September 23, 1960
TO: File (Project A)
FROM: D. C. Engelbart
SUBJECT: A possible research activity toward a technique for teaching coordinate physical skills.
The current widespread interest in teaching machines presages a sizeable activity in research and application of automated techniques for teaching people what might be termed "mental skills" (or often called "verbal skills"). It seems to me that some very interesting possibilities exist for automating the teaching of physical skills, also. Work toward some aspects of this has been going on for years, with simulators, mockup procedural trainers, and most likely other techniques. For some twelve years I have harbored thoughts about possible techniques for coordinatephysicalskill training which I would like to present now for serious consideration.
GENERAL DISCUSSION OF TEACHING MACHINES
In words and form that may make a psychologist wince, here is my picture of what the mentalskill, teachingmachine art comprises. There are many different ways in which a given portion of a mental discipline may be broken down into small concepts from which the whole may be composed. For teaching someone this whole, we first decompose it into small concepts which are expected already to be familiar to the student, and then we cleverly compose from these elemental concepts a set of "larger" concepts. We strive to give the student enough practice in considering these larger assemblies so that his wonderful human nervous system soon fuses each of these compound affairs into a single conceptual entity. If one is responsible for teaching quite an extensive segment of some discipline, he may well have to build quite a number of these new secondary entities in the student's repertoire, and then proceed to use these as effective building blocks with which to define a set of tertiarylevel concepts that he seeks to develop into "fused entities" in the student's mind.
We may have a simple operational task to learn. In this case, there may be one fourthlevel concept, e.g., checking out a newly manufactured dithertron. This in turn may be composed of several thirdlevel concepts, such as checking the gridleak faucet, aligning the front wheels, calibrating the phonic dither, etc. These thirdlevel concepts are composed of numerous secondlevel concepts--primarily the special terms and associated pieces of equipment. The first level, the ones assumed known by the student at the outset, are concepts such as "look," "pick up," "reach," "turn until," "don't worry," "careful," "square," "hot end," etc.
I think that it is also important to this art to recognize that these firstlevel concepts are not only known to us as concepts, but they have been identified in our minds with particular code symbols or words. This is important, and in fact is essential, to our being able to teach conceptual matters of any complexity beyond the most primitive. We have to have a quick and direct way to lead the
File (Project A) 9/23/60
student's concepthandling device (his higher mental processes) through a sequence of already developed concepts from which we are composing a new, compound concept, so that the new association and relationships which are involved can impinge smoothly on the lower processes that somehow will fuse this into a single conceptual entity. My view is that it is essential, for an effective teaching technique, to have a neat encoding system which allows us to trigger within the student a desired sequence of elemental concepts in a form which is smooth enough and rapid enough to allow ready association and integration. One might say that a good part of our training time is spent in developing subsidiary coding systems to facilitate the later presentation of concepts and relationships whose communication otherwise would be too awkward to allow ready assimilation, response, and learning.
To elaborate, we might say that the limitations and characteristics of the human cognitive processes often result in our finding that the end concepts which we wish to teach a student can not be composed from his alreadypossessed concepts in assemblies which are simple enough for easy learning. In this case, it is more effective sometimes to develop first a special set of basic concepts (from his already-known ones) that will efficiently serve to compose the particular compound concepts whose learning by the student is our objective. Decomposition of the end concepts into the set of subconcepts and subsubconcepts, etc., which lends itself best to being learned by a particular set of student, is my picture of one of the basic tasks of the educator.
Another basic task is that of presenting the concepts to the student, with suitable stimulation, orientation, sequencing, timing, repetition, reinforcing, etc. In the last four of these, the use of automatic equipment can be of pronounced help in providing a much wider range of practical possibilities than were ever before open to the educator. Different possibilities here will also provide new and different possibilities for effective decomposition of the learning matter. For any given system within which teaching is to go on, the process of decomposing the learning material, choosing the sequencing and timing for its presentation, and establishing the reinforcement procedure, together comprise the task of "programming the material."
A possibility for physical-skill training
Briefly, this involves using "mechanical" means to expand the possibilities for presenting to the Subject the sequences of action codas which lead him through the coordinated sequence of basic actions that compose the compound, skill action to be learned. This is similar to what the teaching machine has donefor the mental training. Further, we note that in the teachingmachine art, we lookfor automatic ways both for detecting incorrect responses and for initiating appropriate correction or reinforcement reactions. I think that these same sorts of things can be provided for the physicalskill learner.
We could achieve a measure of these objectives, as many have done in physical-skill training, by automating the presentation of information in verbal or pictorial form (or some audible or visual form), and b, various automatic detectors of gross action. To my nonpsychologically trained mind and intuition, the coding of the desired component actions in this form, and the accompanying gross correction and reinforcement techniques, do not provide as effective a decomposition and coding of elemental actions as could be obtained with modern technology. There is too much
File (Project A) 9/23/60
highercenter activity that needs be associated with interpreting the codes (i.e., the visual or audible signals) and consciously initiating the corresponding physical actions.
Consider the job of teaching a Subject to operate a typewriter. Present techniques have him first memorizing the positions of the keys (higher-center activity), then slowly and rhythmically pounding out exercise series of characters until his basic responses begin to be established. Then he practices reading natural prose and converting it letter by letter into finger strokes. At this point he is still using highercenter facilities to convert the code into finger-action stimuli, and only gradually as he continues to practice does his finger response become automatic and he manages to increase his speed. Finally, he is not aware of the decoding and fingertriggering process, and whole words and word series tumble through his automatic processor.
The nature of my proposed innovation can be brought out with reference to this typingtraining example. Suppose that we spent some time initially with the Subject in teaching him what his finger responses were expected to be to small "cue signals" given him directly to different points on his hands and fingers. I shall discuss electricalimpulse cue signals, but they could be provided through other convenient stimulus receptors. It shouldn't take long for a Subject to learn to respond correctly to an impulse, applied say on the inside of a finger, near the base, that tells him to strike with this finger. Than a concurrent impulse out nearer the end of the finger, applied on the outside of the finger, would modify the cue to indicate striking on the upper level of keys. Similarly, a concurrent impulse out near the end of the finger, but applied on the inside, would modify the basic cue for that finger to indicate striking the bottom level of keys. Once proper positioning of the hands had been taught, the application of such cue signals would provide a direct stimulus code that should be processed in an automatic sense by the Subject after vary little practice.
The general teaching technique would be as follows: Instruct the Subject regarding the way he is expected to respond, and perhaps give him some directions as to good hand and finger form. Fit him with "gloves" which hold the stimulus contacts next to his skin. Seat him in front of a typewriter, and show him where his hands are supposed to rest. (Masking the key identities would perhaps be a good idearemember, we haven't told him anything about where he is to expect to find any given key). We put a given exercise display in front of him (perhaps a characteratatime or a wordatatime automatic indexing display, or perhaps just clamp in view an entire passage of exercise material). Turn on the teaching machine to which his gloves are connected, and leave him (them) alone. The teaching machine begins to give him the cue signals that correspond to successive characters of the exercise. The Subject sees what character he is expected to type next, and then he sees on his paper the result of his response to the associated cueing signal. We give him a first, slowsequencing period to practice getting firm responses to the cueing signals, but this period can be expected to be much shorter than a corresponding period where one is learning where the keys are and training his fingers to hit them when he thinks of them(as present learningtotype processes go). After this, the machine cues him at a rate which increases on some predetermined schedule.
A number of improvements can easily be visualized. Such thingsas a response-time monitor to give the machine some feedback on the student's progress. Perhaps some feedback from the typewriter to the teaching machine so that error detection can be provided. Then the progression to new exercise material, and the character-sequencing speeds, could be automatically controlled according to the individual student's progress.
File (Project A) 9/23/60
There is already in existence a teaching machine (developed by Gordon Pask, in England, and being developed in this country for commercial sale by Western Design of Santa Barbara) which trains keypunch operators. It's general scheme of operation is to present, sequentially, the visual stimulus of characters to be punched. There is a panel of lights that are arranged in positions corresponding to key positions, and the student gets a cue from the lighting of the light which corresponds to the key he is supposed to hit. Whenever the student makes an error, the machine slows down, and then gradually speeds up until the next error. Whenever an exercise fins has been repeated enough times to attain a certain minimum of errors, automatic progression to the next exercise line is provided. The cueing lights are gradually dimmed, too, until an error is made, whereupon they become brighter again and then start over with the dimming cycle.
The basic new thing, I think, is the principle of using physical-stimulus cues that are more effective for prompting desired physical responses than are audio or visual cues, which generally have to be given more higher-center processing in our brains before they result in the desired physical response than do the direct physicalstimulus cues. In the end, these cueinterpretation-response reactions are going to be supplanted by the prime stimulusinterpretation-response reaction, and the simpler and more direct we can make the intermediate temporary-skill cue interpretation the more efficient the whole learning process would seem to be.
A physicalskill which would seem very appropriate for some initial experimenting would be the use of a five-key binary keyboard for the encoding of alphabetic and numeric information. I suggest the standard teletype fivebit code. We can easily rig up standard teletype papertape readers and printers to make a simple teaching machine. The cues need only select a combination of fingers, and not direct the fingers to selected keys, which minimizes experimentation with cuesignal establishment. It is also intriguing to consider the possibilities for applying such a skill if it becomes easy and automatic to teach. A very direct, manto-machine communication link is handily available, and I for one would like to see some experimentation with such. The Signal Corps might also be interested in being able to teach easily such a skill, to soldiers who can then operate extremely simple field teletype transmitters.
Furthermore, it seems highly likely to me that, once this skill has been developed to a reasonable extent, these cueing signals that are no longer needed to help the man link a character stimulus to a key-combination finger response could become a means of receiving electrically carried information. Would not the mind easily learn to accept these old cueing signals as stimuli to be automaticallyinterpreted as character strings--i.e., a message? If it weren't a directly acquired skill, a natural byproduct of this teachingmachine approach, then it shouldn't take much more or much different teachingmachine application to establish the receiving skill. This possibility appeals very much to me as a start in basically new means for machineto-man communicationsand it would seem that the Signal Corps (or other military agency) would also be interested. Very simple portable transceivers could be directly human operated, and the storage, transmission, or decoding into print of their signals, etc., could be most easily and compatibly done with present signal-handling and digital techniques.
I rather imagine that there is a lot of training being done in our society nowadays, with more coming, in the operation of different kinds of equipment that
File (Project A) 9/23/60
would be amenable to the above-suggested type of automatic techniques. Typing, key punching, adding machine operating, Stenotyping, Morsecode sending and receiving-are some which use principally finger movements. Then there would be quite a list that uses fingers, arms, and legs, for which similar techniques should be quite valuable. It is very stimulating to consider that these basic directcueing principles might well be developed to the point where training for such things as piano playing, skiing, swimming, skating, judo, dancing, and so on-things that we consider to be quite sophisticated physical skillscould be helped immensely.
The use of a computer for monitoring many studentsimple machine setups, or for monitoring a studentmachine setup for a very sophisticated skill learning procedure, seems a very natural thing to consider. A system study of how best to distribute storage, monitoring, and control, between the central computer and the individual teaching stations would be needed before best balance could be determined. But the computer could sample the progress of each student from time to time, consult the record of his performance to date, and determine when it is time to change types of exercise, reinforce a lagging feature of skill development, etc., to provide very flexible individual attention that can allow each student to proceed at an optimum learning rate for him.
This same possibility exists, of course, for the verbalskill teaching machines, and could well be a good research point for us to consider.
In any event, I think that development of project work in the teachingmachine area would be a very good thing for SRI, and would be especially desirable with respect to a' personal interests. The physicalskill cueingtechnique approach might well be a very good way for us to get into this kind of activity, since it seems (from what I know) to be novel, initial experimentation with non-trivial cases would be relatively straightforward, there would seem to be a good market for special applications of this technique from which we could hope to derive sponsorship, and the continuing developments seem highly interesting from several points of view.
I propose to pursue this possibility, then. I am beginning some experimenting with electrical means of providing the cueing signals, and I am trying to line up teletype equipment for some trial runs of the binarykeyboard trainer. I shall also do a little state-oftheart research, as well as try to assess how SRI might best set up and staff a project in this area.