Dynamics, Synergetics, Autonomous Agents

The following sections are included:

• CONTENTS

• FOREWORD

• INTRODUCTION

• WHY DYNAMICS? BECAUSE IT TACKLES REAL ISSUES

The interdisciplinary field of synergetics looks for general principles that govern the behavior of self-organizing systems. These systems may spontaneously produce macroscopic spatial, temporal or functional patterns. This paper outlines general concepts of synergetics and gives examples of its applications to the psychology of perception by use of computer models for pattern recognition, recognition of ambiguous figures, oscillations, hysteresis and the impact of fluctuations. More recent applications deal with associative chains and mechanisms for intention. The appendix contains a brief outline of the mathematics underlying the latter approaches.

Increasing numbers of scientists are producing dynamical models and explanations of various aspects of cognition. The central insight underlying their work is the dynamical hypothesis: cognitive agents are dynamical systems, and can be understood as such. This paper aims at philosophical articulation of the dynamical hypothesis. The hypothesis is clarified and elaborated in the process of responding to a range of standard objections raised. The conclusion is not that the dynamical hypothesis is true, but rather that it is a bold scientific conjecture, worthy of detailed empirical investigation.

This article deals with mathematical models of discrete, identifiable, ‘symbolic’ events in neural and cognitive dynamics. These dynamical symbols are the supposed correlates of identifiable motor action patterns, from phoneme utterances to restaurant visits. In the first main part of the article, models of dynamical symbols offered by dynamical systems theory are reviewed: attractors, bifurcations, spatial segregation and boundary formation, and several others. In the second main part, the concept of transient attractor (TA) is offered as yet another mathematical model of dynamical symbols. TAs share with ordinary attractors a basic property, namely, local phase space contraction. However, a TA can disappear almost as soon as it is created, which could (not very rigorously) be interpreted as a bifurcation induced by quickly changing control parameters. Such ‘fast bifurcation sequences’ standardly occur in neural and cognitive dynamics.

The deficiency of entropy as a measure of disorder for systems whose number of available states can change and the advantage of a recently proposed measure, entropy normalized through the maximum possible entropy, are illustrated by way of simple examples. Results are presented relating the new measure to (1) the development of order in hierarchies, (2) fractal dimensions and (3) complexity. (1) Two categories of hierarchies are considered: in combinatorial hierarchies disorder can not increase, but in splitting hierarchies it may increase or decrease. (2) It is shown that the fractal dimension of Rényi order q is the product of the embedding dimension and the corresponding measure of disorder in the fine-grained limit. (3) A simple measure of complexity based on the measure for disorder is proposed; it may show the qualitative behavior of many previously proposed measures of complexity.

The heuristic value of the synergetic models of evolution and self-organization of complex systems in the field of epistemology is shown in this contribution. Two main synergetic notions are under consideration: the notion of the discreteness of evolutionary paths of complex systems and the notion of very fast evolutionary processes, so called blow-up regimes. Creative thinking and the functioning of creative intuition and productive imagination are viewed in the light of synergetics as a process of self-organization and of self-completion of images and thoughts, filling up gaps in the net of knowledge. Insight, fast and sudden solutions to scientific problems, instabilities when ‘an idea is in the air’, a boom of research in a certain scientific area are all interpreted as blow-up regimes in a cognitive field.

Cognitive psychology encounters various problems, such as the problem of intentionality, the symbol grounding and representation problems, and the problem of cognitive stability. These problems originate from the mind-body dichotomy; thus, foundational philosophical topics have quite concrete consequences for cognitive psychology and autonomous agents design. We argue that cognitive theory and modeling must account for stability, autonomy or completeness (i.e., should not rely on homunculi), and optimality. A synergetic model is proposed that has the desired properties: Cognition rests upon perception-action cycles. Cognition is viewed as ecological, i.e. it is situated by environmental affordances (valences) which signal energy resources. These cycles have an built-in optimality function because they maximize entropy production by dissipating affordances.

This paper considers how conventional theories of attitudes and the self may be reconceptualized from the perspective of chaos theory and work on self-organization. Within attitude theory, there has been a long tradition of research that has treated attitudes as single points on a bipolar evaluative continuum. More recent approaches to attitudes treat attitudes as structures of evaluative associations stored in memory (Fazio, 1990). This work can be linked to chaos theory by regarding attitudes as attractors (Eiser, 1994) within a phase space whose dimensions correspond broadly to features of the attitude object and its context and which is contoured by previous and concurrent associative learning Specifically, it is proposed that attractors are laid down through processes of association. such as may be simulated through connectionist, or parallel distributed processing, systems. Associative learning and memory processes also are implicated in our concept of self, linking with the philosophy of Hume and contemporary research on interactive models of personality (Mischel & Shoda, 1995). Speculatively, it is suggested that consciousness involves representation of the self within a patterned (i.e. self-organized) environment.

At variance with a purely cognitivistic approach, an affect-centered model of mental functioning called ‘fractal affect-logic’ is presented on the basis of current emotional-psychological and neurobiological research. Functionally integrated feeling-thinking-behaving programs generated by action appear in this model as the basic ‘building blocks’ of the psyche. Affects are understood as the essential source of energy that mobilises and organises both linear and nonlinear affective-cognitive dynamics, under the influence of appropriate control parameters and order parameters. Global patterns of affective-cognitive functioning form dissipative structures in the sense of Prigogine, with affect-specific attractors and repulsors, bifurcations, high sensitivity for initial conditions and a fractal overall structure that may be represented in a complex potential landscape of variable configuration. This concept opens new possibilities of understanding normal and pathological psychodynamics and sociodynamics, with numerous practical and theoretical implications.

The premise is the hypothesis that the task of schemata as cognitive process structures and their corresponding neuronal processes is to maintain aspects of the dynamical equilibrium of an organism. The formation and adaptation of these process structures in relation to the sensory input will be related formally to the evolution of self-organized process structures in physical systems far from thermodynamical equilibrium. Consequently, the significance of the optimisation principles that are involved in these processes will be emphasised, including the assumption that cognitive dynamics and the corresponding neuronal dynamics are defined through homologous principles. A functional model will be presented. This model permits a simulation of the self-organized formation and adaptation of schemata by the specific interaction of assimilation and accommodation processes. In this connection it will be shown that an adequate class formation can be realised by an application of the model to its output, i.e. the schemata. Finally, two empirical designs for the examination of the accommodation and the assimilation dynamics are introduced.

This chapter compares Piaget's theory of development with Feigenbaum & Simon's (1962; 1984) EPAM theory. An attempt is made to map the concepts of assimilation and accommodation in Piaget's theory onto the concepts of familiarisation and accommodation in EPAM. An EPAM-like model of the balance scale task is then presented, with a discussion of preliminary results showing how it accounts for children's discontinuous, stage-like development. The analysis focuses on the transition between rules, using catastrophe flags (Gilmore, 1981) as criteria. It is argued that some symbolic models may be described as dynamical systems, in the same way as some non-symbolic models.

We describe the embodied cognitive science approach by summarizing the current state-of-the-art in the field by a set of design principles for building autonomous agents. Autonomous agents are complete systems, i.e., they are autonomous, self-sufficient, embodied and situated. The goal is to extract and describe in a compact way a large part of the insights which have been acquired in the autonomous agents field. The principles have been developed from a cognitive science perspective. Although they represent only a very modest beginning, they make immediately clear what sort of ideas about intelligence and cognition they endorse. They all contrast sharply with classical thinking according to which intelligence is best explained by means of information processing, independent of physical implementation. Moreover, the principles provide powerful heuristics for design. The approach is illustrated by a number of sample projects. We conclude by raising issues concerning the evaluation of autonomous agents modeling.

Psychological experiments on children's development of spatial knowledge suggest experience at self-locomotion with visual tracking as important factors. Yet, the mechanism underlying development is unknown. We propose a robot that learns to mentally track a target object (i.e., maintaining a representation of an object's position when outside the field-of-view) as a model for spatial development. Mental tracking is considered as prediction of an object's position given the previous environmental state and motor commands, and the current environment state resulting from movement. Following Jordan & Rumelhart's (1992) forward modeling architecture the system consists of two components: an inverse model of sensory input to desired motor commands; and a forward model of motor commands to desired sensory input (goals). The robot was tested on the ‘three cups’ paradigm (where children are required to select the cup containing the hidden object under various movement conditions). Consistent with child development, without the capacity for self-locomotion the robot's errors are self-center based. When given the ability of self-locomotion the robot responds allocentrically.

This paper investigates the skills of autonomous agents to reorganize their society as answer to environmental changes. The reorganization of an agent society can be motivated by the desire to reduce conflicts within inter-agent cooperation and to increase the efficiency in achieving goals. Our interest is centered on situations where new agents want to join an existing agent society which has established conventions for agent cooperation. Joining an agent society means that the society can draw benefits from the interaction with the new members having competencies which are complementary to the society ones; both, the society and the new members, need to agree on a new convention. We introduce a method for reorganization where a society awards the agents and reinforces their roles which are beneficial for the society's survival.

This paper originates from my work on ‘social agents’. An issue which I consider important to this kind of research is the dynamic coupling of an agent with its social and non-social environment. I hypothesize ‘internal dynamics’ inside an agent as a basic step towards understanding. The paper therefore focuses on the internal and external dynamics which couple an agent to its environment. The issue of embodiment in animals and artifacts and its relation to ‘social dynamics’ is discussed first. I argue that embodiment is linked to a concept of a body and is not necessarily given when running a control program on robot hardware. I stress the individual characteristics of an embodied cognitive system, as well as its social embeddedness. I outline the framework of a physical-psychological state space which changes dynamically in a self-modifying way as a holistic approach towards embodied human and artificial cognition. This framework is meant to discuss internal and external dynamics of an embodied, natural or artificial agent. In order to stress the importance of a dynamic memory I introduce the concept of an ‘autobiographical agent’. The second part of the paper gives an example of the implementation of a physical agent, a robot, which is dynamically coupled to its environment by balancing on a seesaw. For the control of the robot a behavior-oriented approach using the dynamical systems metaphor is used. The problem is studied through building a complete and co-adapted robot-environment system. A seesaw which varies its orientation with one or two degrees of freedom is used as the artificial ‘habitat’. The problem of stabilizing the body axis by active motion on a seesaw is solved by using two inclination sensors and a parallel, behavior-oriented control architecture. Some experiments are described which demonstrate the exploitation of the dynamics of the robot-environment system.

Collective intelligence can be broadly defined as the study of global intelligent behaviour, arising out of local interactions within a collective and between the collective and its environment, in the absence of a central authority. A formal framework within which to situate models of collective intelligence is presented. Unique to the model is 1) the use of spatiotemporal patterns as the basic objects of interest, 2) the embodiment of agents within an embodied environment, 3) the implicit incorporation of the environment within the dynamics, 4) the use of a functional analytic setting, 5) the calculation of probability components via a path integral on a section of a dynamical bundle.

This paper discusses the application of catastrophe theory to human decision making, and describes the results of three experiments designed to demonstrate certain ‘catastrophe flags’ (Gilmore, 1981)—phenomena indicative of an underlying catastrophe surface. Clear evidence is presented for hysteresis (more often interpreted as framing effects) in a medical setting with a standard gamble technique, and for sudden jumps in consumer preferences. A problem with much experimental research is identified whereby analysis at the group level suggesting smooth changes in preferences obscures discontinuities at the level of the individual. We conclude that catastrophe theory offers a single unifying framework within which seemingly disparate theories about the nature of thinking and decision making may be integrated.

The Speech Cycling Task is a novel experimental paradigm developed together with Robert Port and Keiichi Tajima at Indiana University. In a task of this sort, subjects repeat a phrase containing multiple prominent, or stressed, syllables in time with an auditory metronome, which can be simple or complex. A phase-based collective variable is defined in the acoustic speech signal. This paper reports on two experiments using speech cycling which together reveal many of the hallmarks of hierarchically coupled oscillatory processes. The first experiment requires subjects to place the final stressed syllable of a small phrase at specified phases within the overall Phrase Repetition Cycle (PRC). It is clearly demonstrated that only three patterns, characterized by phases around 1/3, 1/2 or 2/3 are reliably produced, and these points are attractors for other target phases. The system is thus multistable, and the attractors correspond to stable couplings between the metrical foot and the PRC. A second experiment examines the behavior of these attractors at increased rates. Faster rates lead to mode jumps between attractors. Previous experiments have also illustrated hysteresis as the system moves from one mode to the next. The dynamical organization is particularly interesting from a modeling point of view, as there is no single part of the speech production system which cycles at the level of either the metrical foot or the phrase repetition cycle. That is, there is no continuous kinematic observable in the system. Nonetheless, there is strong evidence that the macroscopic behavior of the entire production system is correctly described as hierarchically coupled oscillators. There are many parallels between this organization and the forms of inter-limb coupling observed in locomotion and rhythmic manual tasks.

This paper aims to show possible refinements in time series methods for evaluating the dynamics of personality change. For the study. 13 students attended a course of personality development based on Jungian theory. The course teaches how to contact one's personal self. For four months the students rated their mood, activity, tension, and feeling of inner control on visual analogue scales twice a day. Standard examination with ARIMA models yield that most subjects show a low to moderate correlation to the previous timepoint. About one third of the cases have an additional lag2-relation. Daytime effects are seldom and the residual tests for the ARIMA models suggest that these linear models are sufficient in describing most of the time series. To evalute the expected smooth transformations in personality the data from one subject is analysed and the following hypotheses are empirically tested by the time-variation of parameters in subsequent time windows: 1) Increasing stability in mood and in the feeling of inner control by decreasing standard deviations 2) higher innerpsychic coherence by increasing autocorrelation coefficients 3) dissociation between mood and feeling of inner control by decreasing cross-correlation coefficients between these two dimensions. Application of several statistical tests shows that hypothesis 1 can be accepted while the other two hypotheses cannot be confirmed. Some methodological difficulties emerge when applied to ‘real’ data and some limitations are found in the statistical testing of time-varying parameters. Overall, though, the proposed methods for examining emotional variability have proven valuable and promising for further research.

This research explored the impact of organizational hierarchies on the dynamics of work performance and profitability. Theoretical expectations from game theory, synergetics, and other work on coupled nonlinear dynamical processes were reviewed and were found to make different predictions regarding the impact of hierarchical organization on behavior at any one level of the organization. University students participated in experimentally contrived organizations that had either two or three levels of hierarchy. The dynamics of work performance and profitability were assessed through nonlinear regression. Results showed that, for a two-level system, work performance was chaotic at both levels of organization, with a decrease in both noise and dimensionality at the upper level. For the three-level system, however, work performance at the lowest level showed gradual, non-asymptotic, and non-chaotic increases over time. The middle and upper management behaviors displayed both dampening and accelerating oscillatory pattern over time, an increase in dimensionality between workers and middle management, and a decrease in dimensionality from middle to top management. Nonlinear models outperformed linear alternatives by a margin upwards of 1.36:1 in variance accounted for.

The following sections are included:

• LIST OF CONTRIBUTORS

• AUTHOR INDEX

• SUBJECT INDEX