THEORY OF STRUCTURED INTELLIGENCE
Results on Innovation-based and Experience-based Behaviour
Meike Jipp and Essameddin Badreddin
Automation Laboratory, University of Mannheim, B6, 23-29, Mannheim, Germany
Keywords: Intelligence, intelligent interfaces, innovation, human computer interaction.
Abstract: An agreed-upon general theory of intelligence would enable signif
icant scientific progress in all disciplines
doing research on intelligence. Such a theory, namely the theory of structured intelligence is tested by
relating it to other theories in the field and by empirically testing it. The results demonstrate (1) that the
theory of structured intelligence uses a similar concept of intelligence as do other theories but offers greater
scientific insights in the how intelligent behaviour emerges and (2) that its distinction between innovation-
and experience-based solutions can be found in the behaviour of the study’s participants. This yields the
opportunity to (1) allow technically testing intelligence in an easier and less time-consuming ways as do
traditional intelligence tests, and (2) allow technology classifying the intelligence of its user and using
adaptive interfaces reducing the possibility of serious handling errors.
1 INTRODUCTION
Many different theories of intelligence have been
developed in the varying disciplines (for a summary
see Badreddin & Jipp, 2006). So far, these
approaches have been isolated not sufficiently taking
into account results from other disciplines. An
opposite example is the theory of structured
intelligence developed by Badreddin and Jipp
(2006). The authors make use of research results in
neurophysiology, psychology and system theory and
present the theory itself and discuss ways of
technical implementation. The theory defines
intelligence as the ability to solve problems using
limited space and time resources. The concepts of
Innovation, Experience, Fusion, and Learning are
distinguished to explain problem solving behavior.
Innovation reflects the capability to come up with
totally new, unpredictable solutions to the current
problem. Experience refers to using past, known,
and successful solutions for the current problem. If a
problem is faced, two solutions are worked out, one
based on innovation (“new solution”), the other
based on experience (“past solution”). These two
solutions are fused by appropriate algorithms and the
final solution will be applied to the current problem.
This derived solution is saved, so that it is available
the next time the same or a similar problem is faced.
Hence, any combination of new and well-known
solutions to a problem can be developed. Fig. 1
gives an overview over the described structure.
Figure 1: Theory of structured intelligence (see Badreddin
& Jipp, 2006).
2 PROBLEM STATEMENT
The, by Badreddin and Jipp (2006) developed theory
of structured intelligence needs further testing to see
whether it (1) allows the realization of a
qualitatively different concept of artificial
intelligence, (2) allows easier testing of human
intelligence, (3) enables technology to test their
user’s intelligence and adapt interfaces according to
the level and structure of intelligence to avoid
possible errors, which is especially of importance in
safety-critical systems.
Innovation
Ex
p
erience
Solution
Learning
New solution
Past solution
Fusion
Problem
327
Jipp M. and Badreddin E. (2007).
THEORY OF STRUCTURED INTELLIGENCE - Results on Innovation-based and Experience-based Behaviour.
In Proceedings of the Fourth International Conference on Informatics in Control, Automation and Robotics, pages 327-332
DOI: 10.5220/0001619903270332
Copyright
c
SciTePress
3 SOLUTION APPROACH
To test the theory at hand, a way similar to what is
known as construct validation has been taken
(Campbell & Fiske, 1959): First, the concept of
intelligence as described in Section 1 will be set into
relation with other existing theories (see Section
2.1). To empirically test the defined relations and
related hypotheses, a study has been conducted in a
second step, which results are presented in Section
2.2.
3.1 Theoretical Foundations
In order to perform the construct validation, the
theory of structured intelligence must be put in
theoretical relation with another theory which has
proven empirical adequacy. The theory used here is
the theory of skill acquisition, which was developed
by Ackerman (1988) and is based on research
performed by Fitts (1964), Anderson (1982),
Fleishman (1972), as well as Schneider and Shiffrin
(1977). Ackerman (1988) distinguishes three phases
of skill acquisition:
The first phase is characterized by a
relatively strong demand on the cognitive-attentional
system, so that performance is slow and error prone.
Ackerman (1988) explains this phase as the one in
which potential ways for executing the current task
are worked out and (mentally) tested. Attention is
focused on thoroughly understanding the task’s
constraints in question. With consistent practice,
performance gets faster (see Schneider & Shiffrin,
1977) and attentional demands are reduced (see Fisk
& Schneider, 1983).
During the second phase, the applied and
successful ways of executing the task in question are
strengthened and fine-tuned. More efficient ways of
solving the task in question are found.
Finally, performance is fast and accurate.
The task is automated and can be completed without
much attention.
Performance in each of these three phases is
determined by abilities, namely by general
intelligence, perceptual speed ability, and
psychomotor abilities. General intelligence was
defined by Ackerman (1989) in accordance to
Humphreys (1979), as the ability to acquire, store,
retrieve, combine, compare, and use information in
new contexts. Perceptual speed refers to the ability
to complete very easy cognitive tasks. The core
cognitive activity is to generate very simple potential
solutions to effectively solve tasks as quickly as
possible. The key is the speed with which symbols
can be consistently encoded and compared
(Ackerman, 1989). Last, psychomotor abilities
represent individual differences in the speed of
motor responses to problems without information
processing demands.
Ackerman (1988) proposes that general
intelligence determines initial performance on a task
with new information processing demands, i.e. the
first phase of skill acquisition. The influence of
general intelligence diminishes, when potential ways
for the solution have been formulated (for empirical
support, see e.g., Ackerman, 1988). The learner
proceeds to the second phase of the skill acquisition
process, when an adequate cognitive representation
of the task has been built. Then, performance
depends more on psychosensoric abilities. It is
required to fine-tune and compile the determined
solutions, which equals the definition of the abilities
underlying psychosensoric abilities. Sequences of
cognitive and motor processes get integrated, ways
of solution adapted for successful task performance.
With further practice the impact of psychosensoric
abilities on performance decreases and psychomotor
abilities play a more important role. In this third
phase of the skill acquisition process, the skill has
been automated, so that performance is only limited
by psychomotor speed and accuracy (Ackerman,
1988).
Fig. 2 summarizes the described relationship
between skill acquisition and ability-performance
correlations.
Figure 2: Theory of Skill Acquisition (adapted from
Ackerman, 1989).
Taken into account the above described theory of
structured intelligence it is to be assumed that in the
first phase of skill acquisition, in which the learner is
confronted the first time with a new problem,
innovation processes affect the derived solution. In
contrast, the solution to a well-known problem will
be chosen based on the memory traces of the already
successfully applied solution alternatives. Hence, in
this case, behaviour is experience-based.
1st phase
2nd phase
3nd phase
number of practice
ability-
performance
correlation
ICINCO 2007 - International Conference on Informatics in Control, Automation and Robotics
328
3.2 Conducted Study and Relevant
Research Results
3.2.1 Research Questions and Hypotheses
The, in the previous section described theory of skill
acquisition has been tested by various researchers
(see e.g., Jipp, Pott, Wagner, Badreddin, &
Wittmann, 2004) and proved its empirical adequacy
in various settings and circumstances. The
relationship to the theory of structured intelligence
has been established theoretically in Section 2.1 and
will be tested empirically. Therefore, the following
research questions and hypotheses are of interest:
It is hypothesized that the degree of
consistency with which given tasks are tackled
decreases with the familiarity of the task. This is the
case as formerly applied successful solutions will be
used to solve the problem at hand. Variation does
only occur when the former solution has not lead to
a satisfying result.
It is hypothesized that the transition from
innovation-related processes to experience-based
solutions is determined by intelligence factors as
measured with traditional measurement scales for
intelligence factors. This is based on Ackerman’s
theory (1989) combining skill acquisition processes
with factors of individual differences, i.e., general
intelligence, perceptual ability and motor skills. The
last is not considered in this paper due to the focus
on structured intelligence.
3.2.2 Description of the Sample and Course
of the Study
To be able to answer these research questions, a
study has been conducted at the vocational school of
the Protestant Foundation Volmarstein
(Evangelische Stiftung Volmarstein, Germany).
Data from 13 students (6 male, 7 female students)
was at hand for the present analyses. The students
were wheelchair-users and have been disabled for
more than 12-15 years. Their average age was 22.5
years (SD = 1.6 years). The disabilities of the
participants were spasticity, spina bifida, dysmelia
or incomplete paralysis.
The study was conducted within two sessions.
The first session lasted between one and two hours
depending on the speed with which the participants
performed the designated tasks (see also Jipp,
Bartolein, & Badreddin, 2007). The tasks the
participants conducted referred to leading a little
garden market. More specifically, the participants
had to prepare products potential customers
requested. These customer wishes were sorted in
two categories: sowing seeds (either sunflower or
ramson seeds) and setting in seedlings (either
flowering or foliage plants). The following actions
were required in order to sow the seeds:
The pots had to be placed in a seed box.
The pots had to be filled with loosened
soil.
A hole had to be made into the soil.
One seed had to be put in each hole.
If the seeds were light seeds (as indicated on
the customer wish), the holes had to be
covered with wet pieces of newspaper.
If the seeds were dark seeds (as indicated on
the customer wish), the holes had to be
covered with a 0.5 cm level of soil.
The pots had to be watered. The water had to
be prepared so that it had a temperature of
25°C and a, in the instructions specified acid
value.
For setting in the seedlings, the following actions
had to be performed by the participants:
The required pots had to be filled half with
soil, which had to be loosened before.
The seedlings had to be put into the pot.
The correct fertilizer had to be chosen (as
indicated on the instructions, which were
handed out to the participant).
The pot had to be filled with layers of soil
and fertilizer until the roots of the seedlings
were covered.
The seedling had to be watered with
appropriate water (25°C and an acid value of
5-6).
In order to acquire the task of leading the market
garden, four customer requirements had to be
executed: the first required the participants to sow
sunflower seeds, the second to set in flowering
seedlings, the third to set in foliage plants and the
last one to sow ramson seeds. The two categories of
tasks have been defined based on only minor
differences in order to allow the participants to
acquire the skill in question. Further customer
wishes could not be executed by the participants due
to problems related to maintaining attention for such
a long time frame. The actions of the participants
were filmed with a standard web camera.
In the second session of the study, the
participants performed tasks of the Berlin
Intelligence Structure Test (BIS, Jäger, Süß &
Beauducel, 1997). These tasks were based on the
Berlin Intelligence Structure Model (Jäger, 1982),
which is a hierarchical model of intelligence.
General intelligence, at the top, is composed of two
facets, which are categories for factors at the next
THEORY OF STRUCTURED INTELLIGENCE - Results on Innovation-based and Experience-based Behaviour
329
lower level (Guttman, 1954). Jäger (1982)
distinguished the facet operations and contents. The
last subsumes three content abilities (i.e., numerical
abilities, verbal abilities, and numerical abilities),
which refer to how a person cognitively deals with
the different types of contents. The facet operation
subsumes what is cognitively done with the given
contents. Four operations are distinguished:
Reasoning is the ability to solve complex problems
(Jäger, Süß, and Beauducel, 1997). Memory asks the
participants to memorize pieces of information and
retrieve them from short-term memory or recognize
them after a short time period. Creativity refers to
the ability to produce a variety of differing ideas
controlled by a given item. Last, perceptual speed is
the ability to work as fast as possible on simple
tasks, requiring no or only little information
processing demands. The BIS tests all these factors
of intelligence. However, the original test has been
shortened. Test items were deleted which required
the participants to write a lot, as – due to the given
time constraints for working on the test items –
especially participants with spacticity would have
been disadvantaged. The conducted test comprised
the tasks as indicated in Table 1 and took the
participants about two hours to complete.
Table 1: The, from the BIS chosen and in this study
applied test items and their sorting in the factors of
intelligence according to the Berlin Intelligence Structure
Model.
General
Intell-
igence
Figural
abilities
Verbal
abilities
Numerical
abilities
Perce-
ptual
speed
- Erasing
letters
- Old English
- Number
Symbol Test
- Part-
Whole
- Classi-
fying words
- Incomplete
records
- X-Greater
- Calcul-
ating
characters
Memory - Test of
orientation
- Company’s
symbols
- Remem-
bering routes
- Mean-
ingful text
- Remem-
bering
words
- Language
of fantasy
- Pairs of
numbers
- Two-digit
numbers
Reason-
ing
- Analogies
- Charkow
- Bongard
- Winding
- Word
analogies
- Fact
opinion
- Comparing
conclusions
- Reading
tables
- Arithmetic
thinking
- Arrays of
letters
3.2.3 Data Analyses and Results
The following variables were derived:
general intelligence, perceptual speed,
reasoning, memory, figural abilities, verbal abilities,
numerical abilities, figural perceptual speed, verbal
perceptual speed, numerical perceptual speed,
figural reasoning, verbal reasoning, numerical
reasoning, figural memory, verbal memory,
numerical memory were derived based on the
reduced set of test items applied from the BIS
number of strategic changes in the
participants’ behavior for each of the four customer
wishes
index for the continuity of the order of
actions while performing each of the four customer
wishes
In order to derive the numerical values for the
intelligence factors, the test items were analyzed as
indicated in the BIS’s handbook (Jäger, Süß, and
Beauducel, 1997).
Altogether eight variables have been used to
operationalize the degree of the innovation in the
observable behavior (i.e. in this case the gardening
tasks): the first is the number of strategic changes in
the behavior of the participants for each of the
customer wishes. For this purpose, the videos have
been transliterated: A list of possible actions has
been identified and the order of actions conducted
has been analyzed. Each participant used a typical
order of how to perform the task in question. The
number of changes to this typical order has been
counted as indicating the number of strategic
changes in the behavior.
To derive the four indices for the continuity of
the order of actions while performing the customer
wishes, the number of grouped actions was counted.
A participant conducting all required actions for one
pot received a very low index of continuity (i.e. 1);
whereas a participant executing one action for all
pots received a high level of continuity (i.e., 10).
Participants with a medium-sized index changed
their strategy within the task. In order to be able to
distinguish the participants who did not change their
strategy and did change their strategy, a
dichotomization was performed – the medium-sized
numbers received the final index number 0, the high-
and low-sized numbers received the final index
number 1.
In order to test the hypotheses, repeated
measurement analyses have been performed with
each of the intelligent variables derived (list see
above) as independent variables and either the
number of strategic changes for all four customer
wishes or the index of continuity for all four
ICINCO 2007 - International Conference on Informatics in Control, Automation and Robotics
330
customer wishes as dependent variables. The
significant results are given in Table 2. More
specifically, the tested variables, the value of the
used test statistic with the number of degrees of
freedom, the probability that the expected effect did
not occur due to chance and the size of the detected
effect using a classification presented by Cohen
(1992) are given.
Table 2: Results of the repeated measurement analyses.
Value of
the used
test
statistics
Tested variables Prob-
ability
Effect
size
1) Learning factor
of the number of
strategic changes
F(3,36)
= 5.80
p = 0.00 f² =
0.49
1
2) a) Learning
factor of the number
of strategic changes
b) Two way
interaction between
the learning factor
and the verbal
perceptual speed
factor
F(3, 30)
= 4.09
F(3, 30)
= 2.75
p = 0.02
p = 0.06
f² =
0.41
1
f² =
0.28²
3) a) Learning
factor of the number
of strategic changes
b) Two way
interaction between
the learning factor
and the verbal
memory factor
F (3, 30)
= 4.20
F (3, 30)
= 3.31
p = 0.01
p = 0.03
f² =
0.42
1
f² =
0.33²
4) a) Learning
factor of the index
of continuity
b) Two way
interaction between
the learning factor
and the numerical
memory factor
F (3, 30)
= 4.34
F (3, 30)
= 4.66
p = 0.01
p = 0.01
f² =
0.43
1
f² =
0.47
1
*significant with α < 0.01
1
= large effect, ² = medium-sized effect
As Table 2 indicates, the first analysis testing the
learning factor of the number of strategic changes
over the performed tasks is significant. A large
effect has been found: The number of strategic
changes shrinks with the number of customer wishes
performed.
The second analysis tested the learning factor
and the two way interaction effect between the
learning factor and the verbal perceptual speed
factor. Figure 3 shows this interaction. The
participants with greater verbal perceptual speed
abilities demonstrate more continuity regarding the
strategic changes. The graph showing the course of
the number of strategic changes while performing
the four customer wishes for the participants with
less verbal perceptual speed abilities demonstrates
(1) that the general level of the number of strategic
changes is greater compared to the number of
strategic changes executed by the participants with
higher verbal perceptual speed abilities and (2) that
the variance of change is bigger.
Figure 3: Line graph with standard error bars of the
relationship between the strategic changes for the four
customer wishes and the verbal perceptual speed of the
participants (Graph 0 shows the less intelligent
participants, Graph 1 the more intelligent participants).
The third significant analysis tested the learning
factor and the two way interaction effect between
the learning factor and the verbal memory factor.
Figure 4 shows the direction of the effect. The two
graphs comparing the course of the number of the
strategic changes while performing the four
customer wishes for the two groups of participants
with high and low scores on the verbal memory
factor closely resemble the graphs displayed before
(see Fig. 3). Again, the less able participants
demonstrate (1) a greater number of strategic
changes and (2) a bigger change in the course of the
skill acquisition/learning process.
Figure 4: Line graph with standard error bars of the
relationship between the strategic changes performed for
four customer wishes and the verbal memory factor of the
participants (Graph 0 shows the less intelligent
participants, Graph 1 the more intelligent participants).
THEORY OF STRUCTURED INTELLIGENCE - Results on Innovation-based and Experience-based Behaviour
331
The fourth analyses tested (1) the learning factor
of the index of continuity and (2) the two way
interaction between the learning factor and the
numerical memory factor. The results were
significant as well. Hence, the index of continuity
changes with the number of practice trials performed
and this change depends on the level of numerical
memory abilities of the participants.
4 CONCLUSIONS AND
TECHNICAL IMPACT
Summarizing, relevant research results regarding the
theory of structured intelligence are presented
(Badreddin & Jipp, 2006). More specifically, the
study confirmed the hypotheses: First, the degree of
consistency in the behaviour of the participants gets
greater with the task’s familiarity. This transition
shows the learning effect, i.e. the transition from
innovation-based to experience-based, i.e., well
known solutions. Second, a relationship between this
transition and traditional measures of intelligence
has been found. Significant predictors were
intelligent factors such as the verbal perceptual
speed factor. However, compared to traditional
research on intelligence, the theory of structured
intelligence goes one step further: it provides an
explanation of how intelligent behaviour emerges
and not only a classification of intelligent behavior.
Drawbacks of the study refer to the small number of
participants, which has reduced the power of the
study, so that some possibly existing effects might
not have been detected. Further, the creativity items
had to be deleted from the intelligence test as they
required the participants to draw solutions, which
would disadvantage some of the participants due to
their disability. Hence, future research should
investigate the relationship between traditional
creativity tests with innovation-based behaviour.
The study’s main contribution is twofold: First,
the theory of structured intelligence demonstrates
links to traditional measurements of intelligence, but
also gives an explanation of how intelligent
behaviour emerges and provides the opportunity to
measure intelligence in easier and less time-
consuming ways and. It allows intelligence to be
judged on (1) by using activity detection and (2) by
observing the participants’ actions when being
aware of the familiarity of the task. The degree of
consistency gives valuable information on
intelligence. This has not only the potential to
revolutionize intelligence diagnostics but also
intelligent interface design: if an intelligent machine
were capable of judging on its user’s intelligence,
the interface can be adapted to the user and different
levels of support given. This might have a big
impact on safety-critical applications with the user is
in the loop of controlling e.g., nuclear power plants.
Second, the theory of intelligence gives the artificial
intelligence research a new direction, as not only
experience is relevant, but also innovation-driven
behaviour, which can be modelled as chaotic
behaviour (see also Badreddin & Jipp, 2006).
REFERENCES
Ackerman, P. L., 1988. Determinants of individual
differences during skill acquisition: Cognitive abilities
and information processing. Journal of Experimental
Psychology: General, 117 (3), pp. 288-318.
Anderson, J. R., 1982. Acquisition of cognitive skill.
Psychological Review, 89, pp. 369-406.
Badreddin, E., Jipp, M., 2006. Structured Intelligence.
International Conference on Computational
Intelligence for Modelling, Control, and Automation
CIMCA 2006, Sydney, Australia.
Campbell, D. T., Fiske, D. W., 1959. Convergent and
discriminant validity by the multitrait-multimethod
matrix, Psychological Bulletin, 56, pp. 81-105.
Fisk, A. D., Schneider, W., 1983. Category and word
search: Generalizing search principles to complex
processing. Journal of Experimental Psychology:
Learning, Memory & Cognition, 9, pp. 177-195.
Fleisman, E. A., 1972. On the relations between abilities,
learning, and human performance. American
Psychologist. 27, 1017-1023.
Guttman, L. (ed.), 1954. Mathematical thinking in the
social sciences. Glencoe, IL. The French Press.
Humphreys, L. G., 1979. The construct of general
intelligence. Intelligence, 3, 105-120.
Jäger, A. O., 1982. Mehrmodale Klassifikation von
Intelligenzleistungen: Experimentell kontrollierte
Weiterentwicklung eines deskriptiven
Intelligenzstrukturmodells, Diagnostica, 28(3), 195-
225.
Jäger, A. O., Süß, H.-M., Beauducel, A., 1997. Berliner
Intelligenzstruktur-Test, Göttingen, Hogrefe.
Jipp, M, Bartolein, C., Badreddin, E., 2007. Assisted
Wheelchair Control: Theoretical Advancements,
Empirical Results and Technical Implementation,
Submitted for publication.
Jipp, M., Pott, P., Wagner, A., Badreddin, E., Wittmann,
W. W., 2004. Skill acquisition process of a robot-
based and a traditional spine surgery, 1
st
International
Conference on Informatics in Control, Automation and
Robotics, pp. 56-63.
Schneider, W., Shiffrin, R. M., 1977. Controlled and
automatic human information processing: 1.
Detection, search, and attention. Psychological
Review, 84, pp. 1-66.
ICINCO 2007 - International Conference on Informatics in Control, Automation and Robotics
332