Emotional Learning in a Simulated Model of the Mental Apparatus

David C. Wyld et al. (Eds) : CSITA, ISPR, ARIN, DMAP, CCSIT, AISC, SIPP, PDCTA, SOEN - 2017
pp. 11– 18, 2017. © CS & IT-CSCP 2017 DOI : 10.5121/csit.2017.70102
EMOTIONAL LEARNING IN A SIMULATED
MODEL OF THE MENTAL APPARATUS
Martin Fittner1
and Christian Brandstätter2
1
Institute of Computer Technology, Vienna University of Technology, Vienna,
Austria fittner@ict.tuwien.ac.at
2
Research Institute of Molecular Pathology (IMP) Vienna, Austria
brandstaetter@imp.ac.at
ABSTRACT
How a human being learns is a wide field and not fully understood until now. This paper should
give an alternative attempt to get closer to the answer how human beings learn something and
what the relation to emotions is. Therefore, the cognitive architecture of the project “Simulation
of Mental Apparatus and Applications (SiMA)” is used to fulfill two tasks. One is to give an
answer to the question above and the other one is to enhance the functional model of the mental
apparatus with learning. For that reason, the functions of the model are analyzed in detail for
their ability to enhance them with a learning ability. The focus of the analysis lay on emotions
and their impact on the ability to change memories in the model to determine a different
behavior than without learning.
KEYWORDS
Learning, Emotion, Artificial General Intelligence, Simulation of Mental Apparatus and
Applications (SiMA), Artificial Recognition System (ARS), Cognitive Architectures, Cognitive
Automation, Psychoanalytically Inspired AI, Software Agents
1. INTRODUCTION
The target of the project Simulation of Mental Apparatus and Applications (SiMA) is the
development of a holistic functional model of the human mind. It was founded by Dietmar
Dietrich in 2000 [1] as a new attempt to overcome the lack of solutions for regulation of complex
tasks in buildings that can’t be solved with traditional algorithms. A bionic approach was used to
develop a model of the human mind. As the best available holistic functional model [2] of the
human mind the Metapsychology from Sigmund Freud [3] was taken as a framework for the
model. This has been evaluated by [4] and [5] that stated that psychoanalysis fulfills the needed
criteria of a unified holistic model of the mental apparatus. Like many other projects in cognitive
science, the project moved to a multi-agent simulation to prove the functionality of the model. A
methodology was developed [6] to transfer models from psychoanalysis and neurology into a
holistic functional model. After 16 years, the model was enhanced and refined and remarkable
results were reached [6]. But until now the functionality of learning was still missing. This paper
should give the first outlook how learning can be integrated into the model with the focus on
emotions.

12 Computer Science & Information Technology (CS & IT)
2. BACKGROUND AND RELATED WORKS
To solve specific technical problems, a lot of learning algorithms have to be developed. They
belong to the group of machine learning (reinforced learning, deep learning, neural networks,
etc.) and showed remarkable results. But their strength is the solving of particular problems. The
more they are used away from their application, the more they make failure in their calculations.
For SiMA more general learning functions are needed to solve the problem of learning in a wider
range. Therefore, machine learning can only be used in SiMA for particular data handling
problems but not as a top-level learning model.
But there are more projects that deal with learning in the area of artificial general intelligence
(AGI). These projects have the same target as SiMA to build a general-problem-solver (GPS). As
a delegate of these projects, LIDA (Learning Intelligent Distribution Agent) should be discussed
regarding their similarity to SiMA [7] and its functions for learning and emotions.
LIDA enhances its predecessor IDA (Intelligent Distribution Agent) by the ability to learn
something. It combines several theories of the mind in a cognitive architecture [8]. In LIDA
unconscious and conscious data processing is done due to the global workspace theory (GWT)
that is used there. Learning is more conceptual but some parts are implemented like perceptual
learning and episodic learning. In LIDA new information can only be stored (learned) when they
first become conscious. Emotions and feelings are used as facilitators of learning [8] in the model.
3. THE SIMA PROJECT
The SiMA project uses the metapsychology as a framework for building a functional model of the
human mind. This was done after an intensive search for a holistic unitary model of the human
mental apparatus. To ensure a unified model different experts from other scientific disciplines are
involved in the building of the model. Computer scientists are using such layer model now since
the beginning of the first computers. The split of a system in its physical representation including
dependencies and a pure data processing part without physics is a specialty of computer scientists.
This was described together with the model in detail in [9]. So, with such a layer model, we can
connect the data processing layers with the physical world. And as the physical world can be
changed, as long as the interfaces to the new physics are adapted. So, it must be possible to run a
human mental apparatus on a computer, as long as the hardware can provide enough power for
the data processing.
The model is divided into 3 layers. Layer 1 (L1 in Figure 1) is the neural layer which handles
signals from sensors and actors and represents the physical world of the model. The next layer is
the neurosymbolic layer (L2 in Figure 1) where signals are transformed to neurosymbols as
described by [10]. The neural and the neurosymbolic layer are not in the focus until now and
realized as simple pass-thru functions. The third layer (L3 in Figure 1) was the starting point of
SiMA. In this layer the human mental apparatus was located and divided into unconscious and
conscious or preconscious data processing. The unconscious part was named as primary process
and the conscious or preconscious part as secondary process. This separation has the advantage
that a huge amount of data can be processed with fast automatic functions and only a small
number of data is passed to the conscious processing part. This ensure that the system still can
make decisions in nearly every situation. In Figure 1 the track view of SiMA is shown which is an
abstract view of the model.

Computer Science & Information Technology (CS & IT) 13
Figure 1. Functional model of SiMA in an abstract view [9]
4. AN EMOTIONAL LEARNING MODEL
To create a model for learning the following parts are needed:
• Memory
• Trigger for learning
• Change or creation of new data structure parts
How these parts are realized in SiMA and how they should be adapted to achieve a learning
functionality are described in the following.
4.1. The Memory Model
The memory is located in the 1st layer of the SiMA model [9]. It corresponds to the multi storage
model of Atkinson und Shiffrin [11]. A new interface is needed to access the memory from the
functions. This can be seen as a separate data storage layer model to the data processing layer
model of the functions as shown in Figure 2.

Figure 2. Memory model in SiMA Figure 3. Psychic intensity and its components
The memory is further divided into short-term memory and long-term memory. An ultra-short-
term memory can’t be used due to the missing implementation of the sensor functions, but in the
model the values, that are handed from the simulator to the mind, are static for one processing
cycle which goes a little bit in the direction of a sensory memory. For the hardware storage layer
an ontology is used as a database and without production rules. A detailed description of the
memory model can be found in [9].
4.2 Emotions
Emotions are one form of valuation in the SiMA model. All valuation terms are summarized as
psychic intensity. As stated in [9] “In the SiMA project, psychic intensity is used as an umbrella
term for all valuation quotas…Valuation ultimately serves to prioritize actions in order to mediate
between the demands of the outside world (the own body and the environment) and the inner
(psychic) needs (e.g. satisfying psychic and physiological needs within the environment, or
adapting (psychic) wishes to the external circumstances)”. In Figure 3 all valuations are pictured.
Let us start with valuations of the secondary process. There we have feelings and neutralized
intensity, where feelings are built up on emotions (basic and extended emotions) and neutralized
intensity is created out of the quota of affect. So, the fundamentals of secondary process
valuations reside in the primary process. There we have the quota of affect which is a valuation of
the tension of sexual and self-preservation drives. The quota of affect was already equipped with
a learning functionality. The outcome of the simulation shows that with the adaption of the quota
of affect, that is connected with objects and actions, these objects or actions are more or less
valuated and on the basis of these valuations the agents in the simulation decided how to interact
with these objects. So, the same agent with the same personality in the same environmental setup
chooses e. a. to eat an apple instead of a carrot just depending on the learned values of the quota
of affect. As the valuation of the quota of affect is a part of the primary process it is unconscious.
This matches perfectly with the human decision making when choosing something to eat. There is
often no conscious reason why we sometimes choose this and sometimes that but there is an
undeniable tendency that drags us to choose one of them.

Figure 4. Creation of basic emotion in SiMA [6]
Emotions were separated into basic and extended emotions. Basic emotions are identical in every
mammal and are formed out of four factors which can be seen in Figure 4. These factors evolve
out of changes of the quota of affect of all drives. Basic emotions are further extended by
conflicts of the defense mechanism. In this paper, basic emotions should be in the focus and
therefore extended with a learning functionality to create new or change emotionally weighted
memories of images. Images are snapshots of a current moment which combine all perceived
information together.
4.3 Exemplary Case
The exemplary case [12] is used in the SiMA project to validate the correct function of new
implementations. It also supports discussion of certain functionality of the model in an
interdisciplinary collaboration. In this paper, the exemplary case should show how the change of
experiences effects the agent’s behavior.
The starting point of the exemplary case are two agents (Adam and Bodo) and two food sources
(an apple and a carrot). At first, Adam perceived the situation and decided according to his
hunger to go to the food source that attracts him more (the apple). His valuation of Bodo’s bodily
expressions doesn’t lead to a change in his plan to eat the apple. As he reached the food source,
he starts to eat. In the meantime, Bodo watched Adam and due to his predefined anger, he
decided to beat Adam. When Bodo beats Adam, Adam perceives unpleasure and flees from Bodo.

5. SIMULATION
For the simulation, the above described exemplary case is used and simulated in a multi-agent
environment based on MASON [13]. There are two simulation runs with different emotional
valuations. The valuation is adapted manually until now but should be done by the agent in the
near future. After the first run (Figure 5) the perceived unpleasure of Adam is captured with
inspectors (numerical or graphical descriptions of certain internal states and decisions of the agent
during the simulation) and used to adapt the emotional valuation of the bodily expressions of
Bodo.
Figure 5. Learning phase Figure 6. Behavior after Learning in 2nd run
After this adaption, the simulation is started again with the same parameters except the adapted
emotional valuation. Now Adams emotions don’t match with the plan to eat the apple and he
decides to execute the second-best plan and flees (Figure 6).
6. CONCLUSIONS AND OUTLOOK
In contrast to the results of the learning with quota of affect valuations the valuations of emotions
are unconscious, but with the transformation to feelings they become conscious. So the agent is
also able to reason about the learned feelings that are a result of the changed emotions.
This paper shows a first attempt for the implementation of a learning functionality in regard to
emotions. Now the change of emotional valuations must be implemented to produce the expected
changes in memories. Then more complex exemplary cases can be created to show that the model
is applicable for detecting dangerous situations. Therefore, the system could be used to valuate
certain bodily expressions and actions of human beings. Of course, then we have to include a
capable perception unit and the until now missing self-categorization.
REFERENCES
[1] Dietrich, Dietmar (2000) “Evolution potentials for Feldbus systems. In: Factory Communication
Systems”, Proceedings, 2000 IEEE International Workshop on Bd. 1, 2000, 145-146. - Invited Talk

[2] Kandel, E. R. (1999) “Biology and the future of psychoanalysis: A new intellectual framework for
psychiatry revisited” American Journal of Psychiatry, 505-524.
[3] Freud, S. (1915) “The Unconscious, volume XIV (1914-1916) of On the History of the Psycho-
Analytic Movement, Papers on Metapsychology and Other Works. Vintag.
[4] Jakubec, M., Doblhammer, K., Fittner, M., Wendt, A.: Logical Thought Based on Word
Presentations, in proceedings of EAPCogSci 2015, EuroAsianPacific Joint Conference on Cognitive
Science, Torino, Italy, September 25-27, pp. 95-100, 2015.
[5] Dietmar Dietrich, D., Fodor, G., Zucker, G., Bruckner, D. (2009) “Simulating the Mind - A Technical
Neuropsychoanalytical Approach” (2st ed.), Vienna: Springer.
[6] Schaat, S. (2016) “Simulation of Foundational Human Information-Processing in Social Context”,
Unpublished undergraduate dissertation, Vienna University of Technology.
[7] Wendt, A., Gelbard, F., Fittner, M., Schaat, S., Jakubec, M.: Decision-Making in the Cognitive
Architecture SiMA. To be published at the 2015 Conference on Technologies and Applications of
Artificial Intelligence (TAAI 2015), Taiwan, 2015.
[8] Franklin, S., Madl, T., D’Mello, S. Snaider, J., (2014) LIDA: A Systems-level Architecture for
Cognition, Emotion, and Learning. IEEE Transactions on Autonomous Mental Development 6 (1),
(pp. 19 – 41), IEEE.
[9] Dietrich, D., Brandstätter, C., Doblhammer, K., Fittner, M., Fodor, G, Gelbard, F., Huber, M.,
Jakubec, M., Kollmann, S., Kowarik, D., Schaat, S., Wendt, A., Widholm, R. (2015). Natural
Scientific, Psychoanalytical Model of the Psyche for Simulation and Emulation; Scientific Report III,
http://publik.tuwien.ac.at/files/PubDat_240983.pdf, Vienna University of Technology, Institute of
Computer Technology.
[10] Velik, R. (2008). A Bionic Model for Human-like Machine Perception. Unpublished undergraduate
dissertation, Vienna University of Technology, Institute of Computer Technology, Vienna.
[11] Atkinson, R. C., Shiffrin, R. M. (1968). Chapter: Human memory: A proposed system and its control
processes. In Spence, K. W., & Spence, J. T. The psychology of learning and motivation (Volume 2).
New York: Academic Press. pp. 89–195.
[12] Schaat, S., Kollmann, S., Zhukova, O., Dietrich, D., Doblhammer, K. (2015) Examination of
foundational AGI-agents in Artificial-Life simulations. Conference on Technologies and Applications
of Artificial Intelligence (TAAI), Tainan, 2015, pp. 344-351.
[13] Luke, S., Cio-Revilla, C., Panait, L., Sullivan, K. (2004). Mason: A new multi-agent simulation
toolkit. In Proceedings of the 2004 Swarmfest Workshop. Retrieved August 21, 2016, from
http://cs.gmu.edu/~eclab/projects/mason/publications/SwarmFest04.pdf.
AUTHORS
Martin Fittner was born 1984 in Mistelbach (Austria), studied Embedded Systems at
the University of Applied Science Technikum Wien in Vienna and finished with the
degree Master of Science. He works as a senior developer for software and system
design for automotive electronic control units at the Robert Bosch AG. His main topics
are testing software for factory testing, customer bootloader programming and security
protection of the system. He is a member of the SiMA (Simulation of the Mental
Apparatus and Applications) team. Currently, he works on his dissertation at the
Vienna University of Technology in the field of the SiMA project.

Christian Brandstätter was born 1979 in Tulln (Austria), studied Informatics and
Informatics-Management at the Vienna University of Technology and finished both
studies with the degree Master of Science. He works as System Administrator for
general IT in a pharmaceutical research institute. His responsibility there is the IT
infrastructure (network, storage, and security). Currently, he works on his dissertation
and is a member of the Institute of Computer Technology at the Vienna University of
Technology. Brandstätter, M.Sc. is a member of IEEE.

Emotional Learning in a Simulated Model of the Mental Apparatus

More Related Content

What's hot (19)

Similar to Emotional Learning in a Simulated Model of the Mental Apparatus (20)

More from cscpconf (20)

Recently uploaded (20)

Emotional Learning in a Simulated Model of the Mental Apparatus