspapers aren't likely  to  print. So it  turns out
really  shallow, and if  you consider the  AP correspondent  in a place
like Moscow or Peking or Jerusalem,
you must say that he must get a bit bored with
his job. because he  writes the much same thing every day, and he does  not
write very much. It is just packaged in what a newspaper is likely  to
print.

     Another  desirable  social effect of on-line publication
is to  make  an  effective
right of reply.  Politicians and journalists often engage in hit-and-run
controversy, in what he says or writes
depends to a  large extent  for its effect on the fact that people who
disagree with  it or the people who are specifically attacked will not
be able to  mount an instant reply.  With the on-line system, if you  read
(say) an attack on the "Eastern Establishment" by
a Republican President,
then you  could ask your terminal %2"What do the Democrats reply to
that?"%1.  I think this
would enhance  the quality of debate which has greatly declined in the
United States at least in the past 100 years.  If you read even part of
the  Lincoln-Douglas  debates  you  discover  that  one of  them could
propound to the  other a very complex hypothetical question about what
the country should  do  "if" and  the other would actually  feel
obliged to answer.  That certainly has not happened at all recently.

      The third aspect of such a new system which I consider desirable
is that  it  would enhance  the ability  of a  group  to maintain  its
separateness if it  so desires.  They could  be physically scattered but
intellectually much closer together; the cost of having a newspaper or
magazine or whatever kind of communication seemed  best to  them would
be greatly reduced from what it is now.

	Now I want to  finish off  by giving some advice, unsolicited of
course. We will give this advice to three entities, the first of which
is  IBM.

	The  first
question is when are they going to give up the punch card check?  
The punched card check  is not important in itself but as a symbol
of attitude.  Its use requires an adaptation by humans to efficient
data processing by machines.  It was acceptable in the age of mechanical
accounting machines, when machines were clumsy and difficult to
adapt to people.  Its maintenance today serves no efficiency purpose
but is merely a bureaucratic tradition in organizations that see
computers only as a means of doing old things cheaper and put no
emphasis on using the flexibility of computers too allow he people
with whom the organization deals more convenience and flexibility.

	While the world library application mentioned earlier is cost-effective
even at the present cost of disk files, provided it is well co-ordinated
so that only one copy is stored of the material for which there is
small demand, our society has less ability to co-ordinate
many people to produce efficient systems than it has to obviate the
need by better technology.  IBM is the world leader in magnetic disk
technology, and rumor has it that disks six times as
dense as present disks and no more expensive
are easily feasible.  It would be a real service to
make this technology available as soon as possible, even if the present
applications don't seem to require it.

	A key technological requirement for making computers and
bureaucracies treat people politely is good time sharing, and my recent
experience with IBM computer systems has confirmed my old belief that IBM
is not up to the state of the art.  Its terminals are too complex, it uses
half duplex interaction, its editors are inefficient and hard to use,
etc.,etc.  I don't want to go into that right now; if they ask me I will
tell them.  (Note: they aren't interested).

	In this connection, Lewis Branscomb gave an interesting example
having to do
with an on-line  application  that  was   interrupted   for  some  batch
processing.  If there  is any technology trade-off which has  a  social
impact it is the distinction between batch and on-line processing.
Whatever they did do, updating this data base with batch
processing  should have been done on-line.  Indeed I would
claim that batch processing has been responsible for many of the evils
that  have  been ascribed to computers.  If the information required
to answer a question or correct an error is on a tape sitting
on  a shelf, you can't solve the person's problem
until  the  next time you run the data through.
Whereas if the data are in on-line files available to the people
who receive the requests, they can answer questions and make changes
in the data base on the spot.  Before disk files became large and
cheap, batch processing was necessary.  If I weren't an opponent of
regulation, I would favor making it illegal in applications that
concern current records about people.

	However, one's social conscience should not merely be concerned
with fighting off evils.  This is good enough for unimaginative people,
but scientists and technologists should put most of their efforts into
creatively taking advantage of technological opportunities to improve
the quality of life.

	One idea is to urge businesses and government agencies to make
more and more of their data bases directly accessible to any person
with a terminal at any hour of the day or night.  A customer should be
able to find the state of  his account even at
2:00  in the morning, if  that's  what he  feels  like  doing, of course
with suitable safeguards against wiping out one's debts or reading other
people's files.  A simple password form of security against snooping
will be adequate for almost all people, but those requiring more should
be able to get it by paying the cost in money or in lack of access to
their own data.

	Now I want to  give  some  advice  to governments.  First a piece of
advice to  the U. S. Government: there is all  this information that  is
supposed to be available according  to Freedom of Information, but  of
course it is sitting in libraries or reading  rooms in  the  agencies.
Making Freedom of Information into an effective right
requires that this  information be available on-line on from
terminals anywhere in the country.

	An important negative freedom is that no organization - be it
a government, a business or a university should be allowed to require
someone to answer a question to which they already have the answer.
If they are nervous about whether the answer is still correct they can
ask to have it confirmed, but the U.S. Government has asked me for
my birth date, which cannot change, at least a hundred times.
Of course, the best way in America to get a new freedom is to get some
lawyers to claim it is in the Constitution.  It will require someone
with a really good legal imagination to find this one in the Constitution.

     I have talked enough, so I will skip the rest of my advice.