25 August 1961
PROGRAM ON HUMAN EFFECTIVENESS
STANFORD RESEARCH INSTITUTE
Human beings face ever more complex and urgent problems, and their effectiveness in dealing with these problems is a matter that is critical to the stability and continued progress of society. A human is effective not just because he applies to a problem a good degree of native intelligence or physical strength, with a full measure of motivation and purposefulness, but also because he makes use of good tools, good methods, and good strategies. These latter may be modified for increased effectiveness, and Stanford Research Institute feels that it has a good basic plan for a program to systematically evolve such modifications. It is a fairly ambitious plan, springing from both a realization that this is an extremely important area of research that warrants a strong, coordinated attack, and a feeling that there exist some very promising possibilities around whose pursuit an energetic and apparently sound program can be launched.
In planning our approach, we have tried to pick a practical route along which we could pursue our objective in a manner that is strongly fundamental but yet is designed to produce practically useable results at most of the steps along the way. To do this, we have relied upon a number of models and generalizations that have evolved in the basic work that has gone on at SRI for more than two years (partial support of this work from the AFOSR since early 1961 is gratefully acknowledged).
In the following section, we shall outline some of these generalizations briefly, to make more meaningful the specific plans described later.
II. GENERALIZATIONS AND MODELS
A. Initial Objective for the Program on Human Effectiveness
Statement of Objective: To develop means for making humans maximally effective as comprehending solvers of problems.
Discussion of Statement: "Maximally effective" means we truly seek the maximum, and want to understand what factors limit the human's capability at any given stage of our progress. "Comprehending" means that we want not merely blind manipulation of things or symbols, as a machine can do (and someday will do better than
humans in any given situation), but we mean that we want the human to have maximum possible comprehension of the relevant concepts in the problem situation. By "solvers" we mean more than derivers of possibilities -- we mean responsible initiators of action. By "problems" we by no means refer only to formalized problems in mathematics or science. We refer in general to any of the problems that man faces in trying to understand his physical universe, his society, and himself, and in trying to pursue survival and improvement of the latter two within the former. We speak specifically of the complex and real-time problems of the legislator, executive, administrator, researcher, educator, etc., and we are concerned with increasing the speed with which problems can be solved, with increasing the complexity of the problems that a human can comprehendingly and successfully solve, and with increasing the quality of the solutions.
Comments: We assume that part of the "problem" for our problem solver is to clear the way through the trivia and red tape of everyday life to make time and "mental space" available for his higher-level problems. To meet our objective, we shall have to improve human effectiveness at these lower-order jobs, and so the development of this program should be useable for increasing human effectiveness at all levels of endeavor. While our long-range basic objective is associated with very high-power capability, we expect to work toward this in stages that should provide practical developments for specialized, realworld jobs. The program is designed so that this "spin-off" development work may be incorporated without interfering with main-effort progress toward the long-range objective.
B. Organizations and Individuals
We realize that the really complex and pressing problems of our society are usually handled by an organization of humans, in which the problems are artfully decomposed into components manageable by succeedingly smaller groups of humans, until finally individual humans are faced with some individually manageable portion of either the original problem or of the internal organizational problem. This represents a sophisticated means by which basic human resources can be brought to bear successfully upon complex problems, and the ultimate in human effectiveness will undoubtedly involve increased sophistication in the means by which humans cooperate.
However, in picking a "practical and fundamental" route of attack on our objective (of increasing human effectiveness), we choose to postpone consideration of organized human cooperation in favor of first pursuing increased effectiveness in individual humans. We are considering rather radical changes in the means by
which humans do their problem-solving work, which leads to two good reasons for turning our attention first upon the individual. For one, this gives us a more specific and manageable research domain in which to learn about mixing radical means with responsible humans. And second, it seems obvious that one must first learn how to get individuals to use radical means before one can expect to get groups to do so. This latter is fortified by realizing that the radical means for improving group effectiveness will be quite dependent upon radical developments in the means by which individuals handle information and tackle their problems.
C. System Model of an Individual Problem Solver
Initially, at least, ours is an engineering approach toward the task of redesigning different parts of a functional system in order to increase its performance. We shall assume that our human is willing to learn and relatively eager to become more effective, but our current model leaves out consideration of such factors as the spiritual and emotional characteristics of the human. This doesn't mean that we don't believe in them -- it is rather that we don't know how yet to incorporate them into a suitable functional model. We can only promise that, as our Program matures (and we with it), every factor that seems likely to bear with any significance upon the human-effectiveness problem will be given serious consideration.
Our model stems from the picture of the "neuro-muscular" human, with his basic sensory, mental, and motor capabilities, being matched to the problems he must face by what we call "augmentation means." These means have for the most part been evolved within the culture in which be is born, and he has been training in their use since childhood. We refer here to all of the tricks, tools, techniques, methods, strategies, special skills, etc. that the individual human can bring to bear upon his struggle with his problems. Our initial approach toward making the human more effective is to try to do a coordinated re-vamping of these augmentation means.
We have found four useful categories with which to divide these augmentation means:
We then have a functional model of a trained human, with his Language, Artifacts and Methodology, as the problem-solving system whose effectiveness we want to improve. We are going to try to make coordinated improvements upon the Language, Artifacts, and Methodology that he uses, and the Training that he is given.
In developing our method of approach, we have found it of further use to consider our problem-solving system from a "process" point of view. When this system executes the total process of solving a problem, it generally does this by organizing the sequential execution of a fair number of sub-processes. For the execution of each of these sub-processes, we can generally say the same thing -- i.e., that its execution is accomplished by organizing the sequential execution of a number of "lower-order" processes. Eventually, if we pursued this down through the levels, examining the sub-sub-processes from which the sub-processes were composed, etc., we would find a primitive set of processes that represents the basic neuro-muscular capabilities of the naked, untrained human.
The capability for executing this given total problem-solving process can then be pictured as stemming from a complex hierarchy of process capabilities, with the hierarchy being an intricate composition of primitive human capabilities and the augmentation means. At any given level, the execution of a process represents a problem. For some processes this execution is so routinized that conditioned responses can relieve almost all associated effort from the human's higher centers. The execution of other processes tax the human to the utmost. But it is not assumed that one process taxes human capability more than a second process just because the latter might be one which is utilized within the former.
When the human faces the execution of a process at some given level (and thus faces a problem), his augmentation means are structured for him to a considerable extent by the array of processes that he can utilize. The names of these processes are an important part of his Language, the ways in which he can coordinate their use are an important part of his Methodology, and the physical tools available to implement them are an important part of his Artifacts. It thus becomes apparent that a really serious attempt at basic improvement in his effectiveness at a given level calls for a coordinated redesign of the array of processes which are utilized at that level, with the aim of improving the human's effectiveness at
execution of these processes. But the same statement will hold true when attention is turned to any of these processes, and we will see that the processes at the next lower level require coordinated redesign. The eventual conclusion is, of course, that our objective calls for the coordinated re-design of the entire process hierarchy above the primitive human capabilities -- which is another way to picture the need to redesign the augmentation means.
There is no reason to think that the objective of increased human effectiveness could not be served by improving these primitive human capabilities (i.e., improving the trainable limits of human performance at certain basic operations). We don't happen now to be considering the inclusion of such research in our program -- but we don't rule it out either.
Considering the hierarchy of processes which rests upon the human's primitive capabilities, and upon his Artifacts, and considering that the nature of this hierarchy is still firmly based upon essentially manual methods for execution of the processes, provides us with a very challenging and stimulating question: What might be the possibilities for building up a new structure of processes which a human could use, based upon his primitive capabilities and the best Artifacts which our society can offer, where the processes which he could organize and the concepts with which he could deal would be no longer limited by the speed, accuracy, and fatigue factors associated with human process execution?
III. OUR PROPOSED METHOD OF APPROACH
We feel that there are radical possibilities for re-designing the augmentation means, and that large payoffs are quite probable, but to step right in and re-design the entire hierarchy of processes for a high-level type of human problem solver would be much too imposing a task for us (at our current level of effectiveness in solving complex problems). This leaves a choice, for launching our program, of either tackling the over-all problem with much-lessthan-radical changes in mind, or keeping the radical possibilities in mind but tackling a more-restricted problem-solving-effectiveness task. We feel that the latter is much the better way to proceed, and our method of approach is designed accordingly.
A. Summary of our Plan
Our most active research effort will involve a fairly sophisticated computer-based laboratory facility, which is designed to provide rather elegant real-time services to human Subjects, but which can very easily be "throttled down" to simulate almost any leas-elegant facility. Our Subjects shall be people who are initially going about the execution of processes which represent the lower levels of the process hierarchy that we picture developing someday for our high-level human problem solver. These shall be processes that tax human capability quite satisfactorily, but yet involve a very restricted amount of information. Our researchers shall develop Language and Methodology to improve the effectiveness of the Subjects at the execution of these low-level processes.
This approach will give us valuable experience in being radical about the modifications introduced in the way humans operate, and it will serve as a sort of calibration run for the entire future program. No one can tell from our present standpoint what to expect in the way of improved effectiveness for any particular proposed increase in the sophistication of processes and training given to humans--at least for those changes associated with the future trends that we anticipate. By pushing hard on an appropriate initial human-task area, we hope to get a feeling for the sort of payoffs and problems to expect downstream. Also, we expect to be able to expand the scope of the human-task-study area in a logical and nicely controlled fashion to reach successive stages where practical application of radical techniques becomes feasible for real-world special utilization. So, besides
our basic research team that will be concentrating on improving effectiveness within the limited-task area, we plan to have another team that is scouting the path between there and real-world utilization,--i.e., that is laying the groundwork for early practical utilization of the knowledge and techniques evolving from the more basic research.
B. Laboratory Facilities
Initially we want to equip at least one working station at which a Subject can work comfortably for extended periods. He shall be provided with equipment and associated primitive processes for manipulating symbols (all available to him at maximum convenience, on demand, with essentially no waiting period) that represents the near ultimate of what our technology can offer, and that far exceed in speed and power any that previously have been integrated into a human's operational hierarchy of processes. There should be provisions for both direct and indirect observation of his behavior to a very complete extent.
We intend to assemble this initial system with commercially available equipment wherever possible. While we conduct Language, Methodology and Training research based upon this "Artifact" facility, we shall be doing concurrent research toward improvements in our laboratory facilities--some of which shall provide improved Artifacts for the Subjects, and some of which will make our experimentation easier.
A high-speed digital computer shall be at the heart of the facilities. The work station shall initially have a standard typewriter keyboard with which the human can communicate to the computer. A cathode-ray-tube display system will also be provided, with high-speed capability for arraying something near full-page presentations of alphabetic, numeric, and special symbols. It shall also be capable of displaying line drawings, graphs, curves, etc. There may also be typewriter, or other hard-copy, printout facility, Display and printout facility will be controlled directly by the computer, with or without the direct specification of this action by the Subject. We plan to provide enough storage capacity, with a combination of high-speed internal storage and rapid-access (rapid to a human's time scale) auxiliary storage, so that there will be no anticipated restriction on what the Subject can have the computer "remember" for him--within the human-task domain in which the Subject shall be working.
The computer shall have other input-output facilities than those used by the Subject. Some use of this will be by researchers in this program who are getting computer help in studying the actions of the Subject, and other uses will
be to take normal advantage of this computing capability when that capability is not in demand by a Subject. We expect time-sharing provisions to be developed so that the computer can be efficiently used, by having it turn to these moregeneral computing tasks whenever its full capability is not required by the Subject(s). As the program develops, it is anticipated that more stations would be needed, and that the time-sharing provision would be needed to allow flexible arrangements for dealing with co-incident demands by Subjects.
C. Augmentation Research
This is the aspect of the program upon which we initially will concentrate the most effort. That labelled Search and Planning, and described in the next section, essentially has already been under way for several years. It will grow more slowly, but may well breed considerable activity in the future.
We might say that the augmentation research project has two basic aims. We want to see to what relative levels of effectiveness a human might actually be lifted in the execution of non-trivial problem-solving processes. We also want to learn how to co-ordinate this kind of research, and to become oriented as to the problems and possibilities that would be involved in extending all parts of this type of augmentation development toward the benefit of practical, higher-level problem solving.
We have chosen a type and level of problem-solving process whose execution we feel will provide an excellent "human task area'' within which we initially can concentrate upon making humans more effective. Upon primitive human capabilities that we shall assume (and about which there is yet much to learn), and upon the basic "mechanical" processes that our laboratory facilities inherently will provide, we shall construct "shallow" but non-trivial experimental hierarchies of processes which we shall train Subjects to execute. These experimental hierarchies will be used by Subjects in executing test processes, and the objective will be to design hierarchies that will maximize human effectiveness at these test processes.
We plan to have Subjects do all of their non-cerebral symbol manipulation (i.e., all that isn't done in their heads) with the computer and the display. We have two reasons for this. First, we are seriously trying to get a feeling for the possibilities and payoffs associated with the ultimate in sophisticated-Artifact utilization. And second, we expect thus to be able to develop a very powerful observation, measurement, and analysis system, based upon the computer's being "aware" of all of the overt steps taken by a Subject, upon its capability
to record selectively and almost "intelligently" any of the Subject's overt-step history, and upon its ability to be programmed for either real-time or off-line analysis of the Subject's performance.
The heart of a complete and "fully automated" process hierarchy--one such as we picture someday being utilized by a specially trained and highly effective human solver of general real-life problems--would be a set of fundamental processes by which the human controls the activity of his symbol-manipulating (informationhandling) help. There is a tremendous amount of "speed, agility, and horsepower" available for us to apply in a design for a "personal, intellectual flitmobil." Our feeling is that we should first learn how a human can effectively control such a flitmobil to become quick, deft, and powerful in basic real-time processes such as moving, disassembling, assembling, rearranging, or inspecting structures of information. The very things that it can thus help a Subject do, if he learns how to "steer" his device well, are what can be advantageously utilized in the next step, where we learn how to get this device to help him in the process of training it to follow sequences of such primitive processes on its own, to execute automatically some higher-order process. If our Subjects become really effective at these two types of activity (steering the device through the processes it can do, and training it to do new processes), then we could turn them loose to evolve a complete process hierarchy and become general problem solvers on their own. In a sense, this is what we intend to do--to help some humans pull themselves up by their own bootstraps and show the rest of us the way.
The task of "training" this "device" to do something helpful for the Subject, would be a standard type of sub-problem that he would face in the process of solving more general problems. This training task can have many of the characteristics of general problems, and this is the area of human proficiency in which we will initially apply most of our research efforts toward making humans more effective. We wanted a human-task area for our initial increased-effectiveness research which would tax human capability significantly, but would be of limited enough scope to allow us to make an all-out attack upon it, and the type of task which we chose is that of the human's getting available information-handling processes organized to accomplish a higher-order process. The level we happened to choose, for a starting place, is at the bottom of the (future) process hierarchy which we deem likely to enable a human to be a very effective general problem solver.
If, in this initial experiment, we cannot manage to make humans significantly more effective at this type of task (and we shall have very good, objective methods of measuring effectiveness), then we will have strong reason to believe that we are entirely on the wrong track. If, however, we do manage to make humans significantly more effective at this type of task, then we will be much encouraged, will have gotten some very valuable experience in doing this kind of research, and will have something in the nature of a means of calibrating future expectations. As a bonus in this latter case, we shall also be positioned ideally for taking the next step toward our objective--which would be to develop the next-higher set of processes by which a human can do useful information-handling tasks associated with his problem-solving processes. Actually, this is planned to be a continuing, well-controlled sequence of development, in pursuance of our ultimate objective.
We plan to provide for three basic types of research function. One function is to synthesize innovations in the Subjects, Artifacts, Language, and Methodology, giving him new tools to try at his tasks. Another function is to analyze the Subjects and their augmentation means (system analysis), and thus evaluate the worth of various innovations and uncover bottlenecks where new innovation would be useful. The third function is to develop good techniques for training Subjects in the use of this type of augmentation means. We have indicated how the analysis function can be aided significantly by using the computer to monitor, record, and evaluate. We plan to get aid for the third function from the same source.
For a number of reasons, it seems very attractive to us to use "programmed tutoring" to accomplish the Subjects' training. It will be a natural facility and subject-matter environment in which to use our approach. A subject can be trained by the computer while sitting at the same work Station at which he will use the results of his training. We can have standard ways to train a new Subject, and we can get as complete a record as we wish of the Subject's learning history associated with any given skill, for later consideration in the evaluation of different augmentation means. Since these augmentation means are expected to represent a radical departure from what humane now use, to put our developments to practical use later on will require that it be practical to train humans in their use, and our ready-made teaching-machine techniques will be a very valuable adjunct to our developments. Also, the process of analyzing these augmentation means from the point of view of extracting concepts in best-learned sequences and
developing teaching programs, is expected to develop very valuable side effects in our total understanding and "feel" of what we are working with. Then there is also the consideration that learning is one process that a human has to be good at if he is to be effective in an all-round sense, and our augmentation means would not be complete if we did not give this man-machine team really good methods for its human to learn new information and new skills.
To recapitulate in explicit terms, our Subjects are going to be taught how to get the computer to manipulate symbols on the face of the display screen, to move symbols into and out of storage, and to do other primitive manipulation processes. They also will be taught how to present a sequence of statements to the computer that represent the directions for the computer's execution, upon proper cueing, of a symbol-manipulation process organized as a sequence of the above-mentioned primitive processes (ie., basic computer programming). Then our researchers will develop and provide to the subjects various experimental sets of next-level processes, together with appropriate clever symbology (or notation), terminology, and methodology, so designed as to let the Subjects make good use of the machine's help in becoming more effective at getting a program developed to control the computer in the execution of test processes. Our researchers shall determine the type of test processes for which the Subjects will organize programs for practicing or testing purposes. These test processes shall be chosen as representative of the next-higher level of processes that Subjects would want in their repertoire if the research were to continue in this line, but this is not critical to the experimentation of the moment. We would learn valuable things about making humans more effective by using test processes of any of a number of different sorts, but the sort we choose is critical to the "bootstrapping" concept which we feel to be a very important feature of our approach.
This bootstrapping concept can perhaps be clarified as follows. At any given stage of our research, we shall be developing "prototype processes" which the Subjects can use in their trial efforts at organizing a higher-order type of "test process". Whenever we have made significant progress in a particular stage of our research, then we shall have a particular set of prototype processes which can be incorporated into the Subject's hierarchy to make him particularly effective at composing a type of higher-order process--that type from which his test processes had been chosen. Now, assume that we move on to the next stage of our research, where we will learn how to make Subjects particularly effective
at composing still higher-order processes. The prototype processes which we now are developing and having them test are of the same type that we used for test processes in the preceding stage or stages of our research. So, we train our process-synthesizing researchers to utilize the process hierarchy that has been developed to date, and they become significantly more effective at writing the prototype processes with which we experiment at this stage of research.
Thus, as we make progress, we become more capable of making progress. And knowing that we shall ourselves make use of what we develop will stimulate us personally, as well as make us more realistic and cautious about what we commit the user of our prototype processes to doing. The emphasis on "trainability" of our processes, given by the programmed-tutoring part of our program activity, will tend to curb what might be a serious problem in this bootstrapping process--i.e., that of successively tailoring our processes to unique capabilities and characteristics of a small set of personally involved researchers, until our developments might become essentially unusable by other humans. Further curbing and guidance will come from the work of the Search and Planning part of the Program.
D. Search and Planning
The successive stages of augmentation research, as envisioned within the "bootstrap" concept, would require guidance as to the sequence of processes which are developed. We also want there to be planning, co-ordinated with this augmentation research, that leads to the development of practical means for providing increased effectiveness in economically useful real-life human tasks. These kinds of guidance and planning functions are to be provided by what might be called our Search and Planning activity.
We plan to give balanced consideration, in the guidance of' the augmentation research, both to working toward our primary goal most effectively (i.e., toward providing real help to high-level problem solvers), and to producing intermediate workable systems of augmentation means which can be valuable in real-world applications. It seems now as if these should be easy things to co-ordinate. We can already see that successful progress in augmentation research, as we plan it, could quite soon turn out means that would give a significant increase in human effectiveness in such job categories as computer programmer (a natural application, undoubtedly the earliest), cryptographer, teaching-machine-program writer, information specialist, etc.
We hope to build up a Development and Application activity, based upon the guidance of this Search and Planning activity, that can both (1) take direct products of the augmentation research and develop them in special systems for particular real-world application, and (2) develop new and useable augmentation systems that are based upon much less sophisticated equipment than that used in our initial push on augmentation research. We hope that the "calibration" we get from our augmentation research (with its very sophisticated equipment and other "means") will help us in developing such less-sophisticated means so that they would have a good chance of being compatible with the future trends in the evolution of practical means for making people more effective in given human-task areas.
Our Search and Planning activity will also need to keep oriented relative to questions, possibilities, consideration and the like for the overall Program, and keep a constant eye out for possible Program changes that would give us better progress toward our objective.