=Paper=
{{Paper
|id=None
|storemode=property
|title=Knowledge Acquisition for Simulating Complex Psychological Processes
|pdfUrl=https://ceur-ws.org/Vol-636/kese6-03.pdf
|volume=Vol-636
|dblpUrl=https://dblp.org/rec/conf/ki/NewoBA10
}}
==Knowledge Acquisition for Simulating Complex Psychological Processes==
https://ceur-ws.org/Vol-636/kese6-03.pdf
Knowledge Acquisition for Simulating Complex
Psychological Processes
Régis Newo, Kerstin Bach, and Klaus-Dieter Althoff
University of Hildesheim,
Institute of Computer Sciences,
Laboratory of Intelligent Information Systems
Email: lastname@iis.uni-hildesheim.de
Abstract. In this paper we introduce an approach that elicits infor-
mation from psychologists and transfers this knowledge in knowledge
representations for a knowledge-based system. Based on SIMOCOSTS,
a psychological model for simulating coping strategies in critical situa-
tions, we present a case study on a literary character that exemplifies our
approach. We describe the simulation sequences in an abstract way and
elucidate how we utilise the acquired knowledge. We use Eichendorff’s
literary character of ”From the Life of a Good-for-Nothing” (”Aus dem
Leben eines Taugenichts”) as an example for a psychological analysis,
which is transferred into a simulation scenario according to the SIMO-
COSTS model.
1 Introduction
Artificial Intelligence (AI) has been successfully applied in various research disci-
plines trying to explain complex contexts. Especially when dealing with complex
problems, a simulation of the problem helps to understand interdependencies be-
tween various components. When a person has to cope with a critical situation
in her or his life, many reactions are possible responding this situation, so it
is very interesting how the decision they made have been developed. The work
presented in this paper picks up on this topic and presents an approach for sim-
ulating a person’s behavior coping with critical situations. The approach has
been developed together with development psychologists who have developed a
psychological theory how decision making in stressful situations can be explained
[1]. In this approach the decision making process is explained as a market place
with various buyers and sellers. Each of them has an individual goal and they
bargain with each other trying to receive its goal. The buyers and sellers in
the theory are goals that have been developed over years and the result of the
negotiation is a strategy how the person deals with the situation.
We picked up the idea of a market place idea where different goals meet
and transfered it to an multi-agent-system that simulates the determination of a
strategy. The underlying architecture of our approach is the SEASALT (Sharing
Experience using an Agent-based System Architecture LayouT) architecture [2]
which is based on the CoMES (Collaborating Multi-Expert-Systems) approach
[3], the research vision of our group.
SEASALT consists of five components and is built around a multi-agent sys-
tem containing intelligent agents providing information (Knowledge Provision).
Each agent, so-called Topic Agent, is equipped with a knowledge-based system
that derives it information from some kind of community (Knowledge Sources)
that is pre-processed and formalized in order to be usable by a knowledge based
system (Knowledge Formalization). Furthermore a Knowledge Representation
component provides shared knowledge like rule sets and domain ontologies. The
interaction with the user is realized via a user interface that interacts with the
multi-agent system. The simulation of developing coping strategies focuses on the
multi-agent- system and the according knowledge formalization which is mainly
carried out by psychologists supported by intelligent graphical interfaces. By
talking about knowledge based systems, we are currently focusing on case-based
reasoning (CBR) systems, because CBR is an established methodology that uses
previous experiences to solve new problems. Moreover, CBR already works with
representative examples and thus performs well in interdisciplinary projects as
ours. Experts have to provide initial examples and by using the application the
figure our ”wrong behaviors” and provide more examples. So, the application
develops successively by integrating the new examples. In comparison to other
SEASALT realizations like docQuery [4] that are focusing on co-operating topic
agents where the agents deal with complex problems by dividing them into top-
ics that are solved individually and afterwards the solutions are put together.
However, in this approach we have competitive agents, because each intelligent
agent has its individual goal and they are competing against each other. The
knowledge acquisition is almost the same, because a knowledge engineer is sup-
ported by intelligent agents or processed developing knowledge models and cases
that can further be used by the CBR systems.
The remaining part of this paper presents the SIMOCOSTS model and the
briefly introduced marked place idea in section 2.2, followed by the introduction
of our case study based on Eichendorff’s literary character of ”From the Life
of a Good-for-Nothing” (”Aus dem Leben eines Taugenichts”) [5] in section 3.
Section 4 discusses how knowledge can be acquired and gives examples taken
from the case study. The final Section 5 sums up the paper and points out the
next challenges for us.
2 Simulating human processes
As stated in the previous section, we deal in our project with human behaviour.
In this section, we will first present some psychological background that we use
for the implementation of the simulation. It should be noted that we want to
develop a tool that should be used only by domain experts.
Any time we have to deal with simulations, we have to know exactly which in-
formation we need and what we have to take into account. In our case, it means
that we have to know which information is processed by a person while facing
a critical situation. Yet, as we all know, it is (almost) impossible to capture the
complete knowledge of a human being. This due to the fact that that whole
knowledge is saved in the so called long term memory which can, according to
psychologists, store an infinite amount of information [6].
Luckily, humans never access their whole long term memory at the same mo-
ment. We instead just use our so called short term memory, which can be seen
as a very small activated part of our long term memory. The short term memory
contains information related to the situation we are actually experiencing. More
on memory activation can be read in [7, 6]. That perception of memory legiti-
mates us to just consider situation related information during our simulations.
That means that we do not have to capture all kind of information that a human
can have. We can restrict on the situation related knowledge in order to simulate
a plausible behavior of human beings in critical situations.
Nevertheless, our simulation will be based on the theory of Brandtstädter and
Greve [8]. It is based on the fact that intentions are a key part of psychological
theories of action. Except for knee-jerk or automated behaviours, human actions
are motivated by intentions. When somebody faces a critical situation, his actual
state strongly differs from his goal state (i.e., his intentions). In order to solve
the problem, the person essentially can use one of the following three forms of
coping processes:
– Assimilative processes: the strategy here is to solve the problem by working
directly on the actual state. That is, it is an active art to work through
a problem, in which the person uses the available resources in a problem
oriented way. The available resources can be the person’s own resources or
external ones.
– Accommodative processes: this strategy is used when the person believes he
can not change the actual state (i.e. solve the problem) by himself. He then
tries to adapt his goal state such that the discrepancy to the actual state
can be diminished.
– Defensive processes: in this case, the person just ignores the discrepancy
between the actual state and his goals. He can for example perform actions
that diminishes the meaning of the discrepancy.
It should be noted that a person does not (normally) intentionally apply a given
type of process. The person rather just try to find out, which strategy would
be the best for him at the moment (depending on his capabilities, environment,
etc.). The chosen strategy can then be evaluated to belong to one of the given
processes by experts.
We will consider that the goals of a person play a crucial role for his be-
haviour. We also suppose that these goals are competitive. In fact, we will con-
sider that human’s mind is comparable to some kind of market place which
contains those competitive goals.
Another question which have to be taken into account is to know which part
of the knowledge of a human play a significant or even an essential role when
he faces critical situations. This is actually a question that will be answered
with the aid of the simulation tool, because even psychologists can not give
Fig. 1. Simplified Version of SIMOCOSTS (see [9])
a precise answer. Nevertheless, there are some parts (i.e. components) which
are believed to a play a major role. To that extend, we developed a model,
called SIMOCOSTS [9] (SImulation MOdel for COping STrategy Selection), in
which we represent those components and also the interactions between them.
A simplified version of the model can be seen on Figure 1
Yet that model can not be directly used for an implementation. We still
for example have to figure out how the information is going to gathered and
stored. The gathering of information will be discussed in Section 4. While dealing
with human processes, many knowledge representation techniques have to be
considered, because of the diversity of human knowledge. Yet we believe that
humans mostly rely on past experiences (first or second hand), particularly in
difficult situations. That is why we will mainly rely on Case-Based Reasoning
(CBR) [10] as our main knowledge representation technique.
Considering the organisation the knowledge, we distinguish between two
types of knowledge, which differs in their accessibility for the simulated per-
son. The first type is called shared knowledge and is always accessible for the
person. It contains general facts about the person as well as personal character-
istics. The second type of knowledge is called unshared knowledge and is only
available for the goals. It mainly contains information about the concerned goals
and strategies which can be used if there is a critical situation.
Fig. 2. Architecture for the Knowledge Management in SIMOCOSTS
A generic illustration of the simulation with respect to the knowledge man-
agement can be seen in Figure 2. The market place contains several processes,
in our case the goals, which interact with each other and are competitive. Each
process can access its own unshared knowledge which can in turn consist of mul-
tiple sources. As mentioned earlier, the knowledge sources will rely on CBR. Yet
we do not rule out the use of other techniques (e.g. ontologies or rules).
This architecture actually displays the knowledge line of the SEASALT ar-
chitecture [2]. SEASALT thus provides a very good platform for the organisation
of the (unshared) knowledge, which can be realised with a case factory [11].
In order to give a better insight on the realisation of the simulation, we will
elucidate in the next paragraph the underlying concepts and workflows used for
the implementation.
2.1 Concepts of the Simulation
Person The main concept is the person itself. As can be seen on Figure 2,
the person can access a so-called general knowledge, which contains for instance
general facts about the person. These general fact include psychic as well as
physical facts. It means that we assume that we know how the person feels
physically and which abilities he has as well as the personality of the simulated
person. We use several psychological theories in order to represent all those facts.
In the style of the personality theory developed by Asendorpf [12], we describe
the psychological facts among other things with the self-concept of a human
by using adjectives defined by Müskens in [13]. These adjectives are grouped in
several classes which include for example happiness and creativity. The physical
abilities are similarly represented by defining the hard skills of the simulated
person.
We thus represent all those attributes of the person with attribute value pair in
which the attributes display the characteristics and the values whether and to
which extend they are existent.
Situations In our simulation, a situation is a description of a state. It might
for instance be the description of the actual state of the person, thus indicating
how the person feels. We use situations to model two important parts of our
architecture.
Events An event is a situation which may have an impact on the characteristics
of the person. Events represents the situation that the person faces which might
affect him. The simulation therefore starts with the (external) generation of an
event with is then passed to the person.
Goals Goals, as stated earlier, are things that the person wants to achieve. They
are thus concrete specification of situations that the person wants to achieve.
Goals might have priorities which show their importance for the simulated per-
son. An accommodative strategy to achieve is then capable to change the goal
or adjust the its priority.
Strategies Strategies are the plans that can be used by a person in order to
achieve a certain goal. These plans can consists of several steps which affect either
the actual situation (assimilative) or the concerned goal (accommodative).
2.2 Workflows of the Simulation
A simulation run consists of five main steps:
1. The generated event is compared to the goals of the person in order to find
out whether we have a critical situation.
Fig. 3. Correlation between the concepts and illustration of the workflows
2. The goals gives a feedback of that evaluation to the event. That feedback
contains a so-called influence factor vector which indicates how the event
actually affects the person.
3. The influence factors are applied on the personal characteristics.
4. The best strategies for the affected goals are computed.
5. The influence factor generated by the strategies is applied on the character-
istics of the person.
Afterwards, the psychologists analyse the state of the person after the run
and also have the possibility to start a new run with different parameters. These
workflows are illustrated in Figure 3
3 Case Study
In this section, we will present an example to illustrate the architecture. We
decided to stick with an literary example because the (important) information
is given and accessible for everyone. Another reason is the fact that we also have
the output that our simulation tool should give. It is hence a good way to tune
our implementation.
Our example is a literary character of ”From the Life of a Good-For-Nothing”,
a novella of Joseph von Eichendorff [5]. It is the story of a young man, called
Good-For-Nothing, who is very lazy (hence the name). He likes to sing and play
the violin. Because he does not like to work and does not help at home, his fa-
ther send him to the wide world. His adventure starts there. Although he is quite
happy at the beginning, because he likes journeys, he faces several difficulties
during his adventure. Nevertheless, he always finds a way to come through, even
if he is sometimes lucky.
Our aim is to simulate the main character, Good-For-Nothing, in different
situations. For this example, we will concentrate on the psychological facts about
the character, because the physical facts do not play an important role in his
decision making. The only hard skill that can be mentioned is his ability to play
violin. As for his self-concept, Good-For-Nothing is a positive and preponderant
lucky person. He is also authentic, candid, unorderly, educated and creative.
Furthermore, he can be seen as a sentimental person and in need of affection.
He has several goals as the story goes on. First, as he leaves his father’s
home, he soon realizes that he has to earn his keep. This situation is a difficult
one because he doe not like to work. Later he gets to know a girl, yet she
does not seem to like him at fist sight. Moreover, he has the goal to go to
Italy because he heard many good things about the country. Because of his
laziness, Good-For-Nothing often applies accommodative strategies to overcome
his critical situations. That means, he just changes his goals or at least their
priorities. Most of his assimilative strategies are based on luck.
4 Knowledge Acquisition in SIMOCOSTS
With the previous example, we can see that there are many information that
should be given before an initial simulation run can be made. Because the tool
should be used by domain experts, we have to deal with the knowledge acquisi-
tion from experts. Two main problems arise. First, the expert has to give many
different information before the start. He does not only have to give a complete
and detailed description of the person (i.e. characteristics), he also have to pro-
vide the goals, strategies to achieve these goals and also events, which should be
evaluated by the person (as critical situations or not).
The second main problem concerns the amount of information needed for each
part of the simulation. In order to avoid a trivial simulation, we do not only
need decent algorithms, but also enough information. Because of the fact that
we intend to use CBR as main knowledge representation technique, we need for
example many strategies the person can rely on before we can start simulation
runs. These two points lead to the fact we need a good knowledge acquisition
methodology. That methodology should of course also take into account that
the experts, in our case the psychologists, often have a different perception of
the underlying model and the simulation than computer scientists. Furthermore,
they want to able to have non trivial simulation runs, which is necessary in order
to be able to develop (psychological) theories.
The basic idea of our approach is based upon the fact that the domain ex-
pert should not be overwhelmed by the amount of information that has to be
entered.
In our concrete example, we aim at having as few input masks as possible.
For that purpose, we try to gather similar information in one step instead of
getting each kind of information in different steps. This leads to the need of an
intelligent combination of similar information inputs. Determining the appropri-
ate combination is of course a highly domain dependent process. Another point
that has to be considered is the use of information extraction techniques. The
domain expert should be able to enter the needed information in his preferred
form, because this might encourage him to give more input. It is therefore a ne-
cessity to use information extraction techniques, in order to be able to structure
the given information in our case-based simulation tool.
For our simulation we will have two input masks. The first one allows the expert
to give information about the person. That means that the experts do not only
give general facts about the person, but also his goals. In the second mask, the
expert should be able to enter information about the events as well as some
strategies which can be applied in order overcome the critical situations. This
is advantageous for the experts because the can give events as well as solutions
which should adapted and applied on the event. We should remark that the spec-
ification of strategies does not trivialise the simulation, because it is important
to find out which factors play a role for the adjustment of the strategies.
With both input masks, we will be able to gather information about the four
knowledge areas in our simulation (namely personal characteristics, goals, events
and strategies).
4.1 Discussion
We are still implementing the approach. Therefore we can not provide an evalu-
ation yet. Nevertheless we strongly think that it would facilitate the knowledge
acquisition from the experts.
If we consider the case study from the previous section, the experts would
have to provide in the first mask the characteristics mentioned (lazy, candid, etc.)
as well the goals of Good-For-Nothing (e.g. earn his keep or have a relationship
with the girl he loves). The second mask offers the possibility to enter events
which will start the simulation and also strategies sketches for the event. The
experts thus can provide information for several knowledge bases in the same
iteration.
5 Summary and Outlook
We presented in this paper an architecture for the implementation of the simu-
lation of complex processes. Our generic approach for the simulation is applied
in a psychological realm, namely the simulation of human behaviour in so called
critical situations. The developed is intended to be used by domain experts
(psychologists) with the aim to develop and test theories. The simulation tool is
currently been implemented while using the SEASALT architecture.
One major problem for our simulation is the knowledge acquisition. We gave
an approach and elucidate how we intend to cope with that problem. Further
steps towards realising our simulation tool include the implementation of both
approaches and an evaluation of the tool.
References
1. Greve, W., Wentura, D.: Personal and Subpersonal Regulation of Human De-
velopment: Beyond Complementary Categories. Human Development 50 (2007)
201–207
2. Reichle, M., Bach, K., Althoff, K.D.: The SEASALT Architecture and its Real-
ization within the docQuery Project. In Mertsching, B., ed.: Proceedings of the
32nd Annual Conference on Artificial Intelligence, KI-2009, Paderborn, Springer
(September 2009) 556–563
3. Althoff, K.D., Bach, K., Deutsch, J.O., Hanft, A., Mänz, J., Müller, T., Newo,
R., Reichle, M., Schaaf, M., Weis, K.H.: Collaborative Multi-Expert-Systems –
Realizing Knowlegde-Product-Lines with Case Factories and Distributed Learning
Systems. In Baumeister, J., Seipel, D., eds.: Proc. of the 3rd Workshop on Knowl-
edge Engineering and Software Engineering (KESE 2007), Osnabrück, Germany,
Berlin, Heidelberg, Paris, Springer Verlag (2007)
4. Bach, K., Reichle, M., Althoff, K.D.: Case-based reasoning in a travel medicine ap-
plication. In Bichindaritz, I., Jain, L., eds.: Computational Intelligence in Medicine.
Advanced Information and Knowledge Processing. Springer (2010) to appear
5. von Eichendorff, J.: Aus dem Leben eines Taugenichts. Vereinsbuchhandlung,
Berlin (1826)
6. Anderson, J.: Kognitive Psychologie. Spektrum Akademischer Verlag, Heidelberg
(2001)
7. Anderson, J.R.: A Spreading Activation Theory of Memory. In Collins, A.,
Smith, E.E., eds.: Readings in Cognitive Science - A Perspective from Psychology
an Artificial Intelligence. Morgan Kaufmann Publishers (1988)
8. Brandtstädter, J., Greve, W.: The aging self: Stabilizing and protective processes.
Developmental Review 14 (1994) 52–80
9. Newo, R., Müller, T., Althoff, K.D., Greve, W.: Learning to Cope with Critical
Situations - A Simulation Model of Cognitive Processes using Multiagent Systems.
In Hinneburg, A., ed.: Proceedings of the LWA 2007: Lernen - Wissen - Adaptivität.
(September 2007) 159–164
10. Aamodt, A., Plaza, E.: Case-based reasoning : Foundational issues, methodological
variations, and system approaches. AI Communications 1(7) (March 1994)
11. Althoff, K.D., Hanft, A., Schaaf, M.: Case factory – maintaining experience
to learn. In Göker, M.H., Roth-Berghofer, T., Güvenir, H.A., eds.: Proc. 8th
European Conference on Case-Based Reasoning (ECCBR’06), Ölüdeniz/Fethiye,
Turkey. Volume 4106 of Lecture Notes in Computer Science., Berlin, Heidelberg,
Paris, Springer Verlag (2006) 429–442
12. Asendorpf, J.B.: Psychologie der Persönlichkeit. 4 edn. Springer Verlag (September
2007)
13. Müskens, W.: Sedimente der Selbstbeschreibung: Der lexikalische Ansatz der
Persönlichkeitsforschung, Berlin (2001)