∂31-JUL-74  1000		network site RAND
 JOHN -
    THIS IS A COPY OF A RECENT MESSAGE TO CRAIG FIELDS
 ABOUT SOME OF THE IMPORTANT ISSUES WE THINK SHOULD BE
 ADDRESSED IN THINKING ABOUT A RESEARCH PROGRAM IN
 INTELLIGENT TERMINALS.  IT CAN FORM THE BASIS FOR
 A PORTION OF OUR DISCUSSION ON THURSDAY.
                    BOB ANDERSON
 
 To: Craig Fields
 
 From: Bob Anderson
 
 Date: 28 July 74
 
 Subject: Intelligent Terminal Study -- Update
 
 Copies: P. Weiner, J. Markowitz, E. Feigenbaum
 
        (This message contains 353 lines)
 
 Craig --
 
 The following is a brief progress report on our Intelligent
 Terminal study.  If you have comments which it would be
 useful for us to think about while you're in Europe, I'd
 appreciate feedback in a message before you leave (or
 from a TIP in England).
 
 
 1.  Task Areas
 --------------
 
 We are currently concentrating on the following task areas.
 Each is about the size and scope which can be handled by
 a separate contractor initially.  For each, we are working on:
 
   .  putting bounds and limits on the scope of the area,
      so that a clear, understandable goal with rather
      specific attributes is delineated.  (Or, alternatively,
      defining a graduated set of increasingly difficult
      goals and capabilities in the area);
 
   .  describing benefits (and costs) to be derived from
      this IT feature for each of several example
      application areas (see #2, below);
 
 These task areas seem to recur in much of the I.T. and M.S.T.
 documentation and in the documents by T. Standish and
 W. Sutherland.
 
      The tasks are:
 
      1(a).  A "USER AGENT" or "ALTER EGO".
 
      A continuously or periodically active process that
      monitors for events of interest; it organizes and
      prioritizes incoming data and messages; it represents
      the user in low-level negotiations (e.g., calendar
      scheduling, appointments).  Research questions include:
      how a user can "program" or "instruct" the agent,
      how the agent is best implemented, and the extent
      to which it can be a surrogate for the user.
 
      1(b).  I.T. AS AN INTERFACE TO EXTERNAL SYSTEMS.
 
      -- "intelligent" handling of other systems' defaults,
         cues, error msgs, etc;
 
      -- hide details of subsystem interactions from user
         if he wishes;
 
      -- negotiate with other systems concerning terminal's
         capabilities, data rates, formats, etc.
 
      Research questions include:  how to program or instruct
      I.T. in other systems' vagaries; protocols for inter-
      system negotiations (cf. recent proposals for ARPANET
      graphics protocols); the I.T.'s role in deciding remote
      vs. local for computing or data storage, and in deciding
      among the use of several remote systems (cf. RSEXEC),
      each of which is capable of accomplishing a particular
      task.
 
      1(c).  ADAPTABLE, IDIOSYNCRATIC USER INTERFACE.
 
      This task stresses the trainability of the interface.
      The interface will most probably contain models of
      (1) the user's traits, preferences, idiosyncracies;
      (2) the user's environment (his organizational
      relationships, who he communicates with, etc.);
      (3) the user's tasks (their deadlines, constraints,
      data bases).  These models will allow the interface
      to act "intelligently" in having a semantic under-
      standing of a user's requests.  Another important
      feature of the adaptiveness and semantic understanding
      features of the interface will be the interpretation
      of advanced forms of I/O:  voice, hand-printing or
      -writing, physiological signals, graphic output.
 
      Research questions include:  how best to create,
      store, and access "models" of user, etc. for this
      particular application; trainability of interface --
      by analogy, by inference, by pattern-action rules?;
      what combinations of advanced inputs (e.g., voice
      plus tablet) to use, and how to interpret using
      semantics?  What features of user to monitor, record,
      adapt to?
 
      1(d).  "APPLICATION" PACKAGES.
 
      Applications which are sufficiently ubiquitous to merit
      having built-in capabilities in the I.T.   Candidates include:
 
      -- text management
 
         editing, storage and retrieval,
         search (either loose or constrained), etc.
 
      -- task management
 
         resource allocation tools, exception reporting
         (cf. some ideas in DNLS related to task management)
 
      -- time management
 
         calendar, appointments,
         travel
 
      -- tailored calculator with memory
 
         desk calculator functions, but programmable
         and with graphic output options, defaults, etc.
 
      -- bookkeeping functions
 
         locate persons within the system;  keep records
         of certain activities, under user control
 
      1(e).  IDENTIFICATION OF USERS AND SYSTEMS.
 
      I.T.'s ability to monitor and identify users (bringing
      to bear its user models?); monitoring for erratic or
      very idiosyncratic behavior; identification of external
      systems to the user, and monitoring for proper system
      behavior, and I.T. awareness of which system in a subsystem
      hierarchy user is connected to.  Most of these features
      are primarily for security, but they're also useful
      for "help" assistance to a confused user.
 
      Research questions include:  how to best achieve a
      specified level of assurance concerning the identity
      of a user, and of a system?  How to handle cases of
      possible imposters:  entrapment?
 
      1(f).  EXPERIMENTAL DERIVATION OF HUMAN FACTORS SPECS
             FOR INTERACTIVE SYSTEMS.
 
      (I haven't talked to Joe Markowitz about this yet;
      all his ideas plug in here -- and elsewhere I imagine.
      I won't try at this time to even outline the scope
      and approaches for this component.)
 
      1(g).  ADVANCED INTELLIGENT TERMINAL DESIGN.
 
      The hardware, firmware, and architecture of advanced
      terminals.  How much computing power to build in?
      Projections of hardware cost, size, power, etc.
      Design ideas for a family of terminals, from "office
      model" to "very portable".  Input devices, graphics,
      etc.  (Cf. paragraph VII.G of Groner's draft outline
      of 12 July.)
 
 2.  Applications for Intelligent Terminals.
 -------------------------------------------
 
 There are at least the following seven distinct application
 areas for an I.T. with the features mentioned above.  We
 are in the process of giving an initial cursory look at the
 relevance and benefits of an I.T. in these areas.  Obviously,
 we must soon winnow this list down to 2 or 3 which we will
 then analyze in some detail.
 
 We are in the process of forming a large matrix, with these 7
 application areas on one axis, and the 7 I.T. functional
 characteristics mentioned above on the other axis.  At the
 intersection of application(i) and terminal characteristic(j)
 we will record the particular relevance of that characteristic
 to that application:  its degree of importance, any unique
 requirements, such as response speed, resolution of graphic
 output, user characteristics and training, etc.  We expect
 this chart to show which features are most important, and
 which application areas most need the features of an I.T.
 From this analysis will come the winnowing.
 
 The application areas follow.  For more description of many
 of them, cf. Rein Turn's recent Rand report "Military
 Applications of Speech Understanding Systems", R-1434-ARPA, June l974;
 this contains a very good description of the characteristics
 of many of these applications, with references to more detailed
 literature.
 
      2(a).  INTELLIGENCE ANALYSIS.
 
      Use of I.T. to interface to a variety of computers
      and application packages with minimal hassle to the user.
      The emphasis would be on search and retrieval of
      text strings, possibly with context-dependent
      options.  Rather small user community, but with
      intensive, long-term terminal use, so there is good
      opportunity for intensive relationships to be built up
      between man and machine (i.e., extensive user model,
      good handling of idiosyncrasies, etc.).
 
      I.T. here would play an important role as an interface to
      a very-large-data-base system.
 
      (More data will soon be available on this area from a
      meeting to be held on A.I. plus intelligence.)
 
      2(b).  THE ARMY TACTICAL OPERATIONS SYSTEMS (TOS).
 
      There are plans in an advanced stage of development
      for an electronic communication system for tactical
      battlefield use.  (Cf. pp. 89-94 of Turn's report.)
      TOS uses several types of terminals to originate and
      transmit various formatted messages in a tactical
      environment.  The current terminals have very bad
      interactive characteristics, and are clumsy and
      cumbersome.  I.T. could greatly aid in this environment,
      reducing bandwidth, speeding interaction, reducing errors.
      Potentially large, visible user community.  Requires
      interaction with Army Computer Systems Command, but there
      also exist Navy and Marine counterparts.
 
      2(c).  NON-COMBAT FIELD DATA ENTRY.
 
      I.T. as the interface to advanced logistics systems
      (ALS, STALOGS), maintenance systems, etc.  The stress
      would be on stand-alone use for periods of time (e.g.,
      while on flight line entering maintenance data), and
      on easy, friendly interaction for large numbers of
      relatively computer-naive personnel (privates, sergeants,
      airmen) with high rotation rates.  Voice-response is an
      interesting possibility for "hands-busy" maintenance.
      We know of interest in this application area by appropriate,
      cooperative officers at the A.F. Data Systems Design Center.
 
 
      2(d).  COMPUTER-AIDED INSTRUCTION.
 
      Use of "intelligence" in I.T. in the form of models
      of systems, etc. being studied (e.g., for maintenance,
      repair, debugging).  Cf. John Seely Brown's work at BBN
      and Irvine in this regard.  Applications in this area
      should be coordinated with Col. Kibler's HRRO research program.
 
 
      2(e).  TASK MANAGEMENT.
 
      Use of an I.T. to help manage a complex procurement or
      scheduling process.  (For example, in a System Program
      Office (SPO) in the Air Force.)  Emphasizes task and
      environment models in the I.T. to do "intelligent"
      exception reporting and monitoring.  Relevant to this
      application area is the AFSC AMIS (Acquisition Management
      Information System) being developed at Wright-Patterson;
      what are its deficiencies; how can I.T. give an order-
      of-magnitude improvement?
 
      A related application area is the use of I.T. for a high-
      level management interface.  Stress would be on large-screen
      graphics, voice I/O, very natural, easy-to-learn interface
      (or else one tailored to an "information specialist" who
      manipulates data for high-level managers).
 
      2(f).  OFFICE AUTOMATION.
 
      Coordination with A. F. Project ADMIN and related efforts.
      This is like MST, but with more carefully tailored arguments
      about unique DoD requirements.  (I won't rehash all the
      MST data here.)
 
      2(g).  INTELLIGENT REACTIVE COMMUNICATION CHANNELS.
 
      This is really a component of other applications.  Use
      I.T.'s intelligence to provide adaptive behavior in
      communication (e.g., in tactical situations).  For example,
      I.T. and a remote system negotiate bandwidth, type of
      burst mode, frequency (dynamically changing according to
      subtle algorithms), etc. -- and especially do these things
      adaptively to react to jamming, noise, and other environmental
      effects, like the need for periods of radio silence.
 
 3.  Committees, etc.
 --------------------
 
 There are two useful categories of people helping with the I.T.
 planning effort:  (1) technical contributors; (2) advisory committee
 members.  To date, the following people have been, or are intended
 to be, drawn into the web:
 
      3(a).  TECHNICAL CONTRIBUTORS.
 
 
           i.  Oliver Selfridge, MIT
 
           He intends to drop by Rand on Aug. 1; if not possible,
           we'll get to Cambridge to talk with him;
 
           ii.  Joe Markowitz, SAI
 
           ARPANET communication until his return from European
           trip;
 
           iii.  Tim Standish, BBN
 
           Can't give large amount of time, but will provide
           think-papers on programming languages for interfaces.
 
           iv.  Dick Watson, SRI
 
           In NLS group there, to get their inputs.  Will
           see him at SRI next week.
 
           v.  Alan Kay, Xerox PARC
 
           To blend their ALTO and Dynabook ideas into
           I.T. specs.  Will see him next week.
 
           vi.  Ed Feigenbaum, Stanford
 
           Consulting with us on A.I. applications,
           intelligence area, other general wisdom.
 
      3(b).  ADVISORY COMMITTEE.
 
      Currently being formed.  Expected to include:
 
           i.  NSA representative (Rosenbloom?)
 
           ii.  Bob Taylor, Xerox PARC
 
           iii.  representative from General Crawford's
                 office in Army Computer Systems Command
 
           iv.  representative from A.F. Data Systems
                Design Center
 
           v.  representative from DCA (chief scientist --
               van Trees?)
 
           vi.  Industry representative (e.g. IBM)
 
 
 
 Your feedback to these categories, concepts, and plans is vital.
 I hope these seem like profitable directions to you.
 
                                        Bob Anderson
 :lmc