DISSERTATION Human Bionically Inspired Autonomous Agents The Framework Implementation ARSi11 of the Psychoanalytical Entity Id Applied to Embodied Agents Submitted at the Faculty of Electrical Engineering, Vienna University of Technology in partial fulfillment of the requirements for the degree of Doctor of Technical Sciences under supervision of Prof. Dr. Dietmar Dietrich Institute number: 384 Institute of Computer Technology Vienna University of Technology and Prof. Dr. Vicente Matell´anOlivera Department of Mechanical and Computer Engineering, Autonomous Systems Group University of Le´on,Escuela Ing. Industrial e Inform´atica, Spain and Dr. Dietmar Bruckner Institute number: 384 Institute of Computer Technology Vienna University of Technology by Dipl.-Ing. Tobias Deutsch Matr.Nr. 9625115 Pramergasse 27/4a, A-1090 Vienna May 2011 Abstract Computers that interact with complex systems | for example in building automation | are facing the challenge of having to cope with vast amounts of data. Furthermore, the control systems have to be able to process data within real-time constraints. Classical Artificial Intelligence is not able to provide an appropriate solution yet. The provided algorithms used for isolated tasks can hardly be integrated into a larger, more capable all-purpose system as needed for dealing with complex systems. Opposed to classical approaches in Artificial Intelligence, a novel, bionic approach is followed: The metapsychology | a model of the human psyche defined by psychoanalysis | is examined regarding its technical feasibility. Based on this analysis, a psychoanalytically inspired technical model for decision making is developed. For its evaluation, agents that populate an artificial life simulation are implemented. In the large field, which is formed by this approach, special focus is placed on the technical implementation of Freud's Id and its interaction with a body | the embodiment. The artificial, simulated world used for evaluation purposes provides a dynamic, rich test-platform. Within it, the agents are exposed to use-cases and their performance is analyzed. Although the generated model is evaluated by implementing and testing it, focus is on basic research. The provided concept is a novel approach for control architectures which have to deal with complex situations in real world environments. Future applications for the concept are to be found, next to building automation, in the fields of autonomous vehicles and in speech processing. i ii Kurzfassung Computer, die mit komplexen Systemen interagieren (zum Beispiel in der Geb¨audeautomatisa- tion), stehen vor der Herausforderung, große Mengen an Daten verarbeiten zu mussen.¨ Weiters mussen¨ deren Kontrollsysteme in der Lage sein, eben diese Datenverarbeitung in Echtzeit durch- zufuhren.¨ Die klassische kunstliche¨ Intelligenz bietet hierfur¨ keine hinreichenden L¨osungen an. Diese sind fur¨ isolierte Aufgaben konzipiert und k¨onnen kaum in ein gr¨oßeres, leistungsf¨ahigeres, universell einsetzbares System integriert werden, das den Anspruchen¨ von komplexen Systemen genugt.¨ Im Gegensatz zur klassischen kunstlichen¨ Intelligenz wird hier auf eine bionische Herange- hensweise zuruckgegriffen,¨ indem ein Modell des menschlichen Verstandes, n¨amlich die Metapsy- chologie der Psychoanalyse, hinsichtlich ihrer technischen Anwendbarkeit untersucht wird. Aus dieser Grundidee wird ein psychoanalytisch inspiriertes, technisches Modell zur Entscheidungs- findung entwickelt. Zur Evaluierung wird dieses in Agenten, die eine Artificial-Life-Simulation bev¨olkern, implementiert. Aus dem großen wissenschaftlichen Gebiet, in dem der hier vorgestellte Ansatz anzusiedeln ist, werden die Implementierung des Freudschen Es und dessen Zusammen- spiel mit einem K¨orper, dem Embodiment, als zentrales Thema herausgearbeitet. Die Agenten mussen¨ in einer reichhaltigen, dynamischen, simulierten Umgebung mehrere Testf¨alle absolvieren. Das Verhalten, das die einzelnen Agenten in der Simulation zeigen, wird analysiert. Mit Hilfe die- ser Testumgebung kann gezeigt werden, dass die neu eingefuhrten¨ Konzepte (das Seeking-System, bessere Aufteilung in libidin¨ose und aggressive Triebe und die Bedeutung des K¨orpers und seiner internen Systeme) gem¨aß der anfangs aufgestellten Behauptungen umsetzbar sind. Das entwickel- te Modell ist Teil eines reinen Grundlagenforschungsprogramms und bietet ein neues Konzept fur¨ die Prozesssteuerung komplexer Systeme. Zukunftige¨ Anwendungsgebiete werden, neben der Geb¨audeautomatisation, unter anderem im Bereich der Steuerung autonomer Fahrzeuge, aber auch in der Sprachverarbeitung zu finden sein. iii iv Acknowledgements First of all, I would like to thank my doctoral advisor Prof. Dietmar Dietrich. He has been the driving force behind the research project ARS and supported me throughout the whole time. It was always a very intense collaboration. Further, I would like to thank my second supervisor, Prof. Vicente Matell´anOlivera, and my third supervisor and project leader Dietmar Bruckner. Next, I am grateful for the good and inspiring time I had collaborating with my colleagues in ARS during the last (almost) six years. From the over 30 people who worked in the project over the time, I want to highlight Roland Lang and Heimo Zeilinger. Roland, who graduated and left the project over a year ago, read the whole PhD thesis draft and helped me a lot with his comments (But he should definitely think about different leisure time activities!). And Heimo donated Easter Saturday and Easter Monday to help me debugging the ARSIN simulator code. To everyone in the ARS team: Thank You! Marcel, without your constant pushing during the middle of my PhD studies, it would have taken me much longer to finish the state of the art. You were my daytime, weekly support. The support during the long writing nights came from half around the globe. Thanks to time shift and the Internet, Josef kept me writing. And for B¨arli:LATEX (he knows why . ). My thank goes to my parents and my brothers for their support and for being around. Especially my mother helped me in many different ways to finalize my thesis. Additionally, I would like to thank my \parents-in-law." Having this said and adding my thanks to everyone who I forgot to add I am dedicating this work to my significant other: for Tatjana. I did it! Thank You All! Written in selected Vienna coffee houses and the national library of Austria. v vi Abbreviations AGI Artificial General Intelligence MOSES Meta-Optimizing Semantic AGISim Artificial General Intelligence Evolutionary Search Simulator OpenCog Open Cognition AI Artificial Intelligence PAD Pleasure, Arousal, and Dominance A-Life Artificial Life PAIAS PsychoAnalytically Inspired ARS Artificial Recognition System Automation System ARSi10 ARS implementation number 10 PAM Perceptive Awareness Module ARSi11 ARS implementation number 11 PEAS Performance, Environment, Actuators, BDI Believe-Desire-Intention Sensors BFG Bubble Family Game RePast Recursive Porous Agent Simulation CAAT Cognitive Agent Architecture and Toolkit Theory PDL Propositional Dynamic Logic CogAff Cognition and Affect DB Database PLN Probabilistic Logic Networks GOFAI Good Old Fashioned Artificial SDL Specification and Description Intelligence Language GUI Graphical User Interface SmaKi Smart Kitchen GWT Global Workspace Theory TCP/IP Transmission Control H-CogAff Human-Cognition and Affect Protocol/Internet Protocol HMM hidden Markov model UML Unified Modeling Language HVAC Heating, Ventilating, and Air VAB Virtual Animal Brain Conditioning VRML Virtual Reality Markup Language ICT Institute of Computer Technology WASABI WASABI Affect Simulation for Agents LIDA Learning Intelligent Distribution with Believable Interactivity Agent MASON Multi-Agent Simulator of XML Extensible Markup Language Neighborhoods . or Networks . or Zamin Zoological Agents for Modification and something . Improvement of Neocreatures vii viii Table of Contents 1 Introduction 1 1.1 Overview.........................................1 1.2 Methodology.......................................2 1.2.1 Lessons Learned from Artificial Intelligence..................2 1.2.2 Why Psychoanalysis?..............................4 1.2.3 Objections to Psychoanalysis..........................5 1.2.4 Modeling Process.................................6 1.3 Context..........................................7 1.4 Problem Statement....................................9 1.4.1 Motivation....................................9 1.4.2 Research Considerations............................. 10 1.5 Approach......................................... 11 2 State of the Art 14 2.1 Artificial General Intelligence Architectures...................... 14 2.1.1 Foundational Architecture Frameworks..................... 15 2.1.2 Project Architectures.............................. 17 2.1.3 Evaluation and Comparison of Architectures................. 24 2.2 Emotions, Drives, and Urges.............................. 26 2.2.1 Theoretical Considerations........................... 26 2.2.2 Projects...................................... 28 2.3 Embodiment....................................... 34 2.3.1 Foundations.................................... 34 2.3.2 Design Principles................................. 36 2.3.3 Virtual Embodiment............................... 39 2.4 Related Projects Basing on Psychoanalysis...................... 41 2.5 Artificial Life Simulation................................. 47 2.5.1 Simulation Platforms............................... 47 2.5.2 Related Projects................................