Specific Instructional Design Principles and Effects

Specific Instructional Design Principles and Effects
There are a range of specific instructional design principles and eVects that
flow from the considerations outlined in this chapter. Cognitive load theory,
an instructional theory based on the combination of information structures
and cognitive architecture described earlier, has been used to generate those
instructional eVects.
1. The Goal-Free EVect
This eVect occurs when learners presented a conventional, goal-specific
problem in which the goal might be ‘‘calculate the value of angle ABC’’ in
the case of a geometry problem or ‘‘calculate the final velocity of the
vehicle’’ in the case of a kinematics problem learn less than learners
presented a nonspecific or goal-free problem. Examples of nonspecific goal
problems are ‘‘calculate the value of as many angles as you can’’ or
‘‘calculate the value of as many variables as you can.’’ The goal-free eVect
was obtained by Sweller and Levine (1982) and has been demonstrated on
many occasions (Ayres, 1993; Sweller, & Cooper, 1985; Burns & Vollmeyer,
248 John Sweller2002; Geddes & Stevenson, 1997; Miller, Lehman, & Koedinger, 1999;
Owen & Sweller, 1985; Paas, Camp, & Rikers, 2001; Sweller, 1988; Sweller
et al., 1983; Tarmizi & Sweller, 1988; Vollmeyer, Burns, & Holyoak, 1996).
It can be explained using the cognitive matrix of continua of Fig.1.
Assume a novice problem solver solving conventional problems by
means-ends analysis. As a novice, he or she will be on the left side of the
matrix of continua. To make moves, diVerences between the current state
and the goal state will need to be established, a potential move will need to
be chosen randomly (assuming prior knowledge concerning the eVects of
particular moves is unavailable), and each potential move will need to be
assessed to establish whether it reduces diVerences between the current
problem state and the goal state. Because working memory limitations are
relevant on the left side of the matrix of continua, this complex procedure
may leave few or no resources available to attend to schema acquisition.
When acquiring a schema, learners must engage in the quite diVerent
activity of learning to classify problems and problem states according to
their moves. As a consequence, learning may be inhibited.
In contrast, assume a problem solver who is presented goal-free problems.
The only activity that needs to be engaged in is to choose any potential
moves randomly and determine whether they can be made. Working
memory load is minimal. Furthermore, learning which moves can be made
given a particular problem state is central to schema acquisition. Sweller
(1988) suggested that this interpretation explains the goal-free eVect.
Presenting learners with goal-free problems may appear unusual if the
aim is to present learners with direct, fully guided instruction. Goal-free
problems reduce the guidance provided by a specific goal. For this reason,
the procedure is eVective, but only if all moves made under goal-free
conditions are useful in the sense that they need to be learned and practiced.
Not all problems have this characteristic. Some problems have a large or
even infinite number of moves that could be made with most moves serving
no function. For example, asking learners to make as many manipulations
as possible of the equation ða þ bÞ=c ¼ d can result in an infinite number of
manipulations, as one can legitimately add an infinite number of constants
to each side. Goal-free problems should not be used with such material and
so an alternative is required.
2. The Worked Example EVect
The use of worked examples can overcome the problem of goal-free
problems only being useful for a limited class of problems. There are
probably no classes of problems for which worked examples are not
potentially eVective.
Evolution of human cognitive architecture 249The worked example eVect occurs when learners who are presented with a
large number of worked examples to study learn more than learners
presented an equivalent number of problems to solve. The eVect has been
studied extensively (Carroll, 1994; Cooper & Sweller, 1987; Miller et al.,
1999; Paas, 1992; Paas & van Merrienboer, 1994; Pillay, 1994; Quilici &
Mayer, 1996; Sweller & Cooper, 1985; Trafton & Reiser, 1993).
Worked examples provide problem-solving guidance that can act as a
substitute for schemas that are unavailable to novices. They are the ultimate
form of direct instruction. Rather than engaging in the means-ends
problem-solving search process described earlier, learners can be guided by
a worked example acting as a substitude schema-based central executive.
The lack of such a central executive necessitates problem-solving search,
with its inevitable random components and working memory load found on
the left side of the matrix of continua. While psychologically the learner is
on the left side of the matrix of continua, a worked example allows him or
her to perform as though they are on the right side of the matrix. A good
example acts as a substitute for a schema-based central executive, eliminates
the problem-solving search with its random base, and reduces diYculties
imposed by a limited working memory because all necessary information is
incorporated within the example (see later sections on split-attention,
modality, and redundancy eVects). As a consequence, learning can be
facilitated by an emphasis on worked examples resulting in the worked
example eVect.
3. The Problem Completion EVect
Most demonstrations of the worked example eVect involve presenting
worked examples paired with very similar problems. Learners are presented
a worked example and are then immediately presented a very similar
problem to solve. This procedure ensures that learners are motivated to
study the worked example in order to ensure that they can solve the
following problem. The extent to which they can solve the following
problem also provides them with some feedback concerning their ability to
solve problems of that type.
Completion problems were invented as an alternative to this procedure.
Rather than presenting learners with full worked examples followed by
similar problems, they are presented with partial worked examples that
require completion. The partial worked example provides suYcient guidance
to reduce the problem-solving search and cognitive load, whereas problem
completion ensures that learners are motivated and receive feedback
concerning their knowledge of relevant problem types. Paas (1992), Paas
and van Merrienboer (1994), van Merrienboer (1990), van Merrienboer and
250 John Swellerde Croock (1992), van Merrienboer and Krammer (1987), and van
Merrienboer, Schuurman, de Croock, and Paas (2002) provided evidence
that completion problems have a positive eVect similar to that of worked
examples when compared to full problems. It is reasonable to assume that the
theoretical reasons for the problem completion eVect are identical to those
used to explain the worked example eVect.
4. The Split-Attention EVect
Not all worked examples are eVective. A worked example that is structured
in a manner that ignores working memory limitations may be no more or
even less eVective than solving the equivalent problem. Some worked
examples in some areas are conventionally structured in a manner that
requires learners to split their attention between multiple sources of
mutually referring information before mentally integrating those sources
of information. A conventional geometry worked example consisting of a
diagram and statements provides an instance. The diagram in isolation
provides no instruction. The associated statements, such as angle
ABC ¼ angle XYZ, are unintelligible without a diagram. Meaning can only
be derived from the worked example by mentally integrating the diagram
and the statements. Mental integration requires working memory resources
because learners must search for referents. When a geometry statement
refers to angle ABC, learners must search the diagram for angle ABC in
order to understand the statement. In eVect, the learner is not only on the
left side of the matrix of continua for geometry, but is on the left side of the
matrix for the particular example being studied. An act of problem solving
must be engaged in simply to locate appropriate referents. Locating
referents requires working memory resources that are unavailable for
learning geometry.
Because we do not normally have schemas for the labeling of particular
geometry diagrams, providing guidance in locating referents can be just as
beneficial as guidance in the more general terms discussed previously. Such
guidance can be provided by physically integrating diagrams and
statements. Rather than placing the statement angle ABC ¼ Angle XYZ
below or next to the diagram as normally occurs, the relevant statements can
be incorporated within the diagram so that a search for referents is
eliminated. If conventionally structured worked examples are compared
with physically integrated examples, results normally demonstrate an
advantage for the integrated versions, resulting in the split-attention eVect.
Various versions of the eVect have been demonstrated using a wide variety
of materials under a wide variety of conditions. Furthermore, as might be
expected, it is not restricted to worked examples but applies to any
Evolution of human cognitive architecture 251instructional material (Bobis et al., 1993; Cerpa, Chandler, & Sweller, 1996;
Chandler & Sweller, 1992, 1996; Mayer & Anderson, 1991,1992; Mwangi &
Sweller, 1998; Sweller et al., 1990; Tarmizi & Sweller, 1988; Ward & Sweller,
1990).
5. The Modality EVect
While physical integration of multiple sources of information can be highly
eVective, there is an alternative that is equally eVective and, under some
circumstances, may be preferable. The split-attention eVect relies on visual
modality with visual search being reduced by the use of physical integration.
Visual search means that the visual channel only (the visuospatial sketch
pad of Baddeley, 1992; Baddeley & Hitch, 1974) is being used and
overloaded under split-attention conditions. Considerable evidence, shows
that eVective working memory can be increased by using dual rather than a
single modality (e.g., Penney, 1989). While the visual and auditory
processors of working memory are not fully separate in the sense that one
does not obtain a simple additive increase in processing capacity by
presenting some material visually and some in auditory mode, there is
considerable empirical evidence of a measurable increase in working
memory capacity when using both modalities (Allport, Antonis, &
Reynolds, 1972; Brooks, 1967; Frick, 1984; Levin & Divine-Hawkins,
1974). From a theoretical perspective, capacity should increase to the extent
that visual and auditory processors can function autonomously without
sharing other cognitive structures that limit capacity. Some empirical
evidence of an increase in working memory capacity when using both
modalities also provides evidence for partial autonomy of the auditory and
visual channels.
The possibility of increasing working memory capacity using dual
rather than a single modality should have instructional consequences. For
example, under split-attention conditions, rather than presenting a diagram
and written text that should be integrated physically, it may be possible to
present a diagram and spoken text. Because the diagram uses visual
modality while speech uses auditory modality, the total available working
memory capacity should be increased, resulting in enhanced learning.
The instructional modality eVect occurs when learners, faced with two
sources of information that refer to each other and are unintelligible in
isolation, learn more when presented with one source in visual mode and the
other in auditory mode rather than both in visual mode. This eVect has been
demonstrated on a number of occasions (Jeung, Chandler, & Sweller, 1997;
Mayer & Moreno, 1998; Moreno & Mayer, 1999; Mousavi, Low, & Sweller,
1995; Tindall-Ford et al., 1997).
252 John Sweller6. The Redundancy EVect
Both split-attention and modality eVects occur under very specific
conditions. They are only obtainable when multiple sources of information
refer to each other and are unintelligible in isolation. For example, a
diagram and text will not yield either split-attention or modality eVects if the
diagram is fully intelligible and fully provides the information needed, with
the text merely recapitulating the information contained in the diagram in a
diVerent form. Under such circumstances, the text is redundant. The
redundancy eVect occurs when additional information, rather than having a
positive or neutral eVect, interferes with learning. For example, instead of
integrating a diagram with redundant text or presenting the text in auditory
form, learning is enhanced by eliminating the text.
There are many diVerent forms of redundancy. The previous diagram/text
redundancy occurs when a self-explanatory diagram has additional text
redescribing the diagram (Chandler & Sweller, 1991). Mental activity/
physical activity redundancy occurs when, for example, learning how to use
a computer application by reading a text has the added physical activity of
using the computer (Cerpa et al., 1996; Chandler & Sweller, 1996; Sweller &
Chandler, 1994). Either reading the text in a manual or, surprisingly,
physically using a computer can be redundant and interfere with learning.
Summary/detailed exposition redundancy occurs when a summary alone
results in enhanced learning compared to a full exposition (Mayer, Bove,
Bryman, Mars, & Tapangco, 1996; Reder & Anderson, 1980, 1982) Finally,
auditory/visual redundancy occurs when the same material, presented
simultaneously in written and spoken form, results in a learning decrement
compared to the material presented in written or auditory form alone
(Craig, Gholson, & Driscoll, 2002; Kalyuga, Chandler, & Sweller, 1999,
2000; Mayer, Heiser, & Lonn, 2001).
The redundancy eVect is one of the more surprising cognitive load
eVects, with many people finding it quite counterintuitive. Most of us feel
that even if additional explanatory material is not beneficial, at the very
least it should be neutral. In fact, the addition of redundant information
can have strong, negative consequences. The eVect can be understood
in cognitive load theory terms. If one form of instruction is intelligible
and adequate, providing the same information in a diVerent form will
impose an extraneous cognitive load. Working memory resources will
need to be used to process the additional material and possibly relate it
to the initial information. It is likely to be only after the learner has
processed the additional information that he or she will be aware that
the activity was unnecessary. At that point, the damage may have been
done.
Evolution of human cognitive architecture 2537. The Element Interactivity EVect
Split-attention, modality, and redundancy eVects all occur as a consequence
of instructional procedures designed to reduce working memory load. It
might be expected that the instructional procedures would only be eVective
where the material being learned imposed an intrinsically high cognitive
load. If material does not impose a high cognitive load, the additional load
due to inadequate instructional procedures may not matter a great deal
because working memory capacity may not be exceeded. An extraneous
cognitive load due to inadequate instructional procedures may be irrelevant
if the intrinsic cognitive load imposed by the structure of the information is
low. Because low element interactivity material has a low intrinsic cognitive
load, we can predict that cognitive load eVects may disappear when learning
such material. The eVects may only be obtainable using high element
interactivity material. Chandler and Sweller (1996) and Sweller and
Chandler (1994) demonstrated that split-attention and redundancy eVects
could be demonstrated readily using high element interactivity material but
disappeared when low element interactivity material was used. Tindall-Ford
et al. (1997) similarly found that the modality eVect could only be obtained
using high element interactivity material. Marcus et al. (1996) found that
diagrams for which we have schemas facilitated understanding when
compared to text but only under conditions of high element interactivity.
The finding that cognitive load eVects can only be obtained using high
element interactivity material demonstrates the element interactivity eVect.
It consists of an interaction between the split-attention, redundancy, and
modality eVects and the complexity (as measured by element interactivity)
of the material being learned. While it has not been tested using other
cognitive load eVects, there is every reason to suppose that it could be
obtained with all other eVects based on a limited working memory.
It has been suggested in this chapter that the particular interaction
between a limited working memory and a very large long-term memory had
to evolve in order to handle high element interactivity material. High
element interactivity material must be imbedded in schemas before it can be
handled by a limited working memory. The element interactivity eVect
indicates that when instruction deals with high element interactivity
material, then the characteristics of human cognitive architecture, such as
a limited working memory, become critical.
8. The Isolated Interacting Elements EVect
Consider a learner faced with new material. That learner is on the left side of
the cognitive matrix of continua. Consider further that element interactivity
of the information that must be assimilated is suYciently high to
254 John Swellersubstantially exceed working memory capacity. Understanding cannot
occur because understanding requires all interacting elements to be
processed simultaneously in working memory. All the interacting elements
cannot be processed simultaneously in working memory until schemas have
been formed, but schemas will not be formed until the learner has moved
toward the right of the matrix of continua. Because the learner cannot
possibly understand the material until those schemas have been formed,
understanding and learning may appear impossible at first sight. When the
material is presented with all of its interacting elements, as it needs to be if
understanding is to occur, it cannot be processed in working memory
because it vastly exceeds working memory capacity. How does learning
occur under such circumstances?
One possibility (perhaps the only possibility) is that initially the elements
must be learned as though they are isolated, noninteracting elements. Once
suYciently sophisticated schemas have been constructed, understanding will
occur because the interacting elements can now be processed in working
memory. On this analysis, learning must precede understanding.
If this analysis is valid, it is reasonable to hypothesize that learning might
be facilitated by initially presenting very complex information to students in
isolated elements form without emphasizing or even indicating the
interactions between elements. Understanding of such instruction will be
limited, but once it has been learned, presentation of the full information
may permit understanding to occur. In contrast, presentation of the
complete information that potentially could be understood during initial
instruction may result in very little learning or understanding. Pollock et al.
(2002) obtained precisely this eVect. Learners presented isolated elements to
learn followed by the full set of interacting elements learned more than
learners presented the full set of interacting elements twice, demonstrating
the isolated interacting elements eVect.
9. The Imagination EVect
Assume a novice on the left of the cognitive matrix of continua has
acquired some schemas and is beginning to move toward the right of the
continua. To attain relatively high levels of expertise, further learning will
need to include automation of the previously acquired schemas that
normally includes continuing to study the material until desired levels of
performance have been attained. An alternative is to attempt to imagine
the procedures that have been learned. Imagining requires the learner to
mentally ‘‘run through’’ the procedures in working memory. For high
element interactivity material, processing information in working memory
is impossible until schemas have been acquired. Once they have been
Evolution of human cognitive architecture 255acquired and the learner has moved toward the right of the matrix of
continua, imagination techniques should be feasible and practice through
imagination should assist in automation. Continuing to study the material
should be unnecessary because studying high element interactivity material
is designed to provide the guidance necessary to reduce search while
acquiring schemas, as occurs on the left side of the matrix of continua. If
schemas have already been acquired, there is no longer any need to provide
instructional guidance to reduce search because, on the right of the matrix
of continua, the central executive function of schemas is now able to
operate. Using those schemas to imagine the procedures learned should
facilitate further learning through automation in a manner that studying
the instructions may not.
Cooper, Tindall-Ford, Chandler, and Sweller (2001) tested this hypothesis
and found that learners given instructions to ‘‘imagine’’ a set of procedures
that needed to be learned performed better than learners given conventional
‘‘study’’ instructions. This imagination eVect was only obtained using
learners with suYcient knowledge to be able to process all of the required
information in working memory. For complete novices who were unable to
process the high element interactivity material in working memory, a reverse
imagination or ‘‘study’’ eVect was obtained with ‘‘study’’ instructions
proving superior to ‘‘imagination’’ instructions. In other words, the eVect
obtained depended on the levels of expertise of the learners. Higher levels
of expertise could reverse the eVect obtained. The ideal form of
instruction depended on the expertise of the learners. This reversal eVect
with expertise, as it happens, is general. As described in the next section,
most, perhaps all, of the cognitive load eVects described earlier depend on
the use of novices.
10. The Expertise Reversal EVect
With the exception of the imagination eVect, all of the previously described
eVects were intended to provide new instructional procedures for novices
who were on the far left of the cognitive matrix of continua. Learners, of
course, continue to learn and may require instructional procedures after
they have moved beyond the left of the matrix of continua. It turns out that
frequently, once learners have acquired some knowledge, many of the eVects
described previously reverse. With increased experience, conventional
instructional procedures, such as practice at solving problems, are better
than cognitive load procedures, such as studying worked examples. The
imagination eVect diVers from the other eVects discussed in that the
imagination technique is intended for more knowledgeable learners rather
than complete novices and so reverses when the imagination technique is
256 John Swellerpresented to novices rather than the more experienced learners. In all other
cases, the eVects shown using novices are reversed when using more
experienced learners. The reversal is due to the redundancy eVect and is
called the expertise reversal eVect. It is due to an interaction between simpler
cognitive load eVects and levels of expertise and can be contrasted with the
element interactivity eVect, discussed earlier, which consists of an
interaction between simpler cognitive load eVects and task complexity.
Using diagrams and text, Kalyuga, Chandler, and Sweller (1998) obtained
the normal split-attention eVect with integrated diagrams and text proving
superior to a split-attention format. A group presented the diagrams alone
performed poorly because the text was essential in helping understand the
diagram, a necessary condition for the split-attention eVect. The learners
used were novices on the left side of the cognitive matrix of continua. Over
several months training in the relevant, engineering area, they moved
toward the right of the matrix of continua. The necessary guidance provided
by the text became less and less essential as schemas were acquired to take
over from the text. The superiority of the integrated format decreased with
increased expertise. A point was reached where there was no diVerence
between groups. Eventually, with additional training, the text became
redundant. Learners could understand and learn from a diagram alone.
Having to process unnecessary text increased the cognitive load. The
presence of redundant text, especially in integrated form where it is diYcult
to ignore, interfered with rather than facilitated learning. A redundancy
eVect was obtained with the diagram-alone condition providing the best
learning environment.
Yeung, Jin, and Sweller (1998) obtained a similar eVect using textual
materials. Learners with low levels of language competence were assisted by
explanatory notes integrated into the primary text. Integrated notes retarded
learning for learners with higher levels of language competence because the
notes were redundant but were diYcult to ignore when integrated into the
primary text.
Other cognitive load eVects also disappear and then reverse with increased
expertise. A modality eVect obtained with novices disappeared and then
reversed (Kalyuga, Chandler,&Sweller, 2000) as expertise increased.Novices
required textual material to assist them understand visually presented
material; that textual material was best presented in spoken rather than
written form, demonstrating the modality eVect. As expertise increased, that
modality eVect disappeared and eventually, presenting the visual material
alonewas superior to an audiovisual presentation or, indeed, any presentation
that included the text. Guidance provided by textual material, essential to
students on the left of the cognitive matrix of continua, was provided by the
schemas nowavailable after students hadmoved to the right side of thematrix.
Evolution of human cognitive architecture 257Similarly, Kalyuga, Chandler, Tuovinen, and Sweller (2001) found that
the worked example eVect reversed with increased expertise. Novices require
worked examples to provide them with guidance. Schemas, once they have
been acquired, provide guidance, and worked examples become redundant.
Kalyuga, Chandler, and Sweller (2001) and Tuovinen and Sweller (1999),
using novices, found that direct instruction is superior to discovery learning.
That diVerence disappeared if learners with more experience in the domain
were used.
These results can be used to explain other findings. McNamara, Kintsch,
Songer, and Kintsch (1996) found that when learners were presented a
textual passage to read and assimilate, those who were relatively expert in
the area learned more from reduced passages that had segments omitted
than the full passage. Learners with less experience in the area learned most
using the full passage. On the present interpretation, novices required the
full passage to allow understanding and so the full passage condition was
superior. With increased experience, the added material was redundant and
merely served to obscure critical points. Working memory resources were
required to extract those critical points from the surrounding, redundant
material, reducing learning and resulting in the superiority of the reduced
passage.
11. The Guidance Fading EVect
From an instructional perspective, the expertise reversal eVect suggests that
as learners move from the left of the cognitive matrix of continua to the
right, schemas increasingly provide guidance and so the guidance provided
by instruction should be faded out. Unnecessary guidance has negative, not
simply neutral eVects. Renkl and associates (Renkl, 1997; Renkl, Atkinson,
& Maier, 2000) obtained precisely this result using combinations of worked
examples, completion problems, and full problems. Using novices, they
found that guidance provided by worked examples was the best form of
instruction. With increasing expertise, it was desirable for those worked
examples to be replaced with completion problems and, ultimately, with full
unguided problems.
It was indicated earlier that for novices, instruction should replace the
missing central executive but that with increased levels of expertise, schemas
play the role of a central executive. A guidance fading technique accords
closely with this suggestion. Initially, with no central executive available,
worked examples indicate relations between elements of information. As
rudimentary schemas begin to form, they can take over some of the central
executive function from worked examples and so complete worked examples
are no longer necessary. Completion problems can be used as a substitute
258 John Swellerfor worked examples. Once full schemas have been constructed, they can act
as a central executive and so full problem solving with no other guidance
can be instituted. Additional learning through schema automation should
occur during problem-solving practice.
Renkl, Atkinson, Maier, and Staley (2002) found guidance fading as
levels of expertise increase to be demonstrably superior to using a single
instructional procedure. They compared the presentation of conventional
worked examples with guidance fading. The worked example procedure
incorporated the presentation of several pairs consisting of a worked
example followed by a very similar problem to solve. This pairing of a
worked example followed by a problem was used throughout the learning
period, irrespective of changing levels of expertise. Results indicated that the
guidance fading procedure was superior. The superiority of fading over a
single design procedure (e.g., worked examples alone or problems alone) as
expertise increases constitutes the guidance fading eVect.
The guidance fading eVect, along with the expertise reversal eVect,
indicates that individual diVerences, specifically diVerences in levels of
expertise, are a critical consideration when choosing an instructional design.
A design that is ideal for a person located toward the left of the cognitive
matrix of continua may be quite inappropriate for someone further to the
right of the matrix. Ignoring levels of expertise can result in the use of quite
inappropriate instructional procedures.
The instructional designs described in this section diVer from most
instructional designs in that they are very closely tied to our knowledge of
information structures and human cognitive architecture. Indeed, they were
generated directly from that knowledge. It can be argued that they provide a
degree of validity to the cognitive theories discussed. In any scientific area, it
is diYcult or impossible to generate applications from substantially faulty
theories.
IV. Conclusions
Human cognitive architecture has evolved to permit humans to engage in
activities that range from prosaic to awe inspiring. There are logical
structures that determine the way in which cognitive architecture deals with
information. Those logical structures, along with the structure of infor-
mation itself, must have determined the course of the evolution of human
cognitive architecture. The basic information structures that underlie
human cognitive architecture consist of a very large information store with
limitations to ensure that any changes to that store do not destroy its
functionality. The end result is an architecture designed to store immeasurable
Evolution of human cognitive architecture 259amounts of information in a long-term memory but a very limited ability to
deal with novel information in working memory. Information held in long-
term memory guides most of our activities. Novel information in working
memory can feed information into long-term memory and so alter long-term
memory, but the logic of the governing information systems ensures that the
alterations are relatively small to circumvent the unavoidable random
components.
As might be expected, this system logic is universal. It not only applies
to the manner in which human cognitive architecture has evolved, it
applies to the manner in which information is handled by the processes
of evolution themselves. Evolution by natural selection can be characterized
as an eVective and eYcient system for managing and adapting very complex,
natural information to changing circumstances. Human cognitive architec-
ture must also manage complex information. Accordingly, it would not be
surprising if human cognitive architecture evolved to handle information in
the same way as evolution by natural selection. Similarities in the way that
the two systems function suggest that human cognitive architecture, by the
processes of evolution by natural selection, has itself evolved to duplicate
the manner in which evolution by natural selection deals with information.
The logic of these systems places both restrictions on and generates
opportunities for the manner in which information is presented and the
activities in which learners should engage. Our cognitive architecture is
structured with schemas providing an executive function guiding our mental
activities. Instruction is required when those schemas are unavailable and
must be acquired. Ideally, that instruction should provide an executive
function that mimics the missing schemas as closely as possible in order to
avoid random activities and reduce working memory load. Many
instructional procedures that meet these requirements have now been
devised. The successful generation of instructional procedures from
theoretical principles provides a degree of validity for those principles.
While the logic of the information systems discussed in this chapter places
immense barriers to their alteration, their adaptability to new circumstances,
even if slow and frequently ineVective, is their crowning glory. Evolution
may occur over eons but its whole point is change and adaptability, resulting
in the creation of new functions, processes, and entities. Similarly, learning
is the adaptive engine of human cognitive architecture. It may take many
years, especially if creativity is required because instruction from and
imitation of other humans is unavailable, but it is the foundation function of
human cognitive architecture. Only through learning does the ability to
eYciently process high element interactivity material become possible, and
processing high element interactivity material is characteristic of humans.
Prior to learning, such material can be dealt with but only in an unguided,
260 John Swellerpartially random manner with all complex interactions ignored. Further-
more, there is an inevitability about this limitation. There can be no
mechanism to coordinate the very large number of possible combinations
that can occur when dealing with even a relatively small number of elements
that have not been learned. Because knowledge acquired through learning
provides the only coordinating function, it is essential that our cognitive
architecture evolved to ensure that only a limited amount of uncoordinated
information is considered at any given time prior to learning. This limitation
creates an immediate tension when dealing with high element interactivity
information that cannot be limited or reduced in size without compromising
understanding. Because high element interactivity material must be
coordinated, a mechanism for coordinating such information had to evolve
if it was to be processed. Schematic knowledge acquired through learning is
that mechanism. There are very wide or perhaps no limits to the amount of
previously learned information that humans can process.
On this analysis, long-term memory is the source of human intellectual
skill because long-term memory holds learned material. It may be this
structure that took millions of years to evolve, and at least on earth, is
unique to humans in terms of size. Our huge knowledge base is shared
neither by other living creatures nor, to this point, by artificial devices
created by humans. It may only be shared by the mechanisms that permit
life itself to reproduce and evolve. Other cognitive structures, including ones
not considered in this chapter, such as sensory systems, are both ubiquitous
and frequently superior to their human equivalent. In contrast, our immense
long-term memory, with its close connections to learning, has no cognitive
equivalent on earth. That structure is quintessentially human.
Acknowledgments
The author thanks Paul Ayres, Brett Hayes, Paul Ginns, Slava Kalyuga, and Nadine
Marcus for providing comments on earlier versions of this chapter. This work was supported by
grants from the Australian Research Council.
References
Allport, D., Antonis, B., & Reynolds, P. (1972). On the division of attention: A disproof
of the single channel hypothesis. Quarterly Journal of Experimental Psychology, 24,
225–235.
Anderson, J. R., & Lebiere, C. (1998). The atomic components of thought. Hillsdale, NJ:
Erlbaum.
Aparicio, S. (2000). How to count  human genes. Nature Genetics, 25, 129–130