From: Minsky   MONDAY  MARCH 04,1974 RE: Proposed AI committee.  

   WARNING: I have spent only a few minutes writing this memo. At the
   present time the staff of our laboratory feels harassed and concerned
   about getting done the research we have planned.  

   We do not want to become burdened with any more time-consuming and
   unproductive administrative or "review" or "planning" problems.  I have
   read Newell's memo of 2 MARCH 74.  Generally I do not agree with it,
   because though many of his points are well taken, the objections to
   complex reviews are more serious at this time than "usual".  

   The occasional review of a research program can be a good thing under
   certain conditions. These include:  

        1. It must not happen very often. Managers underestimate the
        resources consumed in preparation for such reviews, especially in
        those cases where the laboratory or project is in fact managed by
        principal active scientists.  

        2. If the program appears to lack coherence or competence than it is
        useful to propose a review within another year. If things seem
        productive, two years is more reasonable.  If a group or laboratory
        is unusually productive, management should make special efforts not
        to meddle too much.  


        3. The various purposes of presentation of results and viewpoints
        should not be confused.  

           Preparation of plans for applications have to mesh with proposed
           users' plans.  

           Preparation of "reports" for the education of the scientific, or
           the popular, public have other problems.  

           Preparation of reports for educating managers is inexcusable if
           it consumes an appreciable fraction of the project's effort! See
           below.  

           Preparation fo a general explanation of AI concepts and status
           for the use of DOD is probably unjustifiable, and any such
           reports should be designed for the general scientific public --
           perhaps through NSF or NAS.  

   In connection with my own laboratory, there has been too much concern and
   activity over the past two years involving the project's principal
   research workers in adjusting to changes in ARPA's proposal and
   management requirements. Vigorous participation in any more reviews and
   reports would be inappropriate for at least a year, probably two.  I










                                                                      PAGE 2





   cannot overemphasize the firmness with which this statement is made.
   There is already far too much concern on the part of staff members with
   the insecurity of the situation. But worse, the "corruption" of good
   scientific analysis by the desire for premature demonstration of useful
   applications is becoming serious, I think.  The problems we are working
   on are still very difficult, have a 5 or so year prognosis, and can not
   but suffer from forced attention to largely unrealistic "application"
   areas. I claim that our contributions are very substantial, but tend to
   be in the "basic" areas that are better developed elsewhere, and that
   this should be plainly obvious to "management" without further complex
   assessments and studies.  

   This is not to say that, for example, our "electronic Circuit Repairman"
   project is a bad thing. But ARPA has pressed our project (by not
   responding to inflation) into a steady contraction of depth, so that
   there is no room any more for the kind of research that led to this
   stage. 

   In case it should not be realized by higher Management, I think it is
   important for everyone involved to realize that many of the major results
   of our own work -- including much upon which -- four or five years later
   -- much of ARPA's plans for work in Natural language, Automatic
   programming, Speech, etc., are partly based, were NOT results of long-
   range organized planning, nor even in "proposals". These include the work
   of Winograd, Winston, Goldstein and Sussman, and others.  

   I am not convinced that either Natural language or Automatic Programming
   is advanced enough to benefit from agressive "management" -- although a
   review might help to allocate resources among different projects and
   prevent some waste.  Applied Robotics could be managed under certain
   circumstances; namely in connection with particular limited applications.
   But it might be a mistake to do this without carefully protecting the
   more "basic" research efforts, for there are problems in too much
   competition within a project for resource allocation when the shallower,
   applied sector appears to be more responsive to program needs suggested
   from the contractor.  

   Finally,  I do not believe that the internal ARPA issues have been
   properly separated from the problems of organization of the AI community.
   To some extent, surely, there is a need to better inform ARPA top
   management about AI. To do this without enormous waste of everyone's
   time, there should be some simple face-to-face discussions with them to
   find out whether there is in fact a problem of misunderstanding about the
   state of AI and its potential. I am not a management specialist, but it
   is my impression as a reasonably successful producer of good research
   that something has gone wrong in the past two or three years and that the
   amount of distracting bureaucracy has grown immensely.  Before
   encouraging a complex and time-consuming system of reviews, we had better
   see if this is as likely to make things worse rather than better for
   everyone -- the AI laboratories and ARPA.  








ββ