Are you experienced? -
a model of learning systems thinking skills
Martin Schaffernicht
fac. de Ciencias Empresariales - University of Talca
Avenida Lircay s/n - Talca - Chile
Phone: +56 71 200 353
Fax: +56 71 200 358
E-mail: martin@utalcad
Abstract
This paper proposes a contribution to the domain of systems thinking skills. Empirical
studies have repeatedly shown surprising misperceptions and inabilities in subjects
confronted with tasks involving very simple stock and flow systems. Here it is proposed to
represent these skills as implicit integration, by which Polanyi modeled our ability to know.
In this framework, Dreyfus and Dreyfus’ five stage model of learning is used to construct
three hypotheses concerning the learning of systems thinking and its importance for
learning from modeling and interaction with models. The tests elaborated by Ossimitz are
adapted for this purpose and some tasks are added, to serve in the experinental
corroboration of the hypotheses. Since the enpirical work is currently under way, only
few results can be presented; consequently the main contribution is the conceptual
construction of the hypotheses.
Keywords: system thinking skills, tacit knowledge, leaming, expertise
Introduction
System Dynamics was developed in order to help decision makers design better policies,
based upon an improved understanding of how structure generates behavior. From the
outset, it was thought that people's understanding of the system - called mental model - is
at the same time the richest source for modeling and meant to change (improve) as a result
of the modeling (Forrester, 1961).
Later on it was also tried to foster leaming by interacting with simulation models or games
rather than modeling, because modeling is more expensive than using a model (Maier and
Gréssler, 2000). Thus the connection between SD and making-leam became stronger.
Also, SD is presented as possible element of inquiry-leaming cycles (Sterman, 2000).
However, the usual SD publications do not focus on the larger inquiry cycles but rather
concentrate on the leaming going on during a modeling effort, like for example inside the
group model building area (Vennix and Rouwette, 2002).
SD helps in several ways. It provides a language with concepts and a representational
system which guide their user to look at situation in terms of stocks and flows, feedback
loops, delays, nonlinearities and borders (see Ossimitz, 2000): it is a language for systems
thinking. It also allows articulate models to be simulated, in order to validate them before
using them to fundament decisions: it is a method for coming to understand a system in a
disciplined manner.
Sadly, many studies indicate that adult humans are not intuitive systems thinkers. It has
been shown that people misperceive feedback (Sterman, 1989; Moxnes, 2000; Jensen and.
Brehmer, 2003; Doyle et al., 1998) and other SD building like those involved in the bathtub
dynamics (Sweeney and Sterman, 2000; Ossimitz, 2002) and especially distinguishing
stocks and flows (Ossimitz, 2002; Kainz and Ossimitz, 2002). However, SD helps
evolving metal models (Doyle et al. 1998) and system thinking leads to better control
performance (Maani and Maharaj, 2004). Better systems thinking skills would probably
help to be a better systems dynamicist as much as modeling skills are needed.
It has to be suspected that the understanding of one system - a conscious mental model -
not the same kind of knowledge as system thinking, which is a skill; accordingly, leaming
about one system will not be the same as leaming to think systemically either. This article
assumes that the two mentioned cases of leaming are different but strongly connected, as
suggested by the theory of logical types of leaming (Bateson, 1979). Seen in this light, the
studies mentioned above and briefly presented in section 2 investigate different aspects of
one underlying theme. The justification for this assumption is given in appendix 1.
Section 3 argues that system thinking skills are well captured by Polanyi’s theory of
implicit integration or tacit knowing (Polanyi, 1966; Neuberg, 1999). According to this
theory the perception of something we recognize in the world is only possible due to the
unconscious integration of uncountable sensory stimuli, which is why the things we
recognize are always already there, and any conscious knowledge is based on this implicit
knowing. This has consequences for how the transformation of a beginner into an
experienced systems thinker is represented. Section 4 introduces a five stage model of
leaming based on Dreyfus (1986), which has been brought into connection with implicit
integration (Neubery, 1999) is presented to conceptually model this progressive
transformation.
Section 5 looks at the different studies through the lens of criteria derived from these
conceptual models. It appears that the systems thinking tests should be adapted in order to
assess the stage-wise construction of system thinking skills. Empirical work in the domain
of the first issue is just under way and first elements are presented in appendix 2.
At asecondary level, the influence of these skills on the modeling-for-leaming process may
become measurable if studies that investigate mental models take into account the stage of
their subjects. Third, the fact that in spite of the importance of modeling for leaming, none
of the studies tested the leaming effects of modeling, calls our attention.
Recent empirical studies concerning systems thinking and learning
The concem for the quality of perception of systemic structures and their mental use is not
new (for example Sterman, 1989); at the same time, the connection between system
dynamics and leaming has been a subject ever since “Industrial Dynamics” (Forrester,
1961; also Morecroft and Sterman, 1994). This paper tries to tie together the topics of
“leaming in SD activities” and “systems thinking”; in recent years, several relevant studies
have been published en the SDR or the SD conference, but they seem to concentrate either
on “systems thinking” (specially perceiving) or on “leaming”.
Doyle et al. (1998) ask if working with a simulation of a complex system helps to bring
about changes in the subjects’ mental models. Their case is the Kondratiev economic
cycle, built into the Stratagem-2 simulator. Their subjects were undergraduate students
without prior exposure to systemic education, who interacted with the simulator during five
sessions of one hour each, spread over two weeks. Although they could work together
during simulation, the raw material was collected individually. The raw materials were
written statements produced by subjects at the beginning and the end of the session. These
statements were analyzed as mental models, looking for changes en detail (elements, links)
and dynamic complexity (basically feedback loops). This study found subjects to develop
more complex mental models (larger models with more feedback loops) but without
challenging or changing the model boundary.
Moxnes (2004) investigated if subjects could use the interaction with a simulator to leam
not do over-exploit natural resources in the case of reindeer range management. His
subjects were graduate students initiating SD studies who had to realize 3 ties with a
simulator in the least time possible; they had to work alone. The raw material was their
behavior and performance, and mental models or rules were inferred in order to explain the
behavior. He finds that result feedback apparently did not lead to go beyond linear
thinking, maybe because of the non-linearity involved.
Jensen and Brehmer (2003) asked whether subjects who have to manage a predator-prey
population would leam to stabilize the populations in the case of a simulation of foxes and
rabbits (a model with feedback loops and non-linearity). Subjects were undergraduate
students who could interact with the simulator during one hour (the article does not
mention clearly if they worked alone or in teams). The raw material was behavior and
performance, but also verbalized reasoning (thinking aloud). About half of the subjects
succeeded, but many felt they could use more mathematics skills. The successful groups’
reasoning went from closed-loop control to open-loop control. They found that especially
higher system thinking activities had a positive influence on performance, and groups that
asked for the system’ s structure before making their plans did achieve higher outcomes.
Sweeney and Sterman (2000) wanted to test stock-flow thinking: would their subjects be
able to come to grips with a task that demands relating the net flow into a stock with the
stock’s changes? The case was a bathtub and a bank account. The subjects were graduate
Management students beginning SD education. They ware asked to respond to apparently
simple questions in an apparently simple format that would not demand higher mathematics
skills. The result of the study was that many subjects had substantial difficulties with
performing the mental/graphical operations required to properly relate flows and stocks.
Ossimitz (2002) took up the work of Sweeney and Sterman and elaborated a set of tests that
would help to distinguish stock-flow thinking problems from issues related with the
representation. His subjects were undergraduate students who performed even worse than
Sweeney and Sterman’s. Ossimitz concludes that a “fundamental aspect is the ability to
grasp that in a stock-flow-context the stock with one inflow and one outflow is increasing
when the inflow is bigger than the outflow (or the net flow is positive, to put it in another
way). Some findings indicate that this might be a key criterion for discriminating between a
stock-flow-thinker and a non-stock-flow-thinker’. In a following investigation, Kainz and
Ossimitz (2002) compared subjects’ performance before and after a 90 minute crash-course
on stocks and flows, finding a notable improvement.
Maani and Maharaj (2004) tried to find out about if and how systems thinking affects
complex decision making; they used a case of an imaginary “computech” firm where
subjects have to strive for revenue, profit and market share. Subjects were graduate
students in a business school, and had been previously gone through systems thinking
education. They could work in teams with a simulator during 2 hours. scmy mate
was their behavior and performance and the verbalized reasoning. The material was coded
following a systems thinking level scheme and a task understanding scheme (which is
based on the “correct” understanding of the underlying model).
Learning to think and act systemically
The empirical investigations suggest that few adults are equipped with these thinking skills
without having passed through a SD or other systemic education: after all, SD was first
proposed as a remedy for the lack of these skills. Also, someone who received this kind of
education should have less of a hard time in becoming aware of the SD building blocks
when confronted with a system or situation.
This means that there is a progression from beginner to expert, and the beginner uses a
different kind of knowledge as the expert. However, it is not clear what the domain of
expertise is: does one become an expert systems perceiver (implicit systems thinker) or an
expert systems modeler (explicit systems thinker), or both? As stated above, a better
systems thinker will have more chances to elaborate a helpful model. However, even the
most expert modeler will have to articulate his understanding and undergo formal
quantification and validation. So it is assumed as prudent here to think that becoming a
system dynamicist is becoming an expert modeler, who masters the modeling skills that
prove helpful for elaborating explicit knowledge of a system, and at the same time
becoming an expert systems thinker (perceiver). However, the two bodies of knowledge
are different (for a brief review, see appendix 1).
We now have to focus on what changes as one transforms oneself into an expert. In order
not to duplicate the presented argument, the following discourse is limited to becoming an
expert modeler. Does the expert modeler know more (of the same type of knowledge) or
differently? Cognitive psychology has been used in SD in order to reflect on the question
of knowledge and leaming and to fundament the interest for and research on mental models
and their change due to SD activities (Doyle and Ford, 1997). This orientation has
generated. studies about the change in mental models - by definition aware - more than
asking for skills. In this paper, it is assumed that indeed skills are a different kind of
knowledge that usually will not appear in mental models: the expert is thought to know
differently.
Polanyi’s model of implicit integration
Polanyi’s work is proposed to devise an improved model for how a person comes to be
experienced and judge and act intuitively. Polanyi elaborated a model of perceiving and
acting intelligently based on Gestalt psychology. The basic assumptions are that there is a
subject in a world. The extemal world does exist, and the subject has but his or her own
body in order to know this world. So in a way, I know the word through the changes that
my encounters with it trigger in my body: electromagnetic waves hit my eyes, the rods and
cones trigger a chain of nervous activities up to the point where “I” see an old friend. I
know ny friend through all of these inner processes. My friend is a distal term, and the
parts of my body (nervous system included) that worked due to the light waves hitting the
retina constitute a proximal term. We perceive the world attending from proximal terms to
distal ones. And our knowledge becomes apparent in these terms: we can consciously
know grace to our implicit knowing.
Our focal awareness (of the distal term) is directed on what we perceive, and how we
perceive (the proximal term) is not in our awareness, it remains in a subsidiary awareness.
We may direct our focal awareness towards our inner processes, but then they cease to be
proximal and become distal. For example, a musician performing a Paganini piece is
supposed to direct his attention towards the music, not his fingers, and if he redirects his
attention to them, his performance will not be a good one.
Naturally, there are many sensory stimuli arriving at each moment, and the subject (each
animal) has to steer himself (itself), that is: his movements, in an ever changing and moving
world. So there is a need for integration, in order to attend to the most important items.
This is done for us by the nervous system without the need for or even the possibility of
conscious control. So it is that when I suddenly am aware of my friend's presence, he is
already there; I do not need to transform the sensory stimuli into this conclusion
consciously, the integration is implicit from the conscious subject’s point of view.
Interestingly, I could not choose not to recognize my friend. It has been experimentally
shown that optical illusions persist even though a subject deliberately tries not to fall victim
of them (Neuweg, 1999: 170). The subject can choose to doubt his perception, but then the
perception - the implicit integration- has already been produced. If for example, one puts
on eyeglasses that invert the optical image, one can leam to perform usual activities (like
car-driving) within a couple of days; however, this does not mean that the eyes invert the
inverted image, but that the rest of the nervous system adapts so that the other relevant parts
Operate together with the inverted image: it adapts its way to integrate. It was not reflection
that brought about the re-integration, but pragmatic attempts to keep on doing usual things
of life. Remarkably, subjects tend to forget that they see everything in a inverted manner
unless they are asked: the question changes the attending through your eyes to attending to
them.
This model seems well harmonic with recent findings about the brain and its self (Llinas,
2004), that show how the brain organizes complexes of perception (and also of action) in
so-called fixed action pattems in order to reduce computational overhead. Llinas believes
that the brain’s function is to help the organism move (act) successfully in a ever more
dynamic word by making accurate predictions, and that intelligence comes from having to
move your body in the world (reflection is but a tool, not an end in itself).
As the example of the inverted image shows, the implicit integration can be leamed, and
certainly has to be leamed from the earliest age on. This goes on all the time. According to
Polanyi, our awareness is not only directed outwards in each moment, it also moves on
outwards as we leam: as a beginner, a car driver will focus on, say, not directing his car off
the road and not becoming too fast or slow; he feels the texture of the steering wheel on his
hands: he attends from his hands to the steering wheel. The experienced car driver will feel
the road's texture through the car, to him the car is like a part of his body. The same
happens with writing on paper with a pencil (you can feel the texture of the paper) and
when you walk down the street, you experience your feet hitting the road (and not your
socks). We literally incorporate our tools and other entities we encounter in the world.
At the side of perception, we develop connoisseurship - a capability to intuitively (or
implicitly) recognize pattems or situations that require the integration of many elements
from the sensory stimuli to the corresponding neural centers, part of which cannot be
precisely stated or related to each other. Examples are judging the quality of vine or
perfume.
At the side of action, we develop skills - the capability to perform complexes of action that
require a great deal of implicit integration from the respective neuronal centers to the motor
units. One possible example is language, even though it has to be reminded that language
also requires comnoisseurship.
The use of Polanyi’s work for System Dynamics
An expert SD practitioner will probably be able to “see” feedback loops and other building
blocks in a way similar to the inverted-image case: looking at a case like the bathtub
(Seewey and Sterman, 2002, Ossimitz, 2002) or the lichen/grazing case (Moxnes, 2004), he
will intuitively see systemic structures: he is a connoisseur. He will also intuitively know
what to do in order to come to grips with the situation, knowing when to rely on qualitative
modeling and when to simulate (and how to): he has the skills, he knows the systemic
language with all its symbols (Ossimitz) and meanings.
The individuals typically tested in experimental situations (Seewey and Sterman, Doyle et
al., 1998; Moxnes, 2000, Maani and Maharaj, 2004; 2004; Jensen and Brehmer, 2003,
Ossimitz, 2002) seem to be rather like starters up in systemic terms: they do not have the
comnoisseur’s seeing capability, nor the systemic representation skills. They may have
gone through mathematics and sciences education at different levels, but apparently there
are some thinks to be leamed (in order to become a connoisseur) that are not part of this
education. If you have leamed German as your mother language, you do not distinguish
the separate words your Danish neighbor pronounces, unless you take a Danish course. It
becomes understandable how a person who manages his bathtub and bank account
(Sweeney and Sterman, 2000) does not manage to perform a task that seems to be the same
one in more abstract tenms: if managing the bathtub is a skillful activity and the capabilities
are proximal (implicit), but the bathtub experiment brings the skill into focal consciousness
as distal term, then the conclusion can only be that human adults are not skillful
connoisseurs of systemic-dynamic situations unless they become one by leaming.
So the question faced by SD and other systemic practitioners is: how can we help them
becoming connoisseurs and skillful?
Following Polanyi, we have to leam the proximal term, the elements and subsidiaries that
have to do the implicit integration as act of subsidiary awareness. This can be done in
different ways:
e the leamer can focus on the distal term, leaving the elements needed to produce the
integration in the implicit, do not become articulate;
e the leamer can focus on the proximal term, trying to make explicit the subsidiary
elements and their way of integration.
If the leamer can draw on previous personal experience, it can be useful to use explicit
leaming, but always in combination with activities that use implicit leaming. Especially, it
becomes important to have a master whom to observe and imitate. This allows the leamer
to dwell inside the master’s mind, thereby producing in him the needed implicit integration.
The leamer will have to trust his master, and be prepared to execute assigned tasks even
though he will not be able to understand them: only by doing so will the have the personal
experience required for focusing on the proximal term.
Leaming should iterate between the implicit and the explicit mode: the proximal term is a-
critical (for un-articulate) and many times, focusing on it me help to improve the process.
Afterwards, the improved elements have to be driven back into the realm of the implicit,
where it operates as proximal term. This process is like a spiral:
a elaborate whole
implicit
reintegration
a diffuse whole aN
analysis
(extraction of
elements)
articulate whole
explicit
integration
—— elements
Fig. 1 - learning as a spiral of integrations (translated from Neuweg, 1999:255)
If implicitly perceived situations stay a diffuse whole, but may be analyzed into elements.
These can become explicitly integrated into an articulate whole, which is like the revealed.
mental model. Later, this knowledge may become re-integrated into the implicit realm.
Dreyfus and Dreyfus
While Polanyi’s model allows us to understand the meaning of the empirical findings in a
way that points at a specific need to foster connoisseurship and skill in a way that honors
the need for implicit leaming in combination with analysis, it does not show how the
leamer progressively transforms himself into an experienced or expert modeler. As system
dynamicists, we may think of the learning as a delay: the inflow are beginners, the outflow
skilled connoisseurs. The length of the delay depends of didactical and personal factors,
but these being equal amongst leamers, it is a pipeline delay: it takes time to transform
oneself from a beginner into an expert. The question rises if this highly aggregate view is
appropriate? Does one tum from beginner into expert in one single step?
Dreyfus and Dreyfus (1986) model of tuming from a beginner into an expert suggests that
there are indeed several phases one runs through. Since each of the phases is characterized
by different and identifiable aspects of the leamer, this model can be useful to design a
stepwise leaming approach for systems thinking and understanding.
Dreyfus and Dreyfus developed a model for the training of aircraft pilots. It tries to make
Clear how intuitive judgment and action - knowing and acting without previous conscious
deliberation- can be developed out of a beginner state where the leamer cannot do much but
follow previously defined rules. The model has the following phases:
Beginner
The beginner does not have personal experience in the field and depends on the availability
of general (context-free) attributes and rules attached to these attributes. His attention will
be absorbed by the analytic search for the attributes and the execution of the rules.
The typical leaming activities are instruction, presentation, simple and reduced exercises.
The beginner modeler will typically study a textbook, say “Business Dynamics” (Sterman,
2000) intemalize what the SD building blocks are (feedback loops, stocks and flows,
delays, nonlinearities and boundaries), and strive to apply the mules extracted from lecture
or/and example models in order to do the exercises or resolve the challenges.
Advanced beginner
As he comes to resolve more situations, the leamer elaborates a growing set of known
situations and comparison between them starts to crystallize situation-specific and holistic
aspects (rather than formally defined, partial and context-free attributes). These aspects are
connected with directives (rather than rules).
The typical leaming activities are aimed at fostering reflection on the similarities between.
situations.
The advanced beginner modeler will now try to attack validation tasks, which will demand
some interaction with a more experienced modeler or teaching personnel. He may be
interested in “best practices” material which helps him to build up his directives.
Competent
Now the leamer deals with many details (attributes and aspects) and he starts to develop his
own perspective which allows him to order and set priorities and weights; he starts to set
goals and plan ahead, taking into account the situational context and also developing an
emotional involvement with his knowledge (since it is not based on extemally defined rules
and directives).
Typical leaming activities are mainly simple case studies.
The competent modeler will work on small modeling projects in order to resolve a given
case, maybe as part of the book’s challenges.
Proficient competent
Now the leamer has come to perceive the situation as a whole and in an intuitive (implicit)
manner. He does not need to take it apart into attributes in order to know “what the case
is’. However, since there are more possible action strategies than situations, he still has to
consciously reflect upon what to do.
Typical leaming activities include more complex case studies and participation in real work
situations.
The proficient competent modeler will resolve real modeling tasks and even if he keeps the
book like a reference manual, he will only use it to recall how some things are done. He is
able to “see” feedback loops and the like -maybe even archetypes- in an effortless manner.
Expert
Finally, the leamer has leamed to intuitively know what he has to do: “an expert does not
have to reflect; he knows” (Frank Lloyd Wright).
The expert modeler may be looked for by some consulting company, since he is able to
model productively. He knows when to model qualitatively and when to use quantitative
modeling in the situation he meets.
The following table (adapted from Neuberg, 1999, p. 311) summarizes the phases:
Beginner Advanced Competent | Proficient Expert
beginner
Elements context-free | context-free | context-free | context-free | context-free
taken into and and and and
account situational | situational situational situational
Sense for| no no consciously | immediate | immediate
what is elaborated | (implicit)
Perception | analytical analytical analytical holistic holistic
asawhole
Selection of | following interpreting | extensive limited intuitive
action tules mules and | planning planning
directives
When read from left to right, the boldface words demarcate the transit from conscious to
implicit mental processing.
On his way, our modeler will have mastered many -progressively more demanding-
leaning 1 situations and he finished a leaming 2 itinerary that made him become the expert
he nowis.
Using the model for System Dynamics
We now tum back to the full question of becoming a system dynamicist, involving skillful
systems thinking and modeling.
It is not clear how much time and effort it takes to develop from one level to the next, but
most people who have served as subjects in the reported studies (Moxnes, 2000; Sweeney
and Sterman, 2000; Ossimitz, 2002) were beginners, and most active workers in the SD
community are proficient competent or experts. If so, what should be expected from test
subjects is no more than skills that all of us embody implicitly (for example managing not
to flood your house just because you wanted to take a bath); being able to avoid hot-cold
oscillations in the shower, as well as more complex and “systems”-specific tasks would
remain the domain for more advanced leamers.
Of cause, it is a rather crude simplification to consider that the leamer(s) in phase
“beginner” remain beginners until the day the flow into the “advanced beginner” level; we
know that each of the leaming experiences triggers a small and progressive transformation,
and that the flow from one level to the next is going on all the time. Anyway, it takes some
time to go through what we distinguish as being a beginner, and some day the leamer is
considered “advanced beginner”.
We may thus represent a class of system dynamics students as a fourth order pipeline delay:
10
start
neccessary time BEGRINERS
for step 1
ae sip
neccessary time ADVANCED
for step 2 BEGINNERS
neccessary time COMPETENT
for step 3
neccessary time PROFICIENT
for step 4
NN stip
rest of their lives EXPERT
Fig. 2 A model of turning from beginner into expert modeller
This is but a conceptual model, since we do not know how much time it takes to transfer
from one stage to the following one. It is plausible to assume that this time varies
depending on the didactical context of the leaming situation and the personal context of the
leamer.
11
In this model, the time it takes to advance from one stage to the following one appears to be
constant and not dependent from any variable recognized as part of the system. However,
we have said that system thinking skills may influence the becoming of a good modeler.
For example, Ossimitz (2002) concludes that a “fundamental aspect is the ability to grasp
that in a stock-flow-context the stock with one inflow and one outflow is increasing when
the inflow is bigger than the outflow (or the net flow is positive, to put it in another way).
Some findings indicate that this might be a key criterion for discriminating between a
stock-flow-thinker and a non-stockflow-thinker”’. If this specific skill act as enabler, say
for tuming from beginner into advanced beginner, the chain for systems thinking leaming
would interact with the one in the above model, and the “necessary time” would change.
This calls for work on the relationship between specific system thinking skills and the
specific needs at the respective stages in the leaming chain. However, this must wait until
the usefulness of this stage-model is established, which would still have to be done.
Empiric studies’ analysis from this viewpoint
Each of the initially synthesized, recently published studies is somehow related to the issue
of thinking, knowing and leaming, even though in different ways. We will now look at
them from the viewpoint of Polanyi’s model of knowing and Dreyfus and Dreyfus’ model
of leaming. The comparison is based on the following items:
e typeof subjects used according to the leaming model used here: TSU
1. beginner,
e clues provided according to subjects: CPS
1. none
2. attributes and mules;
3. aspects and directives;
4. objective;
e intentionality: INT
1. teaching or
2. experimentation/research;
e competencies looked for: CLF
1. control asystem,
2. system thinking;
e competence level looked for. CLLF
12
e leaming required for the competencies: LRC
1. na
2. improve mental model of system,
3. improve decision policies (leaming I; see appendix 1);
4. improve system thinking skills (leaming ID);
5. improve the way complex situations are approached (leaming IT)
e system dynamics action strategy: SDAS
1. model
2. simulate
3. implicit integration
e assessment approach: AS
1. self-reports,
2. attitudes,
3. inference based upon behavior,
4. mental models from narrative,
5. mental models from cognitive mapping;
Study TSU | CPS INT CLF CLLF | LRC SDAS_ | AS
Doyle 1 ai 2 1 i 1,2 4
Moxnes 1 ai 2 1 i 2() 3
Jensen and | 1 L 2 1 a 2(1) 3
Brehner
Sweeney 1 ti 2 2 a na 3 3
and
Sterman,
Ossimitz 1 1 2 i a 3 2
Maani and] 3 1 1,2 1 12,4 3,4
Maharaj
Besides from remarking that many of the possible combinations have not been used by the
studies used here, we mainly see that
e although subjects almost always were beginners in the field of systems thinking
(low TSU), the studies expected them to be (naturally?) skillful, so to say: experts;
e according to this expectation, no clues were provided for beginners (low CPS);
e the majority of the studies used a rather short time horizon; none of the studies
looked at the learning of systems thinking skills;
e the subjects had either to know or to leam from interacting with a simulation;
modeling was not used as leaming strategy;
e the focus is usually set on leaming I;
e there is diversity as for assessment approaches.
13
These studies showed that “ordinary” people tend to be systems thinking beginners. The
above pipeline - delay model indicates that if nothing in the previous life experience of a
subject (including formal education) provides the necessary leaming, this had to be this
way.
Investigating the transformation from beginner to advanced beginner
It is argued here that studies on the process of leaming of systems thinking skills are
needed, and it is hypothesized that the proposed model of leaming is useful for this
purpose. If the stage-model of leaming to think systemically is useful, then subjects who
are beginners and receive help in form of attributes and rules should outperform other
beginners without this help. Kainz and Ossimitz (2002) gave their subjects a 90 minute
presentation of stocks and flows where we find a series of items like examples, attributes
and some properties, but the presentation does not contain rules comparable to those that
underlie the scoring of the test tasks (like “When the inflow exceeds the outflow, the stock
is rising”). It would be important to know how subjects would perform if the help given
includes these rules. It would also be relevant to assess how much experience (tasks) it
takes until they reach the following stage. We can assume that the usual SD-courses do not
allow to develop beyond the advanced beginner stage (in one semester's time). So in this
section, we focus on this transition. Also, what follows is restricted to the “stocks-and-
flow’ part of systems thinking.
There are three hypotheses to be comoborated:
1. if beginners need explicit and context-free attributes and rules, then the first series
of (adapted) tests should detect superior performance of the subjects.
2. if the beginner stage is a necessary phase in order to become an advanced beginner,
then the subjects who participated in the first phase should outperform other
subjects in the second series of tasks.
3. subjects who have arrived at the competent stage should be able to elaborate better
mental models from interacting with a simulator and from constructing a model.
The tests can be adapted to validate the first two hypotheses (the third one is related to
feedback thinking, that is: a different set of rules and directives and has to be done with
different test instruments). The first task is the bathtub situation taken from Sweeney and
Stemman (2000) with constant flows, and it is presented together with the relevant aspects
and rules; rather that limiting time, the time taken by each subject is measured. The other 4
tests are presented without repeating the general mules (that apply to each of the tests);
instead, students can ask for a sheet where the nules are given and are also allowed to use
personal notes taken (which allows to detect who does not recall the rules) and the task
sheet provides a space where subjects are invited to optionally state the rules they use.
Again, each subject’ s response time is taken.
Ina first step, this investigation will reveal the dynamics of the use of explicit and context-
free attributes and mules. The next step will be to use the diversity of experiences (all 5 tests
are different with respect of presentation and specific skills for each type of presentation) in
order to make subjects elaborate aspects and directives, followed by a new series of tests
that take the form of small modeling tasks, in which each subjects’ skills are challenged.
14
Activities to test the first hypothesis are currently being carried out with
business students in a Chilean university (during the fall semester from March 15 through
June 24) since only the fist test has been absolved so far, the results are only briefly
presented in appendix 2 (but will be available by the conference date).
The second hypothesis has to be evaluated once there are subjects who qualify as
“advanced beginners”, and will take a comparison of one group of beginners with the group
of advanced beginners.
The third hypothesis will become meaningful once the first two have passed the
examination. Again, its corroboration will require two groups consisting of subjects in
different stages to realize the same tasks, in order to compare their performance.
Conclusion and outlook
This paper set out to make a contribution to the domain of the leaming of systems thinking
skills. This is justified because empirical studies consistently find subjects to lack these
skills in a sufficient proportion to raise concems.
The contribution consisted of the proposition to consider the basic acts of systems thinking
to be processes of implicit integration, based of the work of Michael Polanyi. In the
framework of this model, the ability to perceive systemically is basically implicit
(unconscious), but it can be deliberately leamed. This leaming process has been suggested.
to consist of five stages, according to the model developed by Dreyfus and Dreyfus:
beginner, advanced beginner, competent, proficient and expert.
Looking at recent studies on the subject of systems thinking and leaming, it was found that
they did not make an explicit distinction of phases of leaming; rather they detect the lack of
systems thinking abilities in beginners.
Three hypotheses have been derived from the leaming model and it was argued that they
make sense in the current situation. The tasks designed by Ossimitz (2002) have been
adapted and some more tests have been added in order to test the hypothesis.
The empirical research has just started, which is a limitation for the paper's message, since
it has to be understood as the construction hypothesis, and not as their corroboration.
However, this construction has opened an interesting possibility for the domain of systems
thinking, its leaming and its importance for becoming a good modeler or systems
dynamicist.
15
References
Bateson, G. Mind and nature - a necessary unity, 1979. (La Nature et la Pensée, Seuil,
1984)
Doyle, J., Radzicki, M. and Trees, W. Measuring Change in Mental Models of Dynamic
Systems: An Exploratory Study, Report No. 14, 1998
Dreyfus, H. and Dreyfus, S., 1986. Mind over machine. The power of human intuition and
expertise in the era of the computer, The Free Press
Jensen, E. and Brehmer, B., 2003. Understanding and control of a simple dynamics
system, System Dynamics Review 19(2): 119-138
Kainz, D. and Ossimitz, G., 2002. Can Students leam Stock-Flow-Thinking? An emprical
Investigation. Submitted for the 2002 Conference of the System Dynamics Society,
Palermo, Italy.
Llinas, R., 2004. The I of the vortex: from neurons to self
Maani, K. and Maharaj, V., 2004. Links between systems thinking and complex decision
making, System Dynamics Review 20(1): 21-48
Maier, F. and Grossler, A., 2000. What are we talking about? - a taxonomy of computer
simulations to support leaming, System Dynamics Review 16(2): 135-148
Morecroft, J. and Sterman, J., 1994, Modeling for leaming organizations, Productivity
Press
Moxnes, E., 2000. Not only the tragedy of the commons: misperceptions of feedback and
policies for sustainable development, System Dynamics Review 16(4):325-348
Moxnes, E., 2004. Misperceptions of basic dynamics: the case of renewable resource
management, System Dynamics Review 20(2): 139-162
Neuweg, G, 1999. Konnerschaft und implizites Wissen, Waxmann
Ossimitz, G., 2000. Systemisches Denken braucht systemische Darstellungsmittel, invitede
paper for the yearly conference of the Society for Social and Economic Cybemetics
("Gesellschaft fiir Sozial- und Wirtschaftskybemetik" - GWS), Mannheim am 30.9.2000
Ossimitz, G., 2002. Stock-Hlow-Thinking and Reading stock-flow-related Graphs: An
Empirical Investigation in Dynamic Thinking Abilities, 2002 System Dynamics
Conference, Palermo, Italy
Polanyi, M., 1966. The tacit dimension, New Y ork: Doubleday
Richmond, B. 1997. The “thinking” in systems thinking: how can we make it easier to
master., The Systems Thinker 8(2): 1-5.
Rouwette, E. and Vennix, J. Process and outcome modeling: an attempt at formulating a
conceptual framework, unpublished paper
Schunk, D., 1996. Leaming theories - an educational perspective, 24 ed, Prentice-Hall
Sterman, JD. 1989. Modeling managerial behavior: misperceptions of feedback in a
dynamic decision making experiment. Management Science 35(35): 321-339
Stemman, J. 2000. Business dynamics: systems thinking and modeling for a complex world,
McGraw Hill
16
Sweeney, L. and Sterman, JD., 2000. Bathtub dynamics: initial results of a systems
thinking inventory, System Dynamics Review 16(4): 249-286
Vennix, J. and Rouwette, E., 2002. Group model building effectiveness: a review of
assessment studies, System Dynamics Review Vol. 18(1): 5-45
Wolstenholme, E., 1990. System enquiry : a system dynamics approach. John Wiley
Wolstenholme, E. 2003. Towards the definition and use of a core set of archetypal
structures in dynamic systems, System Dynamics Review 19(1): 7-27
Wolstenholme, E. 2004. Using generic systems archetypes to support thinking and
modeling, System Dynamics Review 20(4): 341-357
17
Appendix 1: Systems thinking and knowledge about a system as
complementary types of knowledge
It is not unusual for systems dynamicists to reflect upon leaming, which is generally
defined as “generating new behavior or the intemal possibility to perform new behavior as
a product of experience” (Schunk, 1996). For SD use, one can easily translate this into the
following:
e “new behavior’ refers to new policies and organizational structures;
e “new possibility to perform behavior” means new or improved mental models.
These are two forms of knowledge, and leaming is the process of elaborating them.
Leaming is a process of change in the body of knowledge of someone. It is work to be
done by a leamer, and if we wish to help someone to leam, we can only provide him with
helpful experiences. In this sense, Maier and Grossler (2000: 139) state that “leaming
objectives are the mediation of declarative knowledge (knowing that) as well as procedural
knowledge (knowing how) and structural knowledge (knowing why). [...] the meta
purposes of the simulation programs [...] are to help users understand the principles of the
underlying system and to train users in controlling the systen’.
These goals are clearly related to structural knowledge (understanding) and procedural
knowledge (controlling) in the context of one particular system. It is thought that the
interaction with a modeling tool or a simulation model will allow the users to modify their
mental model and/or their skills. According to Doyle and Ford (1998:17), “a mental model
of a dynamic system is a relatively enduring and accessible, but limited, intemal conceptual
representation of an extemal system whose structure maintains the perceived structure of
that system” (emphasis added). As it is articulated, a mental model is declarative
knowledge referring to some combination of procedural and structural knowledge.
What is knowledge that it can be declarative, procedural and structural? Is the capability to
control a system - a skill- comparable to understanding? Is procedural knowledge
accessible in the sense of a subject being able to articulate it as declarative knowledge?
Does mastering one’s bathtub draw on the same knowledge as resolving the bathtub tests,
oris the declarative knowledge used in reflecting on the test different? And: what is known
when one has structural or declarative knowledge? Is it the particular system under study
or some more generic knowledge that enables to better study any such system? Doesn't it
imply system thinking or system modeling procedural knowledge?
Further, what kind of knowledge is meant when one says “system thinking’? Without
being able to discuss different propositions here, Maani and Maharaj (2004: 22) refer to
Richmond's (1997) definition: Dynamic thinking -> Systenras-cause thinking ->Forest
thinking -> Operational thinking -> Closed-loop thinking -> Quantitative thinking ->
Scientific thinking. Thinking is a process that remains for its main part implicit, as our
conscious reflections are only a small part of all the cerebral activity going on. This means
that system thinking would be a set of skills rather that declarative knowledge.
Ossimitz (2002) describes systemic thinking as:
e “Thinking in Interrelated Structures;
e Dynamic Thinking, which means a thinking which is not restricted to grasping just
snapshots of a situation, but takes into account evolution over time.
18
e Thinking in Models, which means that any systems thinker should be aware that he
or she is always dealing with a model of a complex situation, which is usually
massively simplified compared with the "actual" situation.
e Systemic Action, which means the practical ability of steering systems”.
The fact that Ossimitz refers to “thinking” is interpreted here as meaning that it is an
implicit process of knowing more than a conscious knowledge.
Sweeney and Sterman (2000:250) define systems thinking as being able to understand
“behavior that arises from the interaction of a system’ s agents over time”, specifically:
e “understand how the behavior of a system arises from the interaction of its agents
over time (i.e., dynamic complexity);
e discover and represent feedback processes (both positive and negative)
hypothesized to underlie observed pattems of system behavior;
e identify stock and flow relationships;
e recognize delays and understand their impact;
e identify nonlinearities;
¢ recognize and challenge the boundaries of mental (and formal) models”.
This definition is somewhat closed to the SD commumity; it describes what the systems
thinker is able to do: he comes to be aware of feedback loops, stock-and-flow relationships,
delays, nonlinearities and model boundaries. These are the building blocks of the system.
dynamics language. _It is not clear to which point “recognize” and “identify” refers to
spontaneous recognition/identification of these structures or if they have to be consciously
looked for in order to “discover” them (or some of both)
Anyway, systems thinking is always presented as a fluid form of knowledge that enables its
owner to perform better leaming in front of new situations or systems. In other words,
these skills are second order with respect to “understanding a given system X”. In the
terms of Bateson (1979), there are several types of leaming about a system.
e Leaming 0: coming to know how such or such a variable behaves. This takes a
previously existing model in which this variable makes sense. Many people will
feel that this is not really leaming, but it is hard to deny that when your company’ s
profits where “high” last year and now they are told they are “low”, you just came
to know something new: you leamed. Also, reading in a SD textbook that there is
something called “delay” may be called leaming 0. Knowledge at this level would.
be rather declarative: one knows that there are feedback loops, stocks and flows,
delays, nonlinearities and boundaries.
e Leaming 1: when confronted with an unknown system or situation, one has to
generate a way of dealing with it, calling upon and evolving one’s mental model of
it. This problem solving performance is clearly a higher form of leaming, and it
produces the framework inside which leaming 0 can go on. SD students will be
frequently confronted to this kind of situation, and it is our expectation that they
will do better over time. Here, knowledge would be rather procedural in that one
knows how to find these building blocks in a given situation (one knows how to
19
access or articulate a model in terms of these building blocks). The knowledge
produced would be structural, as one now understands the studied situation in terms
of the same building blocks.
e Leaming 2: the building of all the skills that help performing “leaming 1” situations
using less resources. This may comprise knowing that such things like feedback
loops or delays may exist, and certainly knowing how to model step-by-step and
also being able to intuitively grasp part of the system’ s structure or behavioral logic.
This leaming of a context inside which the situations are alike is the explanation for
the observed progresses in leaming 1 situations and thus it is called leaming 2. The
experienced or even expert modeler is one who has passed through a leaming 2
process. The knowledge belonging to this level should be thought of as highly
procedural, and its articulation would be a declarative reproduction of what is
implicitly known.
e Leaming 3: there is no guarantee that there can be only one context: there may be
several ones, and in some cases a subject would have to leam that a given leaming 1
situation no longer belongs to its usual context, but has to be classified as belonging
to anew one. This leaming of new contexts is called leaming 3; since leaming 2
gives us our typical way of approaching life and its situation (our personality),
Bateson thought leaming 3 to be exceptional and rather traumatic for the leamer. It
has to be suspected that tuming from a non-systemic thinker into a systemic thinker
is anact of leaming 3.
According to Bateson, these types of leaming go on at the same time: while working hard
to elaborate a model (leaming 1), the subject also leams at level 2, and this is how
experienced modelers transformed themselves from the beginners they once were into the
experts they now are. For example, when a person is prompted to make a statement with
respect to some situation (like in the study about mental model changes by Doyle et al.,
1998), and the person switches from textual expression to causal loop diagrams, there
would have been leaming at several levels. Now the person:
0. knows that CLD exist,
1. is able to useit in order to resolve a situation;
2. has anew tool for resolving this type of situation, and maybe even.
3. is tuming into a systems thinker - according to Ossimitz (2000) the way we
can represent something influences the way we know and think about it.
The conclusion from this argument is that what studies like Sweeney and Sterman,
Ossimitz or Moxnes investigate is at a metalevel with respect to research on mental
models. It follows that the ability to perceive systemically is relevant for (prior to?) being
able to develop a helpful model, be it qualitative or quantitative. This in tum leads to ask if
the traditional SD activities - modeling or interacting with a simulation - per se would need
to be preceded by an intervention in the systemic perception. According to Sweeney and
Stemman (2000), this has been the reason why “Business dynamics” (Sterman, 2000) offers
extensive treatment of graphic derivation and integration. However, this should also be
20
accounted for when investigating the effectiveness of modeling or “flying” a simulator for
leaming about a system: what level of system thinking competencies do the subjects have?
21
Appendix 2: the study concerning the learning of systems thinking skills
This study consists of two phases. The first one applies a modified version of the tests used.
by Ossimitz (2002) to undergraduate students of business administration in a Chilean
university (two groups corresponding to two different courses). The first group is made up
of 56 students and it is about information systems development. The second course is
about system dynamics and has 20 students.
At the date of march, 18, only one week of courses has passed, so by the time this paper is
sent, it is too early to discuss the measured results and compare them with the ones
observed by Sweeney and Sterman (2000) and Ossimitz (2002).
Anyway, the first test revealed the following performances:
Task S+S 0 Sch ISD sD
(N=49) (N=34)_| (N=15)
When the inflow exceeds the outflow, | 0,87 0,42 | 0,55 0,41 0,87
the stock is rising
‘When the outflow exceeds the inflow, | 0,86 043 | 053 041 0,87
the stock is falling
The stock should not show any | 0,96 0,64 | 0,55 0,38 0,93
discontinuous jumps (it is piecewise
continuous)
The peaks and troughs of the stock | 0,89 0,56 | 0,55 0,38 0,87
occur when the net flow crosses zero
During each segment the net flow is | 0,84 0.38 | 0.55 0,38 0,87
constant so the stock must rise (fall)
linearly
The slope of the stock during each | 0,73 0,26 | 0,29 0,03 0,87
segment is +/- 25 units/time period.
The quantity added to (removed | 0,68 0,27 | 0,31 0,03 0,93
from) the stock during each segment
is 100 units, so the stock peaks at 200
units and falls to a minimum of 100
units.
0,83 042 | 0,47 0,28 0,90
The discrepancy between the two groups is surprising, but not yet explained. The tests will
continue over the next 4 weeks.
22
In phase 2, subjects will be prompted to model a sequence of simple situations; in each, the
subject has to make a prediction based on graphical integration / derivation. The sequence
is:
e onesingle positive feedback loop
e onesingle negative feedback loop
e each of Wolstenholme’ s totally generic systemic archetypes (Wolstenholme, 2003,
2004), which are all 4 combinations of two interacting feedback loops.
In each of the six tasks, the subject is confronted with a description made from a piece of
text and the graph showing the dynamic of one of the variables; the student’s task is to
decide the type of each identified variable, connect the variables and infer one target
variable’s dynamic from the model’s structure and the available graph. The work sheet
prompts the subject to draw an influence diagram (discriminating between stocks and rates;
see Wolstenholme, 1990). The following figure (next page) is a translation of the first task
from Spanish into English language. To resolve this task, one must be able to tell if a given
variable is a stock or a flow, and be able to infer one variable’s dynamic from the other
one’s.
23
You are the owner of a savings account that rearns a monthly interest corresponding to 2% of the account's
balance. This means that the more money there is in your account, the more interests the bank will pay you
in one month; and the more interests they pay you, the more money will be on your account.
Draw an influence diagram of this system:
Assume that for opening the account, the bank required you to deposit $10.000. How will the system's
variables evolve over the first year?
© 1 2 3 4 5 6 7 8 9 10 11 12
Time (months)
25
