Original Source
..
Research Recommendations
[Presented to international conference on
Technology, Knowledge & Society sponsored by
University of California, Berkeley on February 18-20, 2005 and
in support of
Implementing the Engelbart Hypothesis: Discussing the notion
of augmenting human intellect
(
reviewed
March 30, 2005)]
..
Objectives for a Research Program
The report has put forth the hypothesis that the intellectual
effectiveness of a human being is dependent upon factors which are subject
to direct redesign in pursuit of an increase in that effectiveness. A
conceptual framework is offered to help in giving consideration to this
hypothesis, and an extensive and personalized projection into possible
future developments is presented to help develop a feeling for the
possibilities and promise implicit in the hypothesis and conceptual
structure.
..
If this hypothesis and its glowing extrapolations were borne out in
future developments, the consequences would be most exciting and assumedly
beneficial to a problem-laden world. What is called for now is a test of this
hypothesis and a calibration on the gains if any that might be realized by
giving total-system design attention to human intellectual effectiveness. If
the test and calibration proved to be favorable, then we can set to work
developing better and better augmentation systems for our problem solvers.
..
In this light, we recommend a research program approach aimed at (Goal 1)
testing the hypothesis, (Goal 2) developing the tools and tech niques for
designing better augmentation systems, and (Goal 3) producing real-world
augmentation systems that bring maximum gains over the coming years to the
solvers of tough, critical problems. These goals and the resulting design for
their pursuit are idealized, to be sure, but the results nonetheless have
valuable aspects.
..
Basic Research Recomendations
This should be an empirical approach on a total-system basis--i.e., doing
coordinated study and innovation, among all the factors admitted to the
problem, in conjunction with experiments that provide realistic action and
interplay among these variables. The question of limiting these factors is
considered later in the section. The recommended en vironment for this
empirical, total-system approach, is a laboratory providing a
computer-backed display and communication system of the general sort
described in Section III-B. There should be no stinting on the capabilities
provided--it is very important to learn what value any given artifact
feature may offer the total system, and the only way to learn the value is
to experiment with the feature. At this point no time will be taken to
develop elaborate improvements in the art of time sharing, to provide
real-time service to many users. This kind of development should be done as
separate, backup work. The experimental lab should take the steps that are
immediately available to provide all the service to the human that he needs
in the experimental environment.
..
Where economy demands that a computer not be idle during the time the
augmented subject is not using it (which would be a rather large net fraction
of the time, probably), and where sharing the computer with other real-time
users for which demand delays are a problem, then the only sharing that should
be considered is that with off-line computations for which there are no
real-time service demands to be met. The computer can turn away from off-line
users whenever the on-line worker needs attention of any sort.
..
Whom to Augment First
The experimental work of deriving, testing, and integrating innovations
into a growing system of augmentation means must have a specific type of
human task to try to develop more effectiveness for, to give unifying focus
to the research. We recommend the particular task of computer programming
for this purpose--with many reasons behind the selection that should come
out in the following discussion. Some of the more direct reasons are
these:
..
- The programmer works on many problems, including large and
realistic ones, which can be solved without interaction with other
humans. This eases the experimentalproblem. and realistic problems for
the programmer to solve can be posed for experimental purposes that do
not involve large amounts of working and reference in formation. This
also eases the experimental problem.
..
- Much of the programmer's working data are computer programs (he
also has, we assume, his own reasoning and planning notes), which have
unambiguous syntactic and semantic form so that getting the computer to
do useful tasks for him on his working data will be much
facilitated--which helps very much to get early experience on the value
a human can derlve from this kind of computer help.
..
- A programmer's effectiveness, relative to other programmers, can
probably be measured more easily than would be the case for most other
complex-problem solvers. For example, few other complex solutions or
designs beside a program can so easily be given the rigorous test of
'Does it actually work?'
..
- The programmer's normal work involves interactions with a
computer (although heretofore not generally on-line), and this will
help researchers use the computer as a tool for learning about the
programmer's habits and needs.
..
- There are some very challenging types of intellectual effort
involved in programming. Attempting to increase human effectiveness
therein will provide an excellent means for testing our
hypothesis.
..
- Successful achievements in evolving new augmentation means which
significantly improve a programmer's capability will not only serve to
prove the hypothesis, but will lead directly to possible practical
application of augmentation systems to a real-world problem domain that
can use help.
..
- Computer programmers are a natural group to be the first in the
'real world to incorporate the type of augmentation means we are
considering. They already know how to work in formal methodologies
with computers, and most of them are associated with activities that
have to have computers anyway, so that the new tech niques, concepts,
methods, and equipment will not seem so radical to them and will be
relatively easy for them to learn and acquire.
..
- Successful achievements can be utilized within the
augmentation-research program itself, to improve the effectiveness of
the computer programming activity involved in studying and developing
augmentation systems. The capability of designing, implementing, and
modifying computer programs will be very inlportant to the rate of
research progress.
..
Workers in an augmentation-research laboratory are the most natural
people in the world to be the very first users of the augmentation means
they develop, and we think that they represent an extremely important group
of people to make more effective at their work.
..
Basic Regenerative Feature
The feature brought forth in Reason 9 above is something that offers
tremendous value to the research objectives--i.e., the feeding back of
positive research results to improve the means by which the researchers
themselves can pursue their work The plan we are describing here is
designed to capitalize upon this feature as much as possible, as will be
evident to the reader as he progresses through this section. This
positive-feedback (or regenerative) possibility derives from the facts
that: (1) our researchers are developing means to increase the
effectiveness of humans dealing with complex intellectual problems, and (2)
our researchers are dealing with complex intellectual problems. In other
words, they are developing better tools for a class to which they
themselves belong. If their initial work needs the unifying focus of
concentrating upon a specific tool, let that tool be one important to them
and whose improvement will really help their own work.
..
Tools Developed and Tools Used
This close similarity between tools being developed and the tools being
used to do the developing, calls for some care in our terminology if we
want to avoid confusion in our reasoning about their relationship.
'Augmentation means' will be used to name the tools being developed by the
augmentation research. 'Subject lnformation' will be used to refer to
description and reasoning concerned with the subject of these tools (as
opposed to the method of research), and 'subject matter' will refer to both
subject information and physical devices being incorporated as artifacts in
the augmentation means being developed. 'Tools and techniques' will be used
to name the tools being used to do that research, and are likely here to
include special additions to language, artifact, and methodology
that particularly improve the special capabilities exercised in doing the
research.
..
An integrated set of tools and techniques will represent an art of
doing augmentation research. Although no such art exists ready-made for
our use, there are many applicable or adaptable tools and techniques to be
borrowed from other disciplines. Psychology, computer programming and
physical technology, display technology, artificial intelligence,
industrial engineering (e.g., motion and time study), management science,
systems analysis, and information retrieval are some of the more likely
sources. These disciplines also offer initial subject matter for the
research. Because this kind of diagramming can help more later on, we
represent in Fig. 3 the situation of the beginning research drawing upon
existing disciplines for subject matter and tools and techniques.
..
The program begins with general dependence upon other, existing dis
ciplines for its subject matter (solid arrow) and its tools and tech niques
(dashed arrow). Goal 1 has been stated as that of verifying the basic
hypothesis that concerted augmentation research can increase the intellectual
effectiveness of human problem solvers.
..
Fig. 3 Initial Augmentatiuon-Research Program
..
Research Plan for Activity A l
The dominant goal of Activity A 1 (Goal 1, as in Fig 3) is to test our
hypothesis. Its general pursuit of augmenting a programmer is designed to
serve this goal, but also to be setting the stage for later direct pursuit of
Goals 2 and 3 (i.e., developing tools and techniques for augmentation research
and producing real-world augmentation systems).
..
Before we discuss the possible subject matter through which this research
might work, let us treat the matter of its tools and techniques. Not too long
ago we would have recommended (and did), in the spirit of taking the long-range
and global approach, that right from the beginning of a serious program of this
sort there should be established a careful and scientific methodology.
Controlled experiments, with special re search subjects trained and tested in
the use of experimental new aug mentation means, careful monitoring,
record-keeping, and evaluative procedures, etc. This was to be accompanied by
a thorough search through disciplines and careful incorporation of useful
findings.
..
Still in the spirit of the long-range and global sort of planning, but
with a different outlook (based, among other things, upon an increased
appreciation for the possibilities of capitalizing upon regeneration), we would
now recommend that the approach be quite different. We basically recommend A 1
research adhering to whatever formal methodology is required for (a) knowing
when an improvement in effectiveness has been achieved, and (b) knowing how to
assign relative value to the changes derived from two competing
innovations.
..
Beyond this, and assuming dedication to the goal, reasonable maturity,
and plenty of energy, intelligence, and imagination, we would recommend turning
loose a group of four to six people (or a number of such groups) to develop
means that augment their own programming capability We would recommend that
their work begin by developing the capability for composing and modifying
simple symbol structures, in the manner pictured in Section III-B-2, and work
up through a hierarchy of intermediate capabilities toward the single
high-level capability that would encompass computer programming. This would
allow their embryonic and free wheeling "art of doing augmentation research' to
grow and work out its kinks through a succession of increasingly complex system
problems--and also, redesigning a hierarchy from the bottom up somehow seems
the best approach
..
As for the type of programming to tell them to become good at--tell them,
"the kind that you find you have to do in your research." In other words, their
job assignment is to develop means that will make them more effective at doing
their job. Figure 4 depicts this schematically, with the addition to what was
shown in Fig. 3 of a connection that feeds the subject-matter output of their
research (augmentation means for their type of programming problems) right back
into their activity as improved tools and techniques to use in their research.
..
Fig. 4 Regeneration
If they are making head way, it won't take any carefully worded criterion
of effectiveness nor any great sophistication in measurement technique to
tell that they are more effective with the augmentation means than
without--being quicker to "design and build" a running program to meet given
processing specifications or being quicker to pick up a complex existing
program, gain comprehension as necessary, and find its flaws or rebuild it.
On the other hand, if no gains are really obvious after a year or so, then
it is time to begin incorporating more science in their approach. By then
there will be a good deal of basic orientation as to the nature of the
problem to which "science" is to be applied.
..
What we are recommending in a way is that the augmented capability
hierarchy built by this group represent more a quick and rough scaffolding than
a carefully engineered structure. There is orientation to be derived from
climbing up quickly for a look that will be of great value. For instance, key
concepts held initially, that would have been laboriously riveted into the
well-engineered structure, could well be rendered obsolete by the "view"
obtained from higher in the hierarchy. And besides, it seems best to get the
quick and rough improvements built and working first, so that the research will
benefit not only from the orientation obtained, but from the help that these
improvements will provide when used as tools and techniques to tackle the
tougher or slower possibilities. As progress begins to be made toward Goal
l,the diagram of Fig. 3 will become modified by feeding the subject-matter
output (augmentation means for computer programmers) back into the input as new
tools and techniques to be used by the researchers.
..
We would suggest establishing a sub-activity within A 1, whose purpose
and responsibility is to keep an eye on the total activity, assess and evaluate
its progress and try to provide orientation as to where things stand and where
attention might be beneficial.
..
A few words about the subject matter through which Activity A 1 may
progress. The researchers will think of simple innovations and try them in
short order--and perhaps be stimulated in the process by realizing how handy
some new feature would be that would help them ship up trial processes in a
hurry. They will know of basic capabilities they want to work toward for
structuring their arguments, their planning, their factual data, etc., 50 that
they can more easily get computer help in developing them in analyzing and
pursuing comprehension within them and in modifying or extending them. They
will try different types of structuring and see how easy it is to design
computer processes to manipulate them or composite processes to do total useful
work with them.
..
They can work up programs that can search through other programs for
answers to questions about them--questions whose answers serve the processes of
debugging, extending, or modifying. Perhaps there will be ways they adopt in
the initial structuring of a program--e.g., appending stylized descriptive cues
here and there--that have no function in the execution of that program, but
which allow more sophisticated fact retrieval therein by the computer. Perhaps
such cue tagging would allow development of programs which could automatically
make fairly sophisticated modifications to a tagged program. Maybe there would
evolve semi-automatic "super-compilers," with which the programmer and the
computer leap-frog over the obstacles to formulating exact specifications for a
computer (or perhaps composite) process and getting it into whatever
programming language they use.
..
A Second Phase in the Research Program
The research of A 1 could probably spiral upwards indefinitely, but once
the hypothesis (see Section IV-A) has been reasonably verified and the first
of our stated objectives satisfied, it would be best to re-organize the
program. To describe our recommendation here, let us say that two research
activities, A 2 and A 3, are set up in place of A 1. Whether A 1 is split,
or turned into A 2 and a new group formed for A 3, does not really matter
here--we are speaking of separate activities, corresponding to the
responsible pursuit of separate goals, that will benefit from close
cooperation.
..
To Activity A 2 assign the job of developing augmentation means to be
used specifically as tools and techniques by the researchers of both A 2 and A
3. This establishes a continuing pursuit for Objective 2 of Section IY-A. A 2
will now set up a sub-activity that studies the problems of all the workers in
A 2 and A 3 and isolates a succession of capabilities for which the research of
A 2 will develop means to augment. Activity A 2 should be equipped with the
best artifacts available to an experimental laboratory.
..
To Activity A 3 assign the job of developing augmentation systems that
can be practically adopted into real-world problem situations. This provides a
direct and continuing pursuit of Goal 3 of Section IV-A. It is to be assumed
that the first real-world system that A 3 will design will be for computer
programmers. For this it might well be able to clean up the "laboratory model"
developed in A 1, modify it to fit the practical limitations represented by
real-world economics, working environments, etc., and offer it as a prototype
for practical adoption. Or Activity A 3 might do a redesign, benefiting from
the experience with the first model.
..
Activity A 3 will need a subactivity to study its potential users and
guide the succession of developments that it pursues. Activity A 2 in its
continued pursuit of increased effectiveness among workers in idealized
environment, will be the source for basic subject matter in the developments of
A 3, as well as for its tools and techniques. From the continuously expanding
knowledge and developments of A 2, A 3 can organize successive practical
systems suitable for ever more general utilization.
..
We have assumed that what was developed in A 1 was primarily language and
methodology, with the artifacts not being subject to appreciable modification
during the research. By this second phase, enough has been learned about the
trends and possibilities for this type of on-line man-computer cooperation that
some well-based guidance can be derived for the types of modifications and
extensions to artifact capability that would be most valuable. Activity A 2
could continue to derive long-range guidance for equipment development, perhaps
developing laboratory innovations in computers, display systems, storage
systems, or communication systems, but at least experimenting with the
incorporation of the new artifact innovations of others.
..
An example of the type of guidance derived from this research might be
extracted from the concepts discussed in Section-C-5 (Structure Types). We
point out there that within the computer there might be built and manipulated
symbol structures that represent better images of the concept structures of
interest to the human than would any symbol structure with which the human
could work directly. To the human, the computer represents a special
instrument which can display to him a comprehensible image of any
characteristic of this structure that may be of interest. From our conceptual
viewpoint, this would be a source of tremendous power for the human to harness,
but it depends upon the computer being able to 'read' all of the stored
information (which would be in a form essentially incomprehensible to a human).
Now, if this conjecture is borne out there would be considerably less value in
micro-image information-storage systems than is now generally presumed. In
other words, we now conjecture that future reference information will be much
more valuable if stored in computer-sensible form. The validity of this and
other conjectures stemming from our conceptual framework could represent
critical questions to manufacturers of information systems.
..
It is obvious that this report stems from generalized 'large-view'
thinking. To carry this to something of a final view, relative to the research
recommendations, we present Fig. 5, which should be largely self-explanatory by
this time. Activity A 2 is lifting itself by the bootstraps up the scale of
intellectual capability, and its products are siphoned to the world via A 3.
Getting acceptance and application of the new techniques to the most critical
problems of our society might in fact be the most critical problem of
all by then, and Activity A 4 would be one which should be given special help
from A 3.
..
There is another general and long-range picture to present. This is in
regard to a goal for a practically usable system that A 3 would want to develop
as soon as possible. You might call this the first general Computer
Augmentation System--CAUG-I (pronounced "cog-one").
..
Fig. 5: A Total Program
Suggested relationship among the major
activities in achieving the stated objective (essentially, of significantly
boosting human power in A 4 and U 1). Solid lines represent subject
information or artifacts used or generated within an activity, and dashed
lines represent special tools and techniques for doing the activity in the
box to which they connect. Subject product of an activity (output solid) can
be used as working material (input solid) or as tools and techniques (input
dashed). Tools and techniques as used or needed in an activity (output
dashed) can be used either to work on (input solid) or as tools and
techniques to work with (input dashed).
..
It would be derived from what was assessed to be the basic set of
capabilities needed by both a general-problem-solvlng human and an
augmentation researcher. Give CAUG-I to a real-world problem solver in
almost any discipline, and he has the basic capabilities for structuring his
arguments and plans, organizing special files, etc., that almost anyone
could expect to need. In addition to these direct-application on
capabilities, however, are provided those capabilities necessary for
analyzing problem tasks, developing and evaluating new process capabilities,
etc., as would be required for him to extend the CAUG-I system to match to
the special features of his problem area and the way he likes to work.
..
In other words, CAUG-I represents a basic problem-solving tool kit, plus
an auxiliary tool-makers tool kit with which to extend the basic tool kit to
match the particular job and particular worker. In subsequent phases,
Activity A 3 could be turning out successive generations (CAUG-II, CAUG-III,
etc.) each incorporating features that match an ever-more-powerful
capability hierarchy in an ever-more-efficient manner to the basic
capabilities of the human.