my life

Education

The Uses of Computers in Education
The h l g e informution-processing cupcities of computen muke
it possible to use them to adapt mechanical teaching routines
t o the needs and the pust performance of the indi(~idLm1st udent
A s other articles in this issue make
abundantly clear, both the
- processing and the uses of information
are undergoing an unprecedented
technological revolution. Not
only are machines now able to deal
with many kinds of information at high
speed and in large quantities but also
it is possible to manipulate these quantities
of information so as to benefit
from them in entirely novel mays. This
is perhaps nowhere truer than in the
field of education. One can predict that
in a few more years millions of schoolchildren
will have access to what Philip
of hlacedon’s son Alexander enjoyed as
a royal prerogative: the personal services
of a tutor as well-informed and responsive
as Aristotle.
The basis for this seemingly extravagant
prediction is not apparent in many
examinations of the computer’s role in
education today. In themselves, howel’er,
such examinations provide impressive
evidence of the importance of computers
on the educational scene, As an
example, a recent report of the National
Academy of Sciences states that by mid-
1965 more than 800 computers were in
service on the campuses of various
American universities and that these
institutions spent $175 nillion for computers
that year. The report goes on to
forecast that by 1968 the universities’
annual budget for computer operations
will reach $300 million and that their
by Patrick Suppes
total investment in computing facilities
will pass $500 million.
A similar example is represented by
the fact that most colleges of engineering
and even many high schools now
use computers to train students in computer
programming. Perhaps just as
important as the imposition of formal
course requirements at the college level
is the increasingly widespread attitude
among college students that a knomledge
of computers is a “must” if their
engineering or scientific training is to be
up to date. Undergraduates of my generation
who majored in engineering, for
instance, considered a slide rule the
symbol of their developing technical
prowess. Today being able to program a
computer in a standard language such
as FORTRA4N or ALGOL is much more
likely to be the appropriate symbol.
At the graduate level students in the
social sciences and in business administration
are already making use of computers
in a variety of ways, ranging
from the large-scale analysis of data to
the simulation of an industry. The time
is rapidly approaching when a high percentage
of all university graduates will
have had some systematic training in
the use of computers; a significant percentage
of them Will have had quite
sophisticated training. An indication of
the growth of student interest in computers
is the increase in student units of
computer-science instruction we have
had at Stanford University over the past
four years. Although total enrollment at
Stanford increased only slightly during
that period; the number of student units
rose from 2,572 in 1962-1963 to 5,642
in 1965-1966.
The fact that time-sharing programs
are rapidly becoming operational in
many university computation centers
justifies the forecast of another increase
in the impact of computers on the universities
[ see “Time-sharing on Computers,”
by R. M . Fano and F. J. Corbató,
page 1281. Under time-sharing
regimes a much larger number of students
can be given direct “on line’’
experience, which in itself is psychologically
attractive and, from the practical
viewpoint, facilitates deeper study
of the use of computers. There is still
another far from trivial way in which
the computer serves the interests of
education: The large school system that
does not depend on computers for many
administrative and service functions is
today the exception rather than the rule.
The truly revolutionary function of
computers in education, however, lies
in the novel area of computer-assisted
instruction. This role of the computer is
scarcely implemented as yet but, assuming
the continuation of the present pace
of technological development, it cannot
fail to have profound effects in the near
future. In this article I shall describe
some experiments in computer-assisted
instruction that are currently being conducted
at levels ranging from the comparatively
simple to the quite complex
COMPUTER-$SSISTED ISSTRUCTION in elementary arithmetic is illustrated in the pho- and then examine unsuspected
tographs on the opposite page. A first.grade pupil, receiving “readiness” work preparatory
to instruction in addition, is shown tw-o possible answers to a question implicit in the sym. problems that these experiments have
bols occupying the top line of a cathode-ray-tube display. As he watches (top photograph), revealed. First, however, the reader dehis
earphones carry a verbal message asking him to select from the symbolic statements of an Of ‘%’ ‘Omputerunion
shown in the second and third lines of the display the one that is identical with the assisted instruction is desirequation
shown in the top line, The pupil signals his cho(ibcoet tom photograph)b y pointing able at all.
to the statement he prefers with machine’s light pen; the computer then records the answer. The single nlost powerful argument
I -
207
for conlputer-assisted instruction is an
old one in education. It concerns the
advantages, partly demonstrated and
partly conjectured, of individualized instruction.
The concept of individualized
instruction became the core of an explicit
body of doctrine at the end of the
19th century, although in practice it
was known some 2,000 years earlier in
ancient Greece. For many centuries the
education of the aristocracy was primarily
tutorial. At the university level
individualized tutorial instruction has
been one of the glories of Oxford and
Cambridge. hiodern criticisms of the
method are not directed at its intrinsic
merit but rather at its economic inefficiency.
It is widely agreed that the more
an educational curriculum can adapt in
a unique fashion to individual learners
-ench of whom has his own characteristic
initial ability, rate and even "style"
of learning-the better the chance is of
providing the student with a successful
learning experience.
The computer malces the individualization
of instruction easier because it
can be programmed to follow each student's
history of learning successes and
failures and to use his past performance
as a basis for selecting the new problems
and new concepts to which he
should be exposed next. With modern
information-storage devices it is possible
to store both a large body of curriculum
material and the past histories
of many students working in the curriculum.
Such storage is well within the
capacity of current technology, whether
the subject is primary school mathematics,
secondary school French or elementary
statistics at the college level. In
fact, the principal obstacles to computer-
assisted instruction are notechnological
but pedagogical: how to devise
ways of individualizing instruction and
of designing a curriculum that are suited
to individuals instead of groups. Certain
obvious steps that take account of different
rates of learning can be made with
little difficulty; these are the main things
that have been done so far. We have
still, however, cut only a narrow path
into a rich jungle of possibilities. We do
j i FIRST DAY S I
i FOURTH DAY^ i
I 1 SIXTH DAY 1 i
s0 60 40 20 0 2 4 12
STUDENT RESPONSES (PERCENT) RES?ONSE T!ME (MINUTES)
IMPROVEMEKT IN LEARNING is one evident result of drill and practice. The graph
summarizes the results of a six-day drill on the commutative, associative and distributive
laws of arithmetic. The computer program covered 48 concepts; each day's session pre.
sented 24 problems. Two days' drill therefore reviewed all 48 concepts, although no identical
problems were presented during the six days. By the last day student responses were
more than 90 percent correct and the speed of reply was twice what it was at the start.
not have any really clear scientific idea
of the extent to which instruction can be
individualized. It will próbably be mme
time before a discipline of such matters
begins to operate at anythingl ike an appropriately
deep conceptual level.
A second important aspect of computers
in education is closer in character
to such familiar administrative functions
as routine record-keeping. Before
the advent of computers it was extremely
difficult to collect systematic data on
how children succeed in the process
of learning a given subject. Evaluative
tests of achievement at the end of Iearning
have (and will undoubtedly continue
to have) a place both in the process of
classifying students and in the process
of comparing different curriculunl approaches
to the same subject. Nonetheless,
such tests remain blunt and insensitive
instruments, particularly with
respect to detailed problems of instruction
and curriculum revision. It is not
possible on the basis of poor results in
a test of children's mastery of subtraction
or of irregular verbs in French to
draw clear inferences about ways to improve
the curriculum. A computer, on
the other hand, can provide daily information
about how students are performing
on each part of the curriculum as it
is presented, making it possible to evaluate
not only individual pages but also
individual exercises. This use of computers
mill have important consequences for
all students in the immediate future.
Even if students are not themselves
receiving computer-assisted instruction,
the results of such instruction will certainly
be used to revise and improve ordinary
texts and workbooks.
et me now take up some of the work L' in computer-assisted instruction we
llave been doing at Stanford. It should
be emphasized that similar work is in
progress at other centers, including
the University of Illinois, Pennsylvania
State University, the University of
Pittsburgh, the University of Michigan,
the Vniversity of Texas, Florida State
University and the University of California
at Santa Barbara, and within
such companies as the International
Business Machines Corporation, the Systems
Development Corporation and
Bolt, ßeranek and Newman. This list is
by no means exhaustive. The work at
these various places runs from a primary
emphasis on the development of computer
hardware to the construction of
short courses in subjects ranging from
physics to typing. Although all these
efforts, including ours at Stanford, are
208
still in the developmental stage, the instruction
of large numbers of students at
computer terminals will soon (if academic
and industrial soothsayers are
right) be one of the most important
fields of application for computers.
At Stanford our students are mainly
at the elementary school level; the terminals
they use, however, are also suitable
for secondary school and university
students. ilt each terminal there is a
visual device on which the student may
view displays brought up from the computer
memory as part of the instruction
program. A device that is coming into
wide use for this purpose is the cathode
ray tube; messages can be generated
directly by the computer on the face of
the tube, which resembles a television
screen. Mounted with the cathode ray
tube is a typewriter keyboard the student
can use to respond to problems
shown on the screen. At some additional
cost the student can also have a light
pen that enables him to respond directly
by touching the pen to the screen instead
of typing on the keyboard. Such
a device is particularly useful for students
in the lowest elementary grades,
although when only single-digit numerical
responses or single-character
alphabetical ones are required, the use
' of a keyboard is quite easy even for
kindergarten children to learn.
i f t e r the display screen and the key- ' board the next most important element
at a terminal is the appropriate
sound device. Presenting spoken messages
to students is desirable at all educational
levels, but it is particularly
needed for younger children. It would
be hard to overemphasize the importance
of such spoken messages, programmed
to be properly sensitive to
points at which the student may encounter
difficulty in learning. Such
COMPUTER SUMIM$RP of drill results
makes possible the analysis essential for as.
sessment and revision of various study cur.
riculums. The results of 37 children's replies
to 20 questions designed to test elementary
arithmetic skills are summarized graphically
in this illustration. The most troublesome
question proved to be KO. 7 ; not only did
it take the most time to answer but also 26
students failed to answer it at all and only
two answered it correctly. Although question
No. 9 is the exact reverse of question
Xo. 7, it received 13 correct answers. Evidently
obtaining an unknown quantity by
subtraction is harder than obtaining one by
addition, and the students found it harder
to multiply 12 by 6 than to multiply 6 by 12.
! (6 X 11) +'- = 79
i ( 7 ~ 6 ) + - = 5 1
401 608 3 20 O 2 4 6
STUDENT RESPONSES (PERCENT) RESPONSE TIME (SECONDS)
213
messages are the main help a good tutor
gives his pupil; they are the crucial
missing element in noncomputerized
teaching machines. All of us have observed
that children, especially the
younger ones, learn at least as much by
ear as they do by eye. The effectiveness
of the spoken word is probably stronger
than any visual stimulus, not only for
children but also most of the time for
GLOSSARY
a MOIST AIR R SES
p MOIST AIS COOLS OR VV'ILL COOL
y CLOUDS WALL FORM
+ FORMAL I'dPLlCATION
7 h3T
RULES OF INFERENCE
TRI TRANS 'WTY 3- IMPLICATIOU
í120M Xi-Y AN3 Y+Z, DERIVE X+Z)
IF WODUS 2ON:NS
( 3 O M X--Y AND X, DERIVE Y)
CP CON-RAPS TIVE
(FROM X.+Y, DERIVE - Y + - X )
DNEG DOJB-E NFGATION
(FROM 7 TX, DERIVE X )
RED CONTRADICTIOU OF CONSEQJENT
IFPOM Y AN3 X--+ DERIVE -X)
TUTORIAL EXERCISE in mathematical
logic is an example of a more complex
variety of computer-assisted instruction. The
student may proceed from a set of given
hypotheses ( t o p ) to a given conclusion
(bottom) hy any one of several routes. Each
of the illustrated downward paths represents
a legitimate logical attack on the prob=
lem and each constitutes a unique sequence
of inferences (see legend und stutements in
logical notation below euch of the numbered
eerbul statements j . Ideally a tutorial comput.
er program will show no preference for one
path over another but will check the soundness
of each step along any path and tell the
student if he makes any mistakes in logic.
adults. It is particularly significant that
elementary school children, whose reading
skills are comparatively undeveloped,
comprehend rather complicated
spoken messages.
A cathode ray tube, a keyboard and
a loudspeaker or earphones therefore
constitute the essential devices for computer-
assisted instruction. Additional
visual displays such as motion pictures
or line drawings can also be useful at
almost all levels of instruction. Ordinary
film projectors under computer control
can provide such displays.
so fart hree levels of interactionb e- tween the student and the computer
program have received experimental attention.
At the most superficial level
(and accordingly the most economical
IF 2 4 -
214
one) are “drill and practice” systems.
Instruction programs that fall under this
heading are merely supplements to a
regular curriculunl taught by a teacher.
At Stanford we have experimented a
great deal with elementary school mathematics
at the drill-and-practice level, , and I shall draw on our experience for
examples of what can be accomplished
with this kind of supplementation of a
regular curriculum by computer methods.
Over the past 40 years both pedagogical
and psychological studies have provided
abundant evidence that students
need a great deal of practice in order to
master the algorithms, or basic procedures,
of arithmetic. Tests have shown
that the same situation obtains for students
learning the “new math.” There
seems to be no way to avoid a good deal
of practice in learning to execute the
basic algorithms with speed and accuracy.
At the elementary level the most lmportant
way in which computer-assisted
instruction differs from traditional methods
of providing practice is that we are
in no sense committed to giving each
child the same set of problems, as would
be the case if textbooks or other written
materials were used. Once a number
of study “tracks,” representing various
levels of difficulty, have been prepared
as a curriculum, it is only a matter of
computer programming to offer students
exercises of varying degrees of difficulty
and to select the appropriate level of
difficulty for each student according to
his past performance.
In the program we ran in elementary
grades at schools near Stanford during
the academic year 1965-1966 five levels
of difficulty were programmed for each
grade level. h typical three-day block of
problems on the addition of fractions,
for example, would vary in the following
way. Students at the lowest level (Level
1) received problems involving only
fractions that had the same denominator
in common, On the first two days levels
2 and 3 also received only problems in
which the denominators were the same.
On the third day the fraction problems
for levels 2 and 3 had denominators
that differed by a factor of 2. At Level
4 the problems had denominators that
differed by a factor of 2 on the first day.
At Level 5 the denominators differed
by a factor of 3, 4, 5 or 6 on the first
day. Under the program the student
moved up and down within the five
levels of difficulty on the basis of his
performance on the previous day. If
more than SO percent of his exercises
were done correctly, he moved up a
level. If fewer than 60 percent of the
exercises were done correctly, he moved
down a level. The selection of five levels
and of SO and 60 percent has no specific
theoretical basis; they are founded
on practical and pedagogical intuition.
As data are accumulated we expect to
modify the structure of the curriculum.
ur key effort in drill-and-practice
Osystems is being conducted in an
elementary school (grades three through
six) a few miles from Stanford. The
terminals used there are ordinary teletype
machines, each connected to our
computer at Stanford by means of individual
telephone lines. There are eight
teletypes in all, one for each school
classroom. The students take turns using
the teletype in a fixed order; each student
uses the machine once a day for
five to 10 minutes. During this period he
receives a number of exercises (usually
20), most of which are devoted to a
single concept in the elementary school
mathematics curriculum. The concept
reviewed on any given day can range
from ordinary two-digit addition to intuitive
logical inference. In every case
the teacher has already presented the
concept and the pupil has had some
DRILL-AKD-PRACTICE EXERCISE, shown in abbreviated form, is typical of a simple
computer-assisted instruction program that is designed to be responsive to the needs of in.
dividual students. The illustrated exercise is one of five that differ in their degree of di&.
culty; when the student types his name (color): the exercise best suited to him on the basis
cf computer-memory records of his previous performance is selected automatically. The
first three questions and answers exemplify the ways in which the computer is programmed
to deal with various shortcomings. The student fails to answer the first question within the
allotted 10.second time limit; the computer therefore prints TIME IS UP and repeats the
queetion, which the student then answers correctly (color). A wrong answer to the next
question causes the computer to announce the error and repeat the question automatically;
a second chance again elicits a correct answer. .2 wrong answer to the third question is compounded
by failure to respond to the reiterated question within the time limit. Because this
question has now drawn two unsatisfactory responses the automatic TIME IS UP statement
is followed by a printingo f the correct answer. The question is now repeated for a third and
last time. Whether or not the student elects to copy the correct answer (he dsoo eisn this in.
stance), the computer automatically produces the next question. Only soif xt he 20 questions
that compose the drill are shown in the example. After the student’s last answer the com.
puter proceeds to print a summaroyf the student’s score for the drill as well as his combined
average for this and earlier drills in the same series. The drill-and-practice exercise then
concludes with a cheery farewell to the student and an instructoti toena r off the teletype tape.
216
classroonl practice; the computer-assisted
drill-and-practice work therefore
supplements the teacher’s instruction.
The machine’s first instructionthe
teletype paper when the student
begins his drill. The number of characters
required to respond to this instruction
is by far the longest message the
elementary student ever has to type on
’ the keyboard, and it is our experience
that every child greatly enjoys learning
how to type his OUTI name. When the
name has been typed, the pupil’s record
is looked up in the master file at the
computer and the set of exercises he is
to receive is determined on the basis of
his performance the previous day. The
teletype now writes, for example, DRILL
604032. The first digit (6) refers to the
grade level, the next two digits (04) to
the number of the concept in the sequence
of concepts being reviewed during
the year, the next two digits (03)
to the day in terms of days devoted to
that concept (in this case the third day
devoted to the fourth concept) and the
final digit (2) to the level of difficulty
on a scale ranging from one to five.
The real work now begins. The computer
types out the first exercise [see
illustration on opposite page]. The carriage
returns to a position at which the
pupil should type in his answer. At this
point one of three things can happen. If
the pupil types the correct answer, the
computer immediately types the second
exercise. If the pupil types a wrong answer,
the computer types WRONG and
repeats the exercise without telling the
pupil the correct answer. If the pupil
does not answer within a fixed time (in
most cases 10 seconds), the computer
types TIME IS UP and repeats the exercise.
This second presentation of the
exercise follows the same procedure regardless
of whether the pupil mas wrong
or ran out of time on the first presentation.
If his answer is not correct at the
second presentation, however, the correct
answer is given and the exercise is
typed a third time. The pupil is now
expected to type the correct answer, but
whether he does or not the program
goes on to the next exercise. As soon as
the exercises are finished the computer
prints a summary for the student showing
the number of problems correct, the
number wrong, the number in which
time ran out and the corresponding percentages.
The pupil is also shown his
cumulative record up to that point, including
the amount of time he has spent
at the terminal.
A much more extensive summary of
PLEASE TYPE YOCR XAME-iS already On
student results is available to the teacher.
By typing in a simple code the
teacher can receive a summary of the
work by the class on a given day, of
the class’s work on a given concept, of
the work of any pupil and of a number
of other descriptive statistics I shall not
specify here. Indeed, there are so many
questions about performance that can
be asked and that the computer can
answer that teachers, administrators
and supervisors are in danger of being
swamped by more summary information
than they can possibly digest. We
are only in the process of learning what
summaries are most useful from the
pedagogical standpoint.
question that is often asked about
L‘ drill-and-practice systems is whether
we have evidence that learningis improved
by this kind of teaching. We do
not have all the answers to this complex
question, but preliminary analysis of
improvement in skills and concepts looks
impressive when compared with the
records of control classes that have not
received computer-assisted instruction.
Even though the analysis is still under
way, I should like to cite one example
that suggests the kind of improvement
that can result from continued practice,
even when no explicit instructions are
given either by the teacher or by the
computer program.
During the academic year 1964-1965
we noticed that some fourth-grade
pupils seemed to have difficulty changing
rapidly from one type of problem
format to another within a given set of
exercises. We decided to test whetheorr
not this aspect of performance would
improve with comparatively prolonged
practice. Because we were also dissatisfied
with the level of performance on
problems involving the fundamental
commutative, associative and distributive
laws of arithmetic, we selected 48
cases from this domain.
For a six-day period the pupils were
cycled through each of these 48 types of
exercise every two days, 24 exercises
being given each day [see illustration
on page 2081. No specific problem was
repeated; instead the same problem
types were encountered every two days
on a random basis. The initial performance
was poor, with an average probability
of success of .53, but over the
six-day period the advance in performance
was marked. The proportion of
correct answers increased and the total
time taken to complete the exercises
showed much improvement (diminishing
from an average of 630 seconds to
279 seconds). Analysis of the individual
data showed that every pupil in the
class had advanced both in the proportion
of correct responses and in the reduction
of the time required to respond.
The next level of interaction of the
pupil and the computer program is
made up of “tutorial” systems, which
are more complex than drill-and-practice
systems. In tutorial systems the aim
is to take over from thec lassroom teacher
the main responsibility for instruction.
As an example, many children who
enter the first grade cannot properly use
the words “top” and “bottom,” “first”
and “last” and so forth, yet it is highly
desirable thathe first-grader have a
clear understanding of these words
so that he can respond in unequivocal
fashion to instructions containing
them. Here is a typical tutorial sequence
we designed to establish these concepts:
1. The child uses his light pen to point to
the picture of a familiar object displayed
on the cathode-ray-tube screen. 2. The
child puts the tip of his light pen in a
small square box displayed next to the
picture. (This is the first step in preparing
the student to make a standard response
to a multiple-choice exercise.)
3. The words FIRST and LAST are introduced.
(The instruction here is spoken
rather than written; FIRST and LAST refer
mainly to the order in which elements
are introduced on the screen from
left to right.) 4. The words TOP and
BoTTohf are introduced. (An instruction
to familiarize the child with the use of
these words might be: PUT YOUR LIGHT
TOP.) 5. The two concepts are combined
in order to select one of several things.
(The instruction might be: PUT YOUR
LIGHT PEN ON THE FIRST ANIMAL SHOWX
With such a tutorial system we can
individualize instruction for a child
entering the first grade. The bright
child of middle-class background who
has gone to kindergarten and nursery
school for three years before entering
the first grade and has a large speaking
vocabulary could easily finish work on
the concepts I have listed in a single
30-minute session. 4 culturally deprived
child who has not attended kindergarten
may need as many as four or five sessions
to acquire these concepts. It is important
to keep the deprived child from
developing a sense of failure or defeat
at the start of his schooling. Tutorial
“branches” must be provided that move
downward to very simple presentations,
just as a good tutor will use an increasingly
simplified approach when he re-
PEh‘ ON THE TOY TRUCK SHOWK AT THE
AT THE TOP.)
p
alizes that his pupil is failing to understand
what is being said. It is equally
important that a tutorial program have
enough flexibility to avoid boring a
bright child with repetitive exercises he
already understands. \Ve have found it
best that each pupil progress from one
concept in the curriculum to another
only after he meets a reasonably stiff
criterion of performance. The rate at
which the brightest children advance
may be five to 10 times faster than that
of the slowest children.
I n discussing curriculum materials one
commonly distinguishes between
“multiple-choice responses” and “constructed
responses.” Multiple-choice exercises
usually limit the student to three,
four or five choices. A constructed response
is one that can be selected by the
student from a fairly large set of possibilities.
There are two kinds of constructed
response: the one that is
uniquely determined by the exercise and
the one that is not. Although a good
part of our first-grade arithmetic program
a l l o ~ sc onstructed responses, almost
all the responses are unique. For
example, when we ask for the sum of
2 plus 3, we expect 5 as the unique
response. We have, however, developed
a program in mathematical logic that
allows constructed responses that are
not unique. The student can make any
one of several inferences; the main function
of the computer is to evaluate the
validity of the inference he makes.
\.”hether or not the approach taken by
the student is a wise one is not indicated
until he has taken at least one step in an
attempt to find a correct derivation of
the required conclusion. No two students
need find the same proof; the
tutorial program is designed to accept
any proof that is valid [see illzlstsation
on pages 214 and 21 51. When the student
makes a mistake, the program tells
him what is wrong with his response;
when he is unable to take another step,
the program gives him a hint.
It will be evident from these examples
that well-structured subjects such as
reading and mathematics can easily be
handled by tutorial systems. At present
they are the subjects we best understand
how to teach, and we should be
able to use computer-controlled tutorial
systems to carry the main load of teaching
such -subjects. It should be empha-
ESSENTIAL COMPONENTS that allow interaction of computer ticularly important in primary school instruction. Students may
and student are grouped at this terminal console. The cathode ray respond to instructions by use of the terminal’s keyboard or by
tube (right) replaces the earlier teletypewriter roll as a more use of a light pen (extreme right); programs that will enable the
flexible means of displaying computer instructions and questions. computer to receive and respond to the student’s spoken words are
Earphones or a loudspeaker reproduce spoken words that are par. under study. Supplemental displays are shown on the screen at left.
218
sized, however, that no tutorial program
designed in the near future will be able
to handle every kind of problem that
arises in student learning. It will remain
the teacher’s responsibility to attempt
the challenging task of helping students
who are not proceeding successfully
with the tutorial program and who need
special attention.
Thus a dual objective may be
achieved. 9 o t only will the tutorial program
itself be aimed at individualized
instruction but also it will free the
teacher from many classroom responsibilities
so that he will have time to individualize
his own instructional efforts.
At Stanford we program into our tutorial
sessions an instruction to the computer
that we have named TEACHER CALL.
When a student has run through all
branches of a concept and has not yet
met the required criterion of performance,
the computer sends a teacher call
to the proctor station. The teacher at
the proctor station then goes to the
student and gives hin1 as much individualized
instruction as he needs.
the third and deepest level of stu-
L dent-computer interaction are systems
that allow a genuine dialogue
between the student and the program.
“Dialogue systems” exist only as elementary
prototypes; the successful implementation
of such systems will
require the solving of two central problems.
The first may be described as
follows: Suppose in a program on economic
theory at the college level the
student types the question: TVHY ARE
DEMAND CURVES ALWAYS CONVEX WITH
RESPECT TO THE ORIGIS? It is difficult to
write programs that will recognize and
provide answers to questions that are so
broad and complex, yet the situation is
not hopeless, In curriculum areas that
have been stable for a long time and
that deal with a clearly bounded area of
subject matter, it is possible to analyze
the kinds of questions students ask; on
the basis of such an analysis one can
make considerable progress toward the
recognition of the questions by the computer.
Nonetheless, the central iatellectua1
problem cannot be dodged. iIst not
enough to provide information that will
give an answer; what is needed is an
ability on the part of the computer program
to recognize precisely what question
has been asked. This is no less than
asking the computer program to understand
the meaning of a sentence,
The second problem of the dialogue
system is one that is particularly critical
with respect to the teaching of elementary
school children. Here it is essential
that the computer program be able to
recognize the child’s spoken words. A
child in the first grade will probably not
be able to type even a simple question,
but he can voice quite complex ones.
The problem of recognizing speech adds
another dimension to the problem of
recognizing the meaning of sentences,
In giving an example of the kind of
dialogue system we are currently developing
at Stanford I must emphasize
that the program Ia m describing (which
represents an extension of our work in
mathematical logic) is not yet wholly
operational. Our objective is to introduce
students to simple proofs using the
associative and commutative laws and
also the definitions of natural numbers
as successors of the next smallest number
(for example, 2 = 1 + 1, 3 = 2 + 1
and 4 = 3 + 1). Our aim is to enable
the student to construct proofs of simple
identities; the following would be typical
instances: 5 = 2 + 3 and 8 = (4 + 2) +
2. 1%7e want the student to be able to
tell the conlputer boyr al command what
steps to take in constructing the proof,
using such expressions as REPLACE 2 BY
LISE 3. This program is perfectly practical
with our present computer system as
long as the commands are transmitted
by typing a few characters on the keyboard.
A major effort to substitute voice
for the keyboard is planned for the coming
year; our preliminary work in this
direction seems promising.
But these are essentially technological
problems. In summarizing some other
problems that face us in the task of
realizing the rich potential of computer-
assisted individual instruction, I
should prefer to emphasize the behavioral
rather than the technological
ones. The central technological problem
must be mentioned, however; it has to
do with reliability. Computer systems
in education must work with a much
higher degree of reliability than is expected
in computer centers where the
users are sophisticated scientists, or
even in factory-control systems where
the users are experienced engineers. If
in the school setting young people are
put at computer terminals for sustained
periods and the program and nlachines
do not perform as they should, the result
is chaos. Reliability is as important in
schools as it is in airplanes and space
vehicles; when failure occurs, the disasters
are of different kinds, but they
are equally conclusive.
The primary behavioral problem involves
the organization of a curriculum.
1 -f 1 Or USE THE ASSOCIATIVE LAW Oh’
For example, in what order should the
ideas in elementary mathematics be
presented to students? In the elementary
teaching of a foreign language, to
what extent should pattern drill precede
expansion of vocabulary? What mixture
of phonics and look-and-say is appropriate
for the beginning stages of reading?
These are perplexing questions.
They inevitably arise in the practical
context of preparing curriculum materials;
unfortunately we are far from having
detailed answers to any of them. Individualized
instruction, whether under
the supervision of a computer or a
human tutor, must for some time proceed
on the basis of practical judgment
and rough-and-ready pedagogical intuition.
The magnitude of the problem of
evolving curriculum sequences is difficult
to overestimate: the number of possible
sequences of concepts and subject
matter in elementary school mathematics
alone is in excess of 10100, a
number larger than even generous estimates
of the total number of elementary
particles in the universe.
One of the few hopes for emerging
from this combinatorial jungle lies in the
development of an adequate body of
fundamental theory about the learning
and retention capacity of students. It is
to be hoped that, as systematic bodies
of data become available from computer
systems of instruction, we shall be able
to think about these problems in a more
scientific fashion and thereby learn to
develop a more adequate fundamental
theory than we now possess.
Another problem arises from the
fact that it is not yet clear how critical
various kinds of responses may be. I
have mentioned the problem of interpreting
sentences freely presented by
the student, either by the written or by
the spoken word. How essential complex
constructed responses to such questions
may be in the process of learning
most elementary subjects is not fully
known. il problem at least as difficult as
this one is how computer programs can
be organized to take advantage of unanticipated
student responses in an insightful
and informative way. For the
immediate future perhaps the best we
can do with unanticipated responses is
to record them and have them available
for subsequent analysis by those responsible
for improving the curriculum.
The possible types of psychological
“reinforcement” also present problems.
The evidence is conflicting, for instance,
whether students should be immediately
informed each time they malte a mistake.
It is not clear to what extent stu-
219
dents should be forced to seek the right
answer, and indeed whether this search
should take place primarily in what is
called either the discovery mode or the
inductive mode, as opposed to more traditional
methods wherein a rule is given
and followed by examples and then by
exercises or problems that exemplify the
rule. Another central weakness of traditional
psychological theories of reinforcement
is that too much of the theory
has been tested by experiments in which
the information transmitted in the reinforcement
procedure is essentially very
simple; as a result the information content
of reinforcement has not been sufficiently
emphasized in theoretical discussions.
A further question is whether
or not different kinds of reinforcement
and different reinforcement schedules
should be given to children of different
basic personality types. 4 s far as I know,
variables of this kind have not been
built into any large-scale curriculum
effort now under way in this country,
Another pressing problem involves
the effective use of information about
the student's past performance. In standard
classroom teaching it is impossible
to use such records in a sensitive \=,'ay;
we actually have little experience in the
theory or practice of the use of such
information. A gifted tutor will store in
his own memory many facts about the
past performance of his pupil and take
advantage of these facts in his tutorial
course of study, but scientific studies of
how this should be done are in their
infancy. Practical decisions abouthe
amount of review work needed by the
individual, the time needed for the introduction
of new concepts and so forth
will be mandatory in order to develop
the educational computer systems of the
future. Those of us who are faced with
making these decisions are aware of
the inadequacy of our knowledge. The
power of the computer to assemble and
provide data as a basis for such decisions
will be perhaps the most pomerfu1
impetus to the development of education
theory yet to appear. It is likely
that a different breed of education
research worker will be needed to feel
at home with these vast masses of data.
The millions of observational records
that computers now process in the field
of nuclear physics will be rivaled in
quantity and complexity by the information
generated by computers in the field
of instruction.
IVhen students are put to work on an
individualized basis, the problem of
keeping records of their successes and
failures is enormous, particularly when
220
those records are intended for use in
making decisions about the next stage
of instruction. In planning ways to process
the records of several thousand students
at Stanford each day, we found
that one of the most difficult decisions is
that of selecting the small amount of
total information it is possible to record
permanently. It is not at all difficult to
have the data output run to 1,000 pages
a day when 5,000 students use the terminals.
An output of this magnitude is
simply more than any human being can
digest on a regular basis. The problem
is to reduce the data from 1,000 pages
to something like 25 or 30. As with the
other problems I have mentioned, one
difficulty is that we do not yet have the
well-defined theoretical ideas that could
provide the guidelines for making such
a reduction. At present our decisions are
based primarily on pedagogical intuition
and the traditions of data analysis
in the field of experimental psychology.
Neither of these guidelines is very
effective.
body of evidence exists that at-
,!! tempts to show that children have
different cognitive styles. For example,
they may be either impulsive or reflective
in their basic approach to learning.
The central difficulty in research on cognitive
styles, as it bears on the construction
of the curriculum, is that the research
is primarily at an empirical level.
It is not at all clear how evidence for
the existence of different cognitive styles
can be used to guide the design and
organization of individualized curriculum
materials adapted to these different
styles, Indeed, what we face is a fundamental
question of educational philosophy:
To what extent does society want
to commit itself to accentuating differences
in cognitive style by individualized
techniques of teaching that cater to
these differences? The introduction of
computers in education raises this question
in a new and pressing way. The
present economics of education is sllch
that, whatever we may think about the
desirability of having a diverse curriculum
for children of different cognitive
styles, such diversity is not possible because
of the expense. But as computers
become widely used to offer instruction
in the ways I have described here, it
will indeed be possible to offer a highly
diversified body of curriculum material.
'cl'hen this occurs, we shall for the first
time be faced with the practical problem
of deciding how much diversity
we mant to have. That is the challenge
for which we should be prepared.
Systems Applications Analyst-
Investigate and develop information
handling systems based upon the
use of Xerox products, technology,
systems design and engineering
capabilities. Investigate ways of using
equipment in new and novel applications
and prepare proposals. Support
field salesmen in analysis of
customers' needs. Identify and define
needs for new or modified products
to expand applications of current
products. Minimum of four years
experience in business systems
design, applications engineering,
sales, or manufacturing. BS in
technical field, MBA desirable.
Software Development/Systems
Programmers-
Must have awareness and interest in
advanced state of the art software.
Familiarity with query language and
compiler techniques. Work in query
language, data base management,
time sharing, graphics applications,
and information storage and retrieval
applications to specific problem
areas. Analyze requirements and recommend
state of the art solutions with
Xerox products. Requires the ability
to communicate both orally and in
writing. Degree in Engineering,
Physical Sciences, Math, or related
areas. Experience in systems design
of large information retrieval, communications
and time shared applications.
Programming experience in
language development helpful.
Test Engineers-
BSEE or BSME required with5 -8
years' experience in product design,
field engineering, manufacturing
and/or quality control involving complex
electro-mechanical or electronic
systems. Responsible for developing
test programs from initial design
concept to prototype and manufacturing
evaluations on complex systems
in electronic, mechanical and optical
fields. Also design analysis and
process planning and modification.
These positions are in suburban
Rochester, New York. Please forward
your resume, including salary history,
in confidence to: Mr. G. L. Swanson,
Dept. BBJ-118, Xerox Corporation,
P.O. Box 1540, Rochester,
New York 14603.
An Equal Opportunity Employer