Teaching Machines

From the experimental study of learning come deviceswhich arrange optimal conditions for self-instruction.

There are more people in the worldthan ever before, and a far greater partof them want an education. The demandcannot be met simply by building moreschools and training more teachers. Edu-cation must become more efficient. Tothis end curricula must be revised andsimplified, and textbooks and classroomtechniques improved. In any other field ademand for increased production wouldhave led at once to the invention of la-bor-saving capital equipment. Educationhas reached this stage very late, possiblythrough a misconception of its task.Thanks to the advent of television, how-ever, the so-called audio-visual aids arebeing reexamined. Film projectors, tele-vision sets, phonographs, and tape rc-corders are finding their way into Amer-ican schools and colleges.

Audio-visual aids supplement and mayeven supplant lectures, demonstrations,and textbooks. In doing so they serveone function of the teacher: thry pm-sent material to the student and, whensuccessful, make it so clear and interest-ing that the student learns. There is an-other function to which they contributelittle or nothing. It is best seen in theproductive interchange between teachrrand student in the small classroom ortutorial situation. Much of that inter-change has already been sacrificed inAmerican education in order to teachlarge numbers of students. There is ateal danger that it will he wholly ob-Trured if USC o f rquipmcnt dkwrd sim-

B. F. Skinner

ply to PreJenr material becomesspread. The student is becoming

wide-

and more a mere passive receiver of in-struction.

Pressey’s Teaching Machines

There is another kind of capital equip-ment which will encourage the studentto take an active role in the instructionalprocess. The possibility was recognizedin the 192O’s, when Sidney L. Presseydesigned several machines for the auto-matic testing of intelligence and infor-mation. A recent model of one of theseis shown in Fig. 1. In using the devicethe student refers to a numbered item ina multiple-choice test. He presses thebutton corresponding to his first choiceof answer. If he is right, the devicemoves on to the next item; if he is wrong,the error is tallied, and he must continueto make choices until he is right (2).Such machines, Pressey pointed out (Z),could not only test and score, they couldfeat/r. When an examination is correctedand returned after a delay of manyhours or days, the student’s behavior isnot appreciably modified. The immedi-ate report supplied by a self-scoring de-vice, however, can have an importantinstructional effect. Pressey also pointedout that such machines would increaseefficiency in another way. Even in a smallclassroom the teacher usuahy knows thathr is moving too slowly for some studentsand too fast for othrrs. Those who cnuldgo faster are penalized, and those who

should go slower are poorly taught andunnecessarily punished by criticism andfailure. Machine instruction would per-mit each student to proceed at his ownrate.The “industrial revolution in educa-tion” which Pressey envisioned stub-bornly refused to come about. In 1932he expressed his disappointment (3)“The problems of invention are rela-tively simple,” he wrote. “With a littlemoney and engineering resource, a greatdeal could easily be done. The writerhas found from bitter experience thatone person alone can accomplish rela-tively little and he is regretfully dropping further work on these problems.But he hopes that enough may havebeen done to stimulate other workers,that this fascinating field may be devel-oped.”

Pressey’s machines succumbed in partto cultural inertia; the world of educa-tion was not ready for them. But theyalso had limitations which probably con-tributed to their failure. Pressey wasworking against a background of psycho-logical theory which had not come togrips with the learning process. Thestudy of human learning was dominatedby the “memory drum” and similar de-vices originally designed to study forget-ting. Rate of learning was observed, butlittle was done to change it. Why thesubject of such an experiment botheredto learn at all was of little interest. “Fre-quency” and “recency” theories of learn-ing, and principles of ‘Cmassed and spacedpractice,” concerned the conditions un-der which responses were remembered.

Pressey’s machines were designedagainst this theoretical background. .4sversions of the memory drum, they wereprimarily testing devices. They were tobe used after some amount of learninghad already taken place elsewhere. Byconfirming correct responses and byweakening responses which should nothave been acquired, a self-testing ma-chine does, indeed, teach; but it is notdesigned primarily for that purpose.Nevertheless, Pressey seems to have bernthe lirst to emphasize the importance nfimmediate feedback in eduration and topropose a system in which each student

9

could move at his own pace. He saw theneed for capital cquipmcnt in realizingthese objectives. Above all he conceivedof a machine which (in contrast with theaudio-visual aids which were beginningto be developed) permitted the studentto play an active role.

Another Kind of Machine

The learning process is now much bet-ter understood. Much of what we knowhas come from studying the behavior oflower organisms, but the results holdsurprisingly we11 for human subjects.The emphasis in this research has notbeen on proving or disproving theoriesbut on discovering and controlling thevariables of which learning is a function.This practical orientation has paid off,for a surprising degree of control hasbeen achieved. By arranging appropriate“contingencies of reinforcement,” specificforms of behavior can be set up andbrought under the control of specificclasses of stimuli. The resulting behaviorcan be maintained in stre.agth for longperiods of time. A technology based onthis work has already been put to use inneurology, pharmacology, nutrition, psy-chophysics, psychiatry, and elsewhere

(4).The analysis is also relevant to edu-

cation. A student b “taught” in the sensethat he is induced to engage in newforms of behavior and in specific formsupon specific occasions. It is not merelya matter of teaching him zo!zar to do:we are as much concerned with the prob-

ability that appropriate behavior will, in-deed, :ippcar at the proper t i m e - a nissue which \vould be classed tradition-ally under motivation. In education thebehavior to be shaped and maintainedis usually verbal, and it is to be brought

If our current knowledge of the ac-quisition and maintenance of verbal be-

under the control of both verbal and

havior is to be applied to education,

nonverbal stimuli. Fortunately, the spe-

some sort of teaching machine is needed.Contingencies of reinforcement whichchange the behavior of lower organisms

cial problems raised by verbal behavior

often cannot be arranged by hand; ratherelaborate apparatus is needed. The hu-

can be submitted to a similar analysis

man organism requires even more subtleinstrumentation. An appropriate teach-

(5).

ing machine will have several importantfeatures. The student must compose hisresponse rather than select it from a setof alternatives, as in a multiple-choiceself-rater. One reason for this is that wewant him to recall rather than recognize-to make a response as well as see thatit is right. Another reason is that effec-tive multiple-choice material must con-tain plausible wrong responses, whichare out of place in the delicate processof “shaping” behavior because theystrengthen unwanted forms. Although itis much easier to build a machine toscore multiple-choice answers than toevaluate a composed response, the tech-nical advantage is outweighed by theseand other considerations.

A second requirement of a minimal

teaching machine also distinguishes itfrom cariier versions. In acquiring com-plex behavior the student must passthrough a carefully designed sequenceof steps, often of considerable length.Each step must be so small that it can

Several machines with the requiredcharacteristics have been built and

always be taken, yet in taking it the stu-

tested. Sets of separate presentations or“frames” of visual material are storedon disks, cards, or tapes. One frame is

dent moves somewhat closer to fully

presented at a time, adjacent frames be-ing out of sight. In one type of machine

competent behavior. The machine must

the student composes a response by mov-ing printed figures or letters (6). His set-

make sure that these steps are taken in

ting is compared by the machine with a

a carefully prescribed order.

coded response. If the t\vo correspond,the machine automatically presents thenext frame. If they do not, the response iscleared, and another must be composed.The student cannot proceed to a secondstep until the first has been taken. Amachine of this kind is being tested inteaching spelling, arithmetic, and othersubjects in the lower grades.

For more advanced students-fromjunior high school, say, through college-a machine which senses an arrange-ment of letters or figures is unnecessarilyrigid in specifying form of response.Fortunately, such students may be askedto compare their responses with printedmaterial revealed by the machine. In themachine shown in Fig. 2, material isprinted in 30 radial frames on a 12-inchdisk. The student inserts the disk andcloses the machine. He cannot proceeduntil the machine has been locked, and,once he has begun, the machine cannotbe unlocked. All but a corner of oneframe is visible through a window. Thestudent writes his response on a paperstrip exposed through a second open-ing. By lifting a lever on the front of themachine, he moves what he has writtenunder a transparent cover and uncoversthe correct response in the remainingcorner of the frame. If the two responsescorrespond, he moves the lever horizon-tally. This movement punches a hole inthe paper opposite his response, record-ing the fact that he called it correct,and alters the machine so that the framrwill not appear again when the studentworks around the disk a second time.Whether the response was correct or not,a second frame appears when the leveris returned to its starting position. Thestudent proceeds in this way until hehas responded to a11 frames. He then

works around the disk a second time,but only those frames appear to whichhe has not correctly responded. Whenthe disk revolves without stopping, theassignment is finished. (The student isasked to repeat each frame until a cor-rect response is made to allow for thefact that, in telling him that a responseis wrong, such a machine tells him whatis right.)

The machine itself, of course, does notteach. It simply brings the student intocontact with the person who composedthe material it presents. It is a labor-saving device because it can bring oneprogrammer into contact with an in-definite number of students. This maysuggest mass production, but the effectupon each student is surprisingly likethat of a private tutor. The comparisonholds in several respects. (i) There is aconstant interchange between programand student. Unlike lectures, textbooks,and the usua1 audio-visual aids, the ma-chine induces sustained activity. Thestudent is always alert and busy. (ii)Like a good tutor, the machine insiststhat a given point be thoroughly under-stood, either frame by frame or set byset, before the student moves on. Lec-tures, textbooks, and their mcchanizcdequivalents, on the other hand, proceedwithout making sure that the studentunderstands and easily leave him behind.(iii) Like a good tutor the machine pre-sents just that material for which thestudent is ready. It asks him to take onlythat step which he is at the moment bestequipped and most likely to take. (iv)Like a skiilful tutor the machine helpsthe student to come up with the rightanswer. It does this in part through theorderly construction of the program andin part with techniques of hinting,prompting, suggesting, and so on, de-rived from an analysis of verbal be-havior (5). (v) Lastly, of course, themachine, like the private tutor, rein-forces the student for every correct re-sponse, using this immediate feedbacknot only to shape his behavior most effi-ciently but to maintain it in strength ina manner which the layman would de-scribe as “holding the student’s interest.”

Programming Material

The success of such a machine de-pends on the material used in it. Thetask of programming a given. subject isat first sight rather formidable. Manyhelpful techniques can be derived froma general analysis of the relevant be-havioral processes, verbal and nonverbal.

Fig. 2. Student at war!; on a teaching machine. One frame of material is partly visible inthe left-hand window. The student writes his response on a strip of paper exposed at theright. He then lifts a lever with his left hand, advancing his written response under atransparent cover and uncovering the correct response in the upper corner of the frame.If he is right, he moves the lever to the right, punching a hole alongside the response hehas called right and alterifig the machine so ihat that frame will not appear again whenhe goes through the series a second time: A new frame appears when the lever is returnedto its starting position.

Specific forms of behavior are to beevoked and, through differential rein-forcement, brought under the control ofspecific stimuli.

This is not the place for a systematicreview of available techniques, or of thekind of research which may be expectedto discover others. However, the ma-chines themselves cannot be adequatelydescribed without giving a few examplesof programs. We may begin with a setof frames (see Table 1) designed toteach a third- or fourth-grade pupil tospell the word munu@cture. The sixframes are presented in the order shown,and the pupil moves sliders to exposeletters in the open squares.

The word to be learned appears i nbold face in frame 1, with an exampleand a simple definition. The pupil’s firsttask is simply to copy it. When he doesso correctly, frame 2 appears. He mustnow copy selectively: he must identify“fact” as the common part of ‘Cmanu-facture” and “factory.” This helps himto spell the word and also to acquire a

separable “atomic” verbal operant (5).In frame 3 another root must be copiedselectively from “manual.” In frame 4the pupil must for the first time insertletters without copymg. Since he is askedto insert the same letter in two places, awrong response will be doubly conspicuous, and the chance of failure is therebyminimized. The same principle governsframe 5. In frame 6 the pupil spells theword to complete the sentence used asan example in frame 1. Even a poor stu-dent is likely to do this correctly becausehe has just composed or completed theword five times, has made two impor-tant root-responses, and has learned thattwo letters occur in the word twice. Hehas probably learned to spell theswordwithout having made a mistake.

Teaching spelling is mainly a processof shaping complex forms of behavior.In other subjects-for example, arith-metic-responses must be brought underthe control of appropriate stimuli. Un-fortunately the material which has beenprrpared for teaching arithmetic (7)

4

Table 1. A set of frames designed to teach third- or fourth-grade pupil to spell the word

1. Manufacture means to make or buii ?mir factories manufacture chairs. Copy theword here:

ElnElclL-zlElnclcl~2. Part of the word lrt of the word factory. F ,Jrts come from an old word

meaning make or

rnanul-J~~~ure

3. Part of the word is . f the word manual. Both parts come from an old wordfor hand. Many things x made by hand.

~~~~facture

4. The same letter goes in both spaces:manufmcture

.5. The same letter goes in both spaces:manafactcre

does not lend itself to excerpting. Thenumbers 0 through 9 are generated inrelation to objects, quantities, and scales.The operatioil? of addition, subtraction,multiplication, and division are thor-

’ developed before the number 10Ached. In the course of this the

+I composes equations and expressionsn a great variety of alternative forms.He completes not only 5 +4 = -1, butl +4=9, 5@4=9, and so on, ;sded inmost cases by illustrative materials. Noappeal is made to rote memorizing, evenin the later acquisition of the tables. Thestudent is expected to arrive at 9x7=63,not by memorizing it as he Lvouldmemorize a line of poetry, but by put-ting into practice such principles as thatnine times a numbe r ‘3 the same as tenrimes the number minus the numberboth of these being “obvious” or al-

ready well learned), that the digits ina multiple of nine add to nine, that inc.omposing uL:cessive multiples of nineone rounts backwards (nine, eighteen.t\venty-seven. thirty-&, and so on), thatnine times a single digit is a number be-ginning lvith one less than the digit%ne times 3i.y is fifty something), andpossibly even that the product of t!vonumbers separated by only one numberis equal to the square of the separatingnumber minus one (the square of eightalready being familiar from a specialseries of frames concerned with squares).

Programs of this sort run to greatlength. At five or six frames per word,four grades of spelling may requireZ&O00 or 25,000 frames, and three orfour grades of arithmetic, as many again.If these figures seem large, it is only be-cause we are thinking of the normalcontact between teacher and pupil. Ad-mittedly, a teacher cannot supervise

10,000 or 15,000 responses made by eachpupil per year. But the pupil’s time isnot so limited. In any case, surprisinglylittle time is needed. Fifteen minutes perday on a machine should suffice foreach of these programs, the machinesbeing free for other studemq for the restof each day. (It is probably becausetraditional methods are so inefficientthat we have been led to suppose thateducation requires such a prodigiouspart of a young person’s day.)

“Vanishing” can be used in teachingother types of verbal behavior. When a

A simple technique used in program-ming material at the high-school or col-lege level, by means of the machineshotvn in Fig. 2, is exemplified in teach-ing a student to recite a poem. The firstline is presented Lvith several unimpor-tant letters omitted. The student mustread the line “meaningfully” and sup-ply the missing letters. The second,third. and fourth frames present suc-ceeding lines in the same way. In thefifth frame the first line reappears withother letters also missing. Since the stu-dent has recently read the line, he cancomplete it correctly. He does the samefor the second, third, and fourth lines.Subsequent frames are increasingly in-complete, and eventually-say, after 20or ?4 frames-the student reproducesall four lines without external help, andquite possibly without having made atvrong response. The technique is simi-lar to that used in teaching spelling: re-Tponses are first controlled by a text, butthis is slowly reduced (colloquially,“vanished”) until the responses can beemitted without a text, each member ina series of responses being now underthe “intraverbal” control of other mem-bers.

student describes the geography of partof the world or the anatomy of part ofthe body, or names plants and animalsfrom specimens or pictures, verbal re-sponses are controlled by nonverbalstimuli. In setting up such behavior thestudent is first asked to report featuresof a fully labeled map, picture, or ob-ject, and the labels are then vanished.In teaching a map, for example, thetnachinc asks the student to describespatial relations among cities, countries,rivers, and so on, as shown on a fullylabeled map. He is then asked to do thesame rvith a map in which the names areincomplete or, possibly, lacking. Even-tually he is asked to report the samerelations with no map at all. If the ma-terial has been well programmed, hecan do so correctly. Instruction is some-times concerned not so much with im-parting a new repertoire of verbal re-sponses as with getting the student todescribe something accurately in anyavailable terms. The machine can “makesure the student understands” a graph,diagram, chart, or picture by askinghim to identify and explain its features--correcting him, of course, whenever heis wrong.

In addition to charts, maps, graphs,models, and so on, the student may haveaccess to auditory material. In learningto take dictation in a foreign language,for example, he selects a short passageon an indexing phonograph accordingto instructions given by the machine. Helistens to the passage as often as neces-sary and then transcribes it. The ma-chine then reveals the correct text. Thestudent may listen to the passage againto discover the sources of any error. Theindexing phonograph may also be usedkvith the machine to teach other Ian-guagc skills, as well as telegraphic code,music, speech, parts of literary and dra-matic appreciation, and other subjects.

Several programming techniques areexemplified by the set of frames in

.A typical program combines many ofthese functions. The set of frames shownin Table 2 is designed to induce the stu-dent of high-school physics to talk in-telligently, and to some extent techni-cally, about the emission of light froman incandescent source. In using themachine the student will write a wordor phrase to complete a given item andthen uncover the corresponding word orphrase shown here in the column at theright. The reader who wishes to get the“feel.’ of the material should cover theright-hand column with a card, uncov-ering each line only after he has com-pleted the corresponding item.

Table 2. Part of a program in high-school physics. The machine presents one item at a time The student completes the item and thenuncovers the corresoondinu word or nhrase shown at the right.

Sentence to be completedWord to be

supplied

1.

2.

The important parts of a flashlight are the battery and the bulb. When we “turn on” a flashlight, we close aswitch which connects the battery with the -.

When we turn on a flashlight, an electric current flows through the fine wire in the - and causes it to growhot.

3.

4.

5.

6.

7.

When the hot wire glows brightly, we say that it gives off or sends out heat and -.

The fine wire in the bulb is called a filament. The bulb “lights up” when the filament is heated by the passageof a(n) - current.

When a weak battery produces little current, the fine wire, or -, does not get very hot.

A filament which is less hot sends out or givca off -- light.

“Emit” means “send out.” The amount of light sent out or “emitted,” by a filament depends on how - thefilament is.

8. The higher the temperature of the filament the - the light emitted by it.

9.

IO.

If a flashlight battery is weak, the ~ in the bulb may still glow, but with only a dull red color.

The light from a very hot filament is colored yellow or white. The light from a filament which is not very hoti s c o l o r e d - .

Il.

12.

13.

A blacksmith or other metal worker sometimes makes sure that a bar of iron is heated to a “cherry red” beforehammering it into shape. He uses the - of the light emitted by the bar to tell how hot it is.

Both the color and the amount of light depend on the - of the emitting filament or bar.

An object which emits light because it is hot is called “incandescent.” .4 flashlight bulb is an incandescent sourceof -.

14.

15.

16.

A neon tube emits light but remains cool. It is, therefore, not an incandescent - of light.

A candle flame is hot. It is a(n) - source of light.

The hot wick of a candle gives off small pieces or particles of carbon which burn in the flame. Before or whileburning, the hot particles send out, or -, light.

17.

18.

A long candlewick produces a flame in which oxygen does not reach all the carbon particles. Without oxygenthe particles cannot burn. Particles which do not burn rise above the flame as -.

We can show that there are particles of carbon in a candle flame, even when it is not smoking, by holding apiece of metal in the flame. The metal cools some of the particles before they burn, and the unburned carbon- collect on the metal as soot.

19. The particles of carbon in soot or smoke no longer emit light because they are - than when they were inthe flame.

20.

21.

The reddish part of a candle flame has the same colt as the filament in a flashlight with a weak battery. Wemight guess that the yellow or white parts of a candle flame are ~ than the reddish part.

“Putting out” an incandescent electric light means turning off the current so that the filament grows too -to emit light.

22. Setting me to the wick of an oil lamp is called - the lamp.

23. The sun is our principal - of light, as well as of heat.

24. The sun is not only very bright but very hot. It is a powerful - source of light.

25. Light is a form of energy. In “emitting light” an object changes, or “converts,” one form of - into another.

26. The electrical energy supplied by the battery in a flashlight is converted to - and -.

27.

28.

29.

30.

31.

3?.

33.

34.

35.

If WC leave a flashlight on, a11 the energy stored in the battery will finally be changed or - into heat and light.

The light from a candle flame comes from the - released by chemical changes as the candle burns.

A nearly “dead*’ battery may ma!ce a flashlight bulb jvarm to the touch, but the filament may still not be hotenough to emit light-in other words, the filament ~111 not be - at that temperature.

objects. such as a filament, carbon particles, or iron bars, bccomc incandescent when heated to about 800 de-grees Celsius. At that temperature they begin to - -.

ivhcn raised to any temperature above 800 degrees Celsius, an object such as an iron bar will emit light. .4l-thougb the bar may melt or vaporize, its particles wiY be - no matter how hot they get.

About 800 degrees Celsius is the lower limit of the tcm:crature at which particles emit light. There is no upperlimit of the - at which emission of light occurs.

Sumight is - by very hot gases near the surface of :he sun.

Complex changes similar to an atomic explosion geaera:e the great heat which explains the -- of light bythe sun.

Below about - degrees Celsius an object is not an incandescent source of light.

bulb

bulb

light

electric

filament

less

hot

brighter,stronger

filament

red

color

temperature

light

aourcc

incandescent

emit

smoke

particles

cooler, colder

hotter

coid, cool

lighting

source .

incandescent

energy

heat, light;light, heat

converted

energy

incandescent*

emit light

incandescent

temperature

emitted

emission

800

Table 2. Technical terms are introducedslowly. For example, the familiar term“fine wire” in frame 2 is followed by adefinition of the technical term “fila-ment” in frame 4; “filament” is thenasked for in the presence of the non-scientific synonym in frame 5 and with-out the synonym in frame 9. In the sameway “glow,” “give off light,” and “sendout light” in early frames are followedby a definition of “emit” with a synonymin frame 7. Various inflected forms of“emit” then follow, and “emit” itself isasked for with a synonym in frame 16.ft is asked for without a synonym but ina helpful phrase in frame 30, and“emitted” and “emission” are asked forwithout help in frames 33 and 34.The relation between temperature andamount and color of light is developedin several frames before a formal state-ment using the word “temperature” isasked for in frame 12. “Incandescent” isdefined and used in frame 13, is usedagain in frame, 14, and is asked for inframe 15, the student receiving a the-matic prompt from the recurring phrase“incandescent source of light.” A for-mal prompt is supplied by “candle.”In frame 25 the new response “energy”is easily evoked by the words “formof . . . .” because the expression “formof energy” is used earlier in the frame.“Energy” appears again in the next twoframes and is finally asked for, withoutaid, in frame 28. Frames 30 through 35discuss the limiting temperatures of in-candescent objects, while reviewing sev-eral kinds of sources. The figure 800 isused in three frames. Two interveningframes then permit some time to passbefore the response “800” is asked for.

Unwanted responses are eliminatedwith special techniques. If, for example,the second sentence in frame 24 weresimply “It is a(n) - source of light,”the two “very’s” would frequently leadthe student to fill the blank with“strong” or a synonym thereof. This isprevented by inserting the word “pow-erful” to make a synonym redundant.Similarly, in frame 3 the lvords “heatand’ preempt the response “heat,”which would otherwise correctly fill theblank.

The net effect of such material is morethan the acquisition of facts and terms.Beginning with a largely unverbalizedacquaintance with flashlights, candles,and so on, the student is induced to talkabout familiar events, together with afew new facts, with a fairly technicalvocabulary. He applies the same termsto facts which he may never before haveseen to be similar. The emission of light

from an incandescent source takes shapeas a topic or field of inquiry. An undcr-standing of the subject emerges whichis often quite surprising in view of thefragmentation required in item building.

It is not easy to construct such a pro-gram. L+‘here a confusing or ellipticalpassage in a textbook is forgivable be-cause it can be clarified by the teacher,machine material must be self-containrdand wholly adequate. There are otherreasons why textbooks, lecture outlines,and film scripts are of little help in pre-paring a program. They are usually notlogical or developmental arrangementsof material but strategems which theauthors have found successful under ex-isting classroom conditions. The exam-ples they give are more often chosen tohold the student’s interest than to clarifyterms and principles. In composing ma-terial for the machine, the programmermay go directly to the point.

A first step is to define the field. Asecond is to collect technical terms,facts, laws, principles, and cases. Thesemust then be arranged in a plausible de-velopmental order-linear if possible,branching if necessary. A mechanicalarrangement, such as a card filing sys-tem, helps. The material is distributedamong the frames of a program toachieve an arbitrary density. In the finalcomposition of an item, techniques forstrengthening asked-for responses andfor transferring control from one vari-able to another are chosen from a listaccording to a given schedule in order

to prevent the establishment of irrelevantverbal tendencies appropriate to a singletechnique. When one set of frames hasbeen composed, its terms and facts areseeded mechanically among succeedingsets, where they will again be referred toin composing later items to make surethat the earlier repertoire remains ac-tive. Thus, the technical terms, facts, andexamples in Table 2 have been distrib-uted for reuse in succeeding sets on re-flection, absorption, and transmission,where they are incorporated into itemsdealing mainly with other matters. Setsof frames for explicit review can, ofcourse, be constructed. Further researchwill presumably discover other, possiblymore effective, techniques. MeanwhiIe,it must be admitted that a considerablemeasure of art is needed in composinga successful program.

Whether good programming is to re-main an art or to become a scientifictechnology, it is reassuring to know thatthere is a final authority-the student.An unexpected advantage of machineinstruction has proved to be the feedbackto the progrummer. In the elementaryschool machine, provision is made fordiscovering which frames commonlyyield wrong responses, and in the high-school and college machine the paperstrips bearing written answers are avail-able for analysis. A trial run of the firstversion of a program quickly revealsframes which need to be altered, or se-quences which need to be lengthened.0ne or two revisions in the light of a

7

few dozen responses work a great im-provement. No comparable feedback isavailable to the lecturer, textbook writer,or maker of films. Although one text orfilm may seem to be better than another,it is usually impossible to say, for ex-ample, that a given sentence on a givenpage or a particular sequence in a filmis causing trouble.

Difficult as programming is, it has itscompensations. It is a salutary thing totry to guarantee a right response at everystep in the presentation of a subject mat-ter. The programmer will usually findthat he has been accustomed to leavemuch to the student-that he has fre-quently omitted essential steps and neg-lected to invoke relevant points. The re-sponses made to his material may revealsurprising ambiguities. Unless he islucky, he may find that he still hassomething to learn about his subject. Hewill almost certainly find that he needsto learn a great deal more about the be-havioral changes he is trying to inducein the student. This effect of the ma-chine in confronting the programmerwith the full scope of his task may initself produce a considerable improve-ment in education.

Composing a set of frames can be anexciting exercise in the analysis of knowl-edge. The enterprise has obvious bear-ings on scientific methodology. Thereare hopeful signs that the epistemologi-cal implications will induce experts tohelp in composing programs. The expertmay be interested for another reason.We can scarcely ask a topflight mathe-matician to write a primer in second-grade arithmetic if it is to be used bythe average teacher in the average class-room. But a carefully controlled ma-chine presentation and the resulting im-mediacy of contact between programmerand student offer a very different pros-pect, which may be enough to inducethose who know most about the subjectto give some thought to the nature ofarithmetical behavior and to the variousforms in which such behavior should beset up and tested.

Can Material Be Too Easy?

The traditional teacher may viewthese programs with concern. He may beparticularly alarmed by the effort tomaximize success and minimize failure.He has found that students do not payattention unless they are worried aboutthe consequences of their work. The cus-tomary procedure has been to maintainthe necessary anxiety by inducing errors.

In recitation, the student who obviouslyknows the answer is not too often asked;a test item which is correctly answeredby everyone is discarded as nondiscrimi-nating; problems at the end of a sectionin a textbook in mathematics generallyinclude one or two very difficult items;and so on. ( T h e teacher-tumed-pro-grammer may be surprised to find thisattitude affecting the construction ofitems. For example, he may find it diffi-cult to allow an item to stand which“gives the point away.” Yet if we cansolve the motivational problem withother means, what is more effective thangiving a point away?) Making sure thatthe student knows he doesn’t know is atechnique concerned with motivation,not with the learning process. Machinessolve the problem of motivation in otherways. There is no evidence that what iseasily learned is more readily forgotten.If this should prove to be the case, re-tention may be guamnteed by subsequentmaterial constructed for an equallypainless review.

The standard defense of “hard” ma-terial is that we want to teach more thansubject matter. The student is to bechallenged and taught to “think.” Theargument is sometimes little more thana rationalization for a confusing presen-tation, but it is doubtless true that lec-tures and texts are often inadequate andmisleading by design. But to what end?What sort of “thinking” does the studentlearn in struggling through difficult ma-terial? It is true that those who learnunder difficult conditions are better stu-dents, but are they better because theyhave surmounted difhculties or do theysurmount them because they are better?In the guise of teaching thinking we setdifficult and confusing situations andclaim credit for the students who dealwith them successfully.

The trouble with deliberately makingeducation difficult in order to teachthinking is (i) that we must remain con-tent with the students thus selected, cvcnthough we know that they are only asmall part of the potential supply ofthinkers, and (ii) that we must continueto sacrifice the teaching of subject mat-ter by renouncing effective but “easier”methods. A more sensible program is toanalyze the behavior called “thinking”and produce it according to specifica-tions. A program specifically concernedwith such behavior could be composedof material already available in logic,mathematics, scientific method, and psy-chology. Much would doubtless be addedin completing an effective program. Themachine has already yielded important

relevant by-products. Immediate feed-back encourages a more careful readingof programmed material than is the casein studying a text, where the conse-quences of attention or inattention arcso long deferred that they have littleeffect on reading skills. The behavior in-volved in observing or attending to de-tail-as in inspecting charts and modelsor listening closely to recorded speech-is efficiently shaped by the contingenciesarranged by the machine. And when animmediate result is in the balance, a stu-dent will be more likely to learn how tomarshal relevant material, to concen-trate on specific features of a presenta-tion, to reject irrelevant materials, torefuse the easy but wrong solution, andto tolerate indecision, all of which areinvolved in effective thinking.

Part of the objection to easy mate-rial is that the student will come to de-pend on the machine and will be lessable than ever to cope with the ineffi-cient presentations of lectures, textbooks,films, and “real life.” This is indeed aproblem. All good teachers must “wean”their students, and the machine is no ex-ception. The better the teacher, the moreexplicit must the weaning process be.The final stages of a program must beso designed that the student no longerrequires the helpful conditions arrangedby the machine. This can be done inmany ways-among others by using themachine to discuss material which hasbeen studied in other forms. These arequestions which can be adequately an-swered only by further research.

No large-scale “evaluation” of ma-chine teaching has yet been attempted.We have so far been concerned mainlywith practical problems in the designand use of machines, and with testingand revising sample programs. The ma-chine shown in Fig. 2 was built andtested with a grant from the Fund forthe Advancement of Education. Mate-rial has been prepared and tested withthe collaboration of Lloyd E. Homme,Susan R. Meyer, and James G. Holland(8). The self-instruction room shown inFig. 3 was set up under this grant. Itcontains ten machines and was recentlyused to teach part of a course in humanbehavior to Harvard and Radcliffe un-dergraduates. Nearly 200 students com-pleted 48 disks (about 1400 frames)prepared with the collaboration of Hol-land. The factual core of the course wascovered, corresponding to about 200pages of the text (9). The median timerequired to finish 48 disks was 145hours. The students were not examinedon the material but were responsible

for the text which overlapped it. Theirreactions to the material and to self-instruction in general have been studiedthrough interviews and questionnaires.Both the machines and the material arenow being modified in the light of thisexperience, and a more explicit evalua-tion will then be made.

Meanwhile, it can be said that theexpected advantages of machine instruc-tion were generously confirmed. Unsus-pected possibilities were revealed whichare now undergoing further exploration..4lthough it is less convenient to reportto a self-instruction room than to pickup a textbook in one’s room or else-where, most students felt that they hadmuch to gain in studying by machine.Most of them worked for an hour ormore with little effort, although theyoften felt tired afterwards, and they re-ported that they learned much more inless time and with less effort than in con-ventional ways. No attempt was made topoint out the relevance of the materialto crucial issues, personal or otherwise,but the students remained interested.(Indeed, one change in the reinforcingcontingencies suggested by the experi-ment is intended to reduce the motiva-tional level.) An important advantageproved to be that the student alwaysknew where he stood, without waitingfor an hour test or final examination.

Some Questions

Several questions are commonly askedwhen teaching machines are discussed.Cannot the results of laboratory researchon learning be used in education withoutmachines? Of course they can. Theyshould lead to improvements in text-books, films, and other teaching mate-rials. Moreover, the teacher who reallyunderstands the conditions under whichlearning takes place will be more effec-tive, not only in teaching subject matterbut in managing the class. Nevertheless,some sort of device is necessary to ar-range the subtIe contingencies of rein-forcement required for optimal learningif each student is to have individual at-tention. In nonverbal skills this is usuallynbvious; texts and instructor can guidethe learner but they cannot arrange thefinal contingencies which set up skilledbehavior. It is tme that the verbal skillsat issue here are especially dependentupon social reinforcement, but it mustnot be forgotten that the machine sim-ply mediates an essentially verbal rela-tion. In shaping and maintaining verbal

knowledge we are not committed to thecontingencies arranged through immedi-ate personal contact.

Machines may still seem unnecessarilycomplex compared with other media-tors such as workbooks or self-scoringtest forms. Unfortunately, these alterna-tives are not acceptable. When materialis adequately programmed, adjacentsteps are often so similar that one framereveals the response to another. Onlysome sort of mechanical presentationwiil make successive frames indepen-dent of each other. Moreover, in self-instruction an automatic record of thestudent’s behavior is especially desirable,and for many purposes it should be fool-proof. Simplified versions of the presentmachines have been found useful-forexample, in the work of Ferster andSapon, of Porter, and of Gilbert (8)-but the mechanical and economic prob-lems are so easily solved that a machinewith greater capabilities is fully war-ranted.

WilI machines replace teachers? Onthe contrary, they are capital equipmentto be used by teachers to save time andlabor. In assigning certain mechanizablefunctions to machines, the teacheremerges in his proper role as an indis-pensable human being. He may teachmore students than heretofore-this isprobably inevitable if the world-widedemand for education is to be satisfied-but he will do so in fewer hours andwith fewer burdensome chores. In re-turn for his greater productivity he canask society to improve his economic con-dition.

The role of the teacher may well bechanged, for machine instruction willaffect several traditional practices. Stu-dents may continue to be grouped in“grades” or “classes,” but it will be pos-sihle for each to proceed at his ownlevel, advancing as rapidly as he can.The other kind of “grade” will alsochange its meaning. In traditional prac-tice a C means that a student has asmattering of a whole course. But if ma-chine instruction assures mastery at everystage, a grade will he useful only inshowing how /a~ a student has gone. Cmight mean that he is halfway througha course. Given enough time he will beable to get an A; and since A is no longera motivating device, this is fair enough.The quick student will meanwhile havepicked up A’s in other subjects.

Differences in ability raise other ques-tions. A program designed for the slow-est student in the school system willprohably not seriously delay the fast stu-

dent, who will be free to progress athis own speed. (He may profit from thefull coverage by filling in unsuspectedgaps in his repertoire.) If this does notprove to be the case, programs can beconstructed at two or more levels, andstudents can be shifted from one to theother as performances dictate. If thereare also differences in “types of think-ing,” the extra time available for ma-chine instruction may be used to presenta subject in ways appropriate to manytypes. Each student will presumably re-tain and use those ways which he findsmost useful. The kind of individual dif-ference which arises simply because astudent has missed part of an essentialsequence (compare the child who has no“mathematical ability” because he wasout with the measles when fractions werefirst taken up) will simply be eliminated.

Other Uses

Self-instruction by machine has manyspecial advantages apart from educa-tional institutions. Home study is an ob-vious case. In industrial and militarytraining it is often inconvenient to sched-ule students in groups, and individualinstruction by machine should he a feasi-ble alternative. Programs can also beconstructed in subjects for which teach-ers are not availahle-for example, whennew kinds of equipment must be ex-plained to operators and repairmen, orwhere a sweeping change in methodfinds teachers unprepared (ZO). Educa-tion sometimes fails because studentshave handicaps which make a normalrelationship with a teacher difficult orimpossihle. (Many blind children aretreated today as feeble-minded becauseno one has had the time or patience tomake contact with them. Deaf-mutes,spastics, and others suffer similar han-dicaps.) A teaching machine can headapted to special kinds of communica-t ion-as, for example, Brai l le-and,ahove all, it has infinite patience.

Conclusion

An analysis of education within theframework of a science of hehavior hasbroad implications. Our schools, in par-ticular our “progressive’ schools, areoften held responsible for many currentproblems-including juvenile delin-quency and the threat of a more pow-erful foreign technology. One remedyfrequently suggested is a return to older

techniques, especially to a greater “dis-cipline” in schools. Presumably this isto be obtained with some form of pun-ishment, to be administered either withcertain classical instruments of physicalinjury-the dried bullock’s tail of theGreek teacher or the cane of the Englishschoolmaster-or as disapproval or fail-ure, the frequency of which is to be in-creased by “raising standards.” This isprobably not a feasible solution. Notonly education but Western culture asa whole is moving away from aversivepractices. We cannot prepare youngpeople for one kind of life in institutionsorganized on quite different principles.The discipline of the birch rod mayfacilitate learning, but we must remem-ber that it also breeds followers of dic-tators and revolutionists.

In the light of our present knowledgea school system must be called a failureif it cannot induce students to learn ex-cept by threatening them for not learn-ing. That this has always been thestandard pattern simply emphasizes theimportance of modern techniques. John

9

Dewey was speaking for his culture andhis time when he attacked aversive edu-cational practices and appealed to teach-ers to turn to positive and humane meth-ods. What he threw out should have beenthrown out. Unfortunately he had toolittle to put in its place. Progressive edu-cation has been a temporizing measurewhich can now be effectively supple-mented. Aversive practices can not onlybe replaced, they can be replaced withfar more powerful techniques. The pos-sibilities should be thoroughly exploredif we are to build an educational systemwhich will meet the present demandwithout sacrificing democratic principles.

I.

2.

3.4.

5.

6.

The Navy’s “Se l f -Rater” i s a l arger ver s ionof Pressey’s m a c h i n e . T h e i t e m s are p r i n t e don code-punched plastic cards fed hy the tna-chine . The t ime required to answer i s takeninto account in scoring.S . L . Pressey, S c h o o l and S o c i e t y 2 3 , S36(19261.-, i&d. 36, 934 (1932).B. F . Skinner , “The experimental analys i s c~fbehav ior ,” Am. scienttit 45, 4 (1957).-, Verbal Be-havim ( A p p l e t o n - C e n t u r y -Crofts, New York, 1957).-, “The sc ience of learning and the art

o f t e a c h i n g , ” Harvard Eductvional Reu. 24, 2(1934).This material was prepared with the assistanceof Susan R. Meyer.Dr. Homme prepared sets of frames for teach-ing part o f coIlege physia ( k i n e m a t i c s ) , andMrs. Meyer has prepared and informally testedmaterial in remedial reading and vocabule~building at the junior high school level. Otherwho have contributed to the development oft e a c h i n g m a c h i n e s s h o u l d b e m e n t i o n e d .N a t h a n H . Azrin c o o p e r a t e d w i t h m e i nt e s t i n g a vemion o f a m a c h i n e t o t e a c ha r i t h m e t i c . C . B. Ferster a n d S t a n l e y M.Sapon used a simple “machine” to teach Ger-man [see “An appl i ca t ion o f recent deve lop-ments in psychology to the teaching of Ger-man ,” Harmrd Educat ional Reu. 28, 1 (1958)].D o u g l a s P o r t e r , o f t h e G r a d u a t e School ofEducation at Harvard, has made an in&pen&ent schoolroom test of machine instruction ins p e l l i n g [ s e e “ T e a c h i n g m a c h i n e s , ” HarvardG r a d u a t e S c h o o l of Educ. Assoc. Bull. 3, 1(1956)]. Dewa Cooper has experimented withthe teaching of Eiglish compos i t ion for fresh-xnen a t t h e U n i v e r s i t y o f Kenh~cky. ThomasF. Gilbert , of the University of Georgia, hascompared standard and machine instruction inan introductory course in psychology, and withthe collaboration of J. E. Jewett hau preparedmaterial in algebra. The U.S. Naval TrainingDevices Center has recent ly contracted withthe University of Pennsylvania for a study ofprograms relat ing to the machine instruct ionof servicemen, under the direction of EugeneH. Galanter.B. F. Sk inner , Sc ience and Human B e h a v i o r(Macmil lan, New York, 1953) .K . Menger, “New approach to teaching inter-m e d i a t e m a t h e m a t i c s , ” S c i e n c e 1 2 7 , 1 3 2 0(1956).