Sample Title
Total Page:16
File Type:pdf, Size:1020Kb
Die approbierte Originalversion dieser Dissertation ist an der Hauptbibliothek der Technischen Universität Wien aufgestellt (http://www.ub.tuwien.ac.at). The approved original version of this thesis is available at the main library of the Vienna University of Technology (http://www.ub.tuwien.ac.at/englweb/). DISSERTATION From Neuro-Psychoanalysis to Cognitive and Affective Automation Systems Submitted at the Faculty of Electrical Engineering and Information Technology, Vienna University of Technology in partial fulfillment of the requirements for the degree of Doctor of Technical Sciences under supervision of Prof. Dr. Ing. Dietmar Dietrich Institute of Computer Technology Vienna University of Technology Austria and Prof. Etienne Barnard Ph.D. Human Language Technologies Research Group Meraka Institute Republic of South Africa by Dipl.-Ing. Brigitte Palensky Matr.Nr. 8925308 Rembrandtstraße 22, 1020 Vienna Vienna, February 28, 2008 Kurzfassung Der Bedarf nach besserer Prozessinformation f¨uhrt zu einer stetigen Erh¨ohung der Anzahl der Sensoren in Automatisierungssystemen. Die dadurch entstehende Komplexit¨atwird durch dy- namische, ungewisse und komplexe Einsatzumgebungen noch verst¨arkt. Traditionelle, regel- basierte Steuerungen stoßen dabei an ihre Grenzen, neue, adaptive und flexible L¨osungen sind zuk¨unftig gefragt. Diese Arbeit pr¨asentiert einen neuen Ansatz technische Systeme mit den Erkenntnisf¨ahigkeiten des menschlichen Geistes auszustatten. Die Basis daf¨ursind Ergebnisse einer jungen, aber produktiven Wissensdisziplin – der Neuro-Psychoanalyse. Diese erm¨oglicht die Entwicklung einer funktionellen, kognitiven Architektur – angelehnt an das Ich/Es/Uber-Ich¨ Modell von Sigmund Freud – mit der technische Systeme zur Entscheidungsfindung ausgestattet werden k¨onnen. Ein zentraler Punkt ist die ganzheitliche Sicht von Wahrnehmung und Ak- tion. Automatisierungssysteme oder Roboter werden mit affektiven Mechanismen der Bewertung (Triebe, Emotionen, W¨unsche, etc.) versehen, die es ihnen erlauben, wahrgenommene Sensor- daten in bedeutungsbehaftete Informationen und kontext-spezifisches Wissen zu verwandeln, das wiederum die Basis f¨urdie Auswahl geeigneter Handlungen bildet. Der Kern der Architektur ist ein Ged¨achtnis, das individuelle Erfahrungen eines konkreten Systems auf emotional besetzte Weise abspeichert. Der stetige Fluss der Wahrnehmungen wird durch diese bewertet abgespe- icherten Erinnerungen von Ereignissen und eigenen Handlungen, inklusive deren Folgen, gefiltert, um hochwertige Entscheidungen f¨urdie aktuelle Situation zu finden. Erfahrungen, die in der Ver- gangenheit als positiv empfunden wurden, sowie Entscheidungen, die indirekt zu einer positiven Empfindung gef¨uhrt haben, werden wieder angestrebt. Bereits erfahrene und abgespeicherte Se- quenzen dienen auch dazu, den Ausgang von aktuellen Vorg¨angenzu antizipieren und im eigenen Sinne zu beeinflussen. Es wird untersucht, wie der gew¨ahlte neuro-psychoanalytische Ansatz das Design der kognitiven Architektur bestimmt, sowohl bez¨uglich der konstituierenden Elemente (Konzepte, Funktionen, Datenstrukturen, etc.), als auch der strukturellen Organisation und der darauf ablaufenden Prozesse. Erste prototypische Tests der neuen Konzepte werden beschrieben um zu demonstrieren, wie die einzelnen Teile der viele Ebenen umfassenden Architektur inter- agieren und w¨ahrenddes Betriebs aufeinander einwirken. Eine zuk¨unftige, vollst¨andige Imple- mentierung der Architektur stellt in Aussicht, dass technische Systeme dieser Art, trotz aller Komplexit¨at,nicht nur ein kontext-sensitives Verst¨andnis ihrer Umgebung entwickeln k¨onnen, sondern auch ein Bild ihrer selbst als handelnde Akteure. Abstract Automation systems are becoming increasingly complex, driven by a steadily increasing numbers of sensors for better process information. Additionally, such systems will be required to act in dynamic, uncertain, and complex environments. Traditional, rule-based models are too limited to create suitable adaptive systems; consequently, more flexible descriptions and solutions are necessary. This thesis presents a new approach of functionally translating useful human mental capabilities to technical systems via the construction of a unified cognitive architecture based on a combination of neurological and psychoanalytic findings and concepts – two sciences which have only recently embarked in promising cooperative directions. Psychoanalytically, out of the many possible frameworks, the architecture is inspired by the id-ego-superego model of S. Freud. A central aspect of the approach is an integrative view of perception and action. Automation systems or robots are equipped with evaluative psychic mechanisms (e.g. drives, emotions, and desires) that enable them to autonomously and adaptively turn perceived sensor information into meaningful pieces of knowledge, needed for the selection of appropriate actions. An important part of the architecture is the introduction of a system-specific memory storing individual expe- riences in an emotionally tagged way to constantly process the perceptual present through the filter of the past in order to reach decisions on ’what is currently the best thing to do’. Previously successful experiences are desired to be repeated. Known sequences of actions and events are pro- jected ahead to anticipate what will happen next, and to evaluate alternative behavioral paths in an off-line fashion. The proposed cognitive architecture is informed by several aspects of the chosen approach concerning its constitutive elements (e.g. concepts, functions, data structures), its organization, and its processes. A simple prototypical implementation of the architecture is described to illustrate how the various functions on the many levels of the architecture work together. This serves to demonstrate the potential of the proposed architecture when ‘in action’ and supports the hope that – despite all complexities – one day, when effectively implemented, the architecture can lead to technical systems that construct a context-sensitive picture about their environment – and also about themselves as subjectively planning and feeling agents. Preface The number of sensor values automation systems have to deal with per time unit will increase dramatically in the not so distant future. Moreover, there is also the demand for systems that can act in highly dynamic, complex, and uncertain environments. Traditional, rule-based models mainly used in the field so far are not adaptive enough to meet these requirements, more flexible descriptions and solutions are necessary. The fields of artificial intelligence (AI) is vast and has already seen several changes of the prevailing paradigms, from classical symbolic AI, over neuronal nets and other distributed and statistical approaches, to embodied cognitive science. Several cognitive mechanisms and architectures have been proposed. More recently, there has been an increased focus on the role of emotions in AI. Again several systems have been proposed. Most of them are either too low-level, ethology- inspired, or too rule-based, appraisal-oriented. What is missing is a comprehensive model unifying low and high-level capabilities. Some people have already suggested such models, however, almost no one (with very rare exceptions) has done this by consulting the insights of psychoanalysis. The so far suggested comprehensive models are either a) not coherent enough, or b) they stay too vague, just arguing the need for this or that mechanism without specifying how it could be realized in detail. Psychoanalysis can remedy both of these shortcomings. The work intends to design a complex, autonomous control system by taking a prominent bio- logical system as inspiration: the human mind, being able to filter vast amounts of information and to make good decisions in confusing and conflicting situations. The suggested cognitive ar- chitecture is based on neurological findings as well as on psychoanalytic principles, translated into a technical language. Drives and desires to motivate actions, basic and complex emotions to evaluate situations, different types of memories, planning (‘acting-as-if’) capabilities, and conflict resolution mechanisms are introduced as important functional elements. All these elements are arranged using the id-ego-superego model of Freud as template. The model helps to determine how to combine the processing of current external demands with the processing of current internal needs and currently available actions. A successful combination of these three elements (that is, one that serves the system’s goals, the most fundamental of which being ‘survival’) makes up the core of intelligent behavior. Of particular importance for situation recognition and categorization – two key capacities of intelligent behavior – is the use of predefined images as templates and the introduction of an emotionally afflicted episodic memory. Moreover, the system shall not only passively react, but actively build up expectations about what is supposed to happen next (‘focus of attention’). One intended target area of application is smart building automation, in particular care for handicapped and elderly people. Another potential application are mobile service robots. The suggested neuro-psychoanalytically inspired cognitive architecture is not only supposed to de- liver more context-aware autonomous systems than the ones existing today. It shall also produce technical systems that possess some ‘insight into the psychological functioning’ of human beings.