Advances in Cognitive Systems
2: Control Systems Centre, University of Manchester, UK
This book has been inspired by the portfolio of recent scientific outputs from a range of European and national research initiatives in cognitive science. It presents an overview of recent developments in cognition research and unites the various emerging research strands within a single text as a reference point for progress in the subject. It also provides guidance for new researchers on the breadth of the field and the interconnections between the various research strands identified here. Advances in Cognitive Systems brings together a wide range of material from leading workers in the field as well as the outputs from research groups around the world, covering the two principal cognition paradigms of cognitivism and emergence. Furthermore, it suggests some interesting lines of thought about the challenges and opportunities for future work in the field and for promoting the various research agendas highlighted within. This compilation will be of interest not only to those working in the fields of computer science and AI but also to psychologists, neural scientists, researchers in information science, and engineers involved in the development of advanced robotics, mechatronic systems and HCI systems.
Inspec keywords: artificial intelligence; cognitive systems
Other keywords: cognitive robotics system; adaptive systems; neurocomputational cognitive architectures; cooperative decision model; dynamical enactive cognitive architecture; distributive cognition; investor behavior; virtual humans; neural networks; swarm intelligence techniques; robotic agents
Subjects: Artificial intelligence (theory); Robotics; Neural computing techniques; Expert systems and other AI software and techniques; Virtual reality
- Book DOI: 10.1049/PBCE071E
- Chapter DOI: 10.1049/PBCE071E
- ISBN: 9781849190756
- e-ISBN: 9781849190763
- Page count: 528
- Format: PDF
-
Front Matter
- + Show details - Hide details
-
p.
(1)
-
1 Embodiment in cognitive systems: on the mutual dependence of cognition and robotics
- + Show details - Hide details
-
p.
1
–12
(12)
Cognitive systems anticipate, assimilate, and adapt. They develop and learn, predict, explain, and imagine. In this chapter, we look briefly at the two principal paradigms of cognition, cognitivism and emergence, to determine what embodied form each entails, if any. We highlight one specific emergent approach to embodied cognition enaction and discuss the challenges it poses for the advancement of both robotics and cognition.
-
2 Intelligence, the interaction of brain, body and environment design principles for adaptive systems
- + Show details - Hide details
-
p.
13
–36
(24)
There have been two ways of approaching intelligence: the traditional approach, where the focus has been on the study of the control or the neural system itself and a more recent approach that is centered around the notion of 'embodiment,' the idea that intelligence always requires a body, a complete organism that interacts with the real world. This chapter explores the deeper and more important consequences, concerned with connecting brain, body, and environment, or more generally with the relation between physical and information (neural, control) processes. Often, morphology and materials can take over some of the functions normally attributed to the brain (or the control), a phenomenon called 'morphological computation.' A number of case studies are presented to illustrate how 'morphological computation' can be exploited for designing intelligent, adaptive robotic systems, and to understand natural systems. We conclude with a theoretical scheme that can be used to embed the diverse case studies and that captures the essence of embodiment and morphological computation.
-
3 Compliant actuation: enhancing the interaction ability of cognitive robotics systems
- + Show details - Hide details
-
p.
37
–56
(20)
The growth of cognitive capabilities in biological systems and particularly in humans is assisted by the extensive and safe interaction of their bodies with their living environment. This ability to interact safely with the environment is mainly due to a set of inherent behaviours/properties that allow the biological systems to intelligently react to the interaction at the body/actuation level with minimum active control. One fundamental intrinsic characteristic of the bio logical systems is the compliance. Looking at the diverse range of actuation techniques, the natural world appears to have developed a much more ubiquitous actuation design, with the organic muscle that provides power to animals ranging from the largest whales to microbes with adaptation to cope with environmental extremes. The potential benefits, which can be gained in any mechanism with the incorporation of biologically inspired compliant actuation concepts, are well known. However, the majority of today's robots lack these characteristics. This chapter reviews some of the solution adopted to implement the compliant behaviour in robotic systems and introduces a case study of a new compact soft actuation unit intended to be used in multi-degree of freedom and small-scale robotic systems such as the cognitive child humanoid robot 'iCub'.
-
4 Toward a dynamical enactive cognitive architecture
- + Show details - Hide details
-
p.
57
–86
(30)
The increasing complexity of humanoid robots and their expected performance in dynamic and unconstrained environments demand an equally complex, autonomous, adaptable, and dynamic solution. In this article, motivated by the enactive paradigm of cognitive systems, we develop the first steps toward the creation of such a solution. Enactive approaches assert that the primary basis of cognition is anticipative skill construction by embodied systems. The system's ontogenetic development is central to this approach since it is the system's experience - its dynamics and interaction with its environment - that defines its cognitive understanding of the world in which it is embedded rather than any a priori rules or representations of the system designer. Enactive systems are typically realized by dynamical systems theory, connectionism, and self-organization. In this article, we survey briefly the main attributes of nonlinear dynamical systems and then introduce Dynamic Field Theory (DFT) - a mathematical framework grounded in dynamical systems and neurophysiology - as a plausible way of combining the power of nonlinear dynamical systems theory and the autonomous developmental aspects of enactive systems. The potential of this approach is demonstrated by replicating in an anthropomorphic robotic platform the responses found in 7- to 12-month-old infants when exhibiting the A-not-B error, a classic simple cognitive behavior in developmental psychology.
-
5 A computational model of object affordances
- + Show details - Hide details
-
p.
87
–126
(40)
The concept of object affordances describes the possible ways whereby an agent (either biological or artificial) can act upon an object. By observing the effects of actions on objects with certain properties, the agent can acquire an internal representation of the way the world functions with respect to its own motor and perceptual skills. Thus, affordances encode knowledge about the relationships between action and effects lying at the core of high-level cognitive skills such as planning, recognition, prediction and imitation. Humans learn and exploit object affordances through their entire lifespan, by either autonomous exploration of the world or social interaction. Building on a biological motivation and aiming at the development of adaptive robotic systems, we propose a computational model capable of encoding object affordances during exploratory learning trials. We represent this knowledge as a Bayesian network and rely on statistical learning and inference methods to generate and explore the network, efficiently dealing with uncertainty, redundancy and irrelevant information. The affordance model serves as base for an imitation learning framework, which exploits the recognition and planning capabilities to learn new tasks from demonstrations. We show the application of our model in a real-world task in which a humanoid robot interacts with objects, uses the acquired knowledge and learns from demonstrations. Results illustrate the success of our approach in learning object affordances and generating complex cognitive behavior.
-
6 Helping robots imitate: metrics and technological solutions inspired by human behaviour
- + Show details - Hide details
-
p.
127
–168
(42)
In this chapter we describe three lines of research related to the issue of helping robots imitate people. These studies are based on observed human behaviour, technical metrics and implemented technical solutions. The three lines of research are: (a) a number of user studies that show how humans naturally tend to demonstrate a task for a robot to learn, (b) a formal approach to tackle the problem of what a robot should imitate and (c) a technology driven conceptual framework and technique, inspired by social learning theories, that addresses how a robot can be taught. In this merging exercise we will try to propose a way through this problem space, towards the design of a human-robot interaction (HRI) system able to be taught by humans via demonstration.
-
7 Models for cooperative decisions in Prisoner's Dilemma
- + Show details - Hide details
-
p.
169
–210
(42)
The Prisoner's Dilemma (PD) game is a two-person game that is widely used as a model of social interactions. The structure of the PD game represents a dilemma between the individual and the collective rationality. The interest in studying PD game arises from the idea that many social situations and problems such as overpopulation, pollution, energy savings, etc. have such a dilemma structure. PD games are specifically used as a tool for studying cooperative behavior.
-
8 Virtual humans made simple
- + Show details - Hide details
-
p.
211
–248
(38)
Virtual humans, or intelligent virtual agents, born from the mould of academic, industrial, and science-fiction communities, have been surrounding us for more than a decade. Theoretical knowledge allowing many practitioners and enthusiasts to build their own agents is now available. However, the initial learning curve imposed by this specialized knowledge is quite steep. Even more fundamentally, there is a lack of freely available software assisting in the actual building of these agents. In this chapter, we introduce the Pogamut 3 toolkit. Pogamut 3 is a freeware platform for building the behaviour of virtual humans embodied in a 3D virtual world. It is designed for two purposes: first, to help newcomers to build their first virtual agents; second, to support them as they advance in their understanding of the topic. On the one hand, Pogamut 3 allows for rapid building of simple reactive agents and on the other hand, it facilitates development of challenging agents exploiting advanced artificial intelligence techniques. Pogamut 3 is primarily tailored to the environment of the Unreal Tournament 2004 video game, but it can be connected to other virtual world as well. It is suitable both for research and educational projects.
-
9 More from the body: embodied anticipation for swift readaptation in neurocomputational cognitive architectures for robotic agents
- + Show details - Hide details
-
p.
249
–270
(22)
The coupling between a body (in an extended sense that encompasses both neural and non-neural dynamics) and its environment is here conceived as a critical substrate for cognition. We propose and discuss the plan for a neurocomputational cognitive architecture for robotic agents, so far implemented in its minimal form for supporting the behavior of a simple simulated robotic agent. A non neural internal bodily mechanism (crucially characterized by a timescale much slower than the normal sensory-motor interactions of the robot with its environment) extends the cognitive potential of a system composed of purely reactive parts with a dynamic action selection mechanism and the capacity to integrate information over time. The same non-neural mechanism is the foundation for a novel, minimalist anticipatory architecture, implementing our bodily anticipation hypothesis and capable of swift readaptation to related yet novel tasks.
-
10 The strategic level and the tactical level of behaviour
- + Show details - Hide details
-
p.
271
–300
(30)
We introduce the distinction between a strategic (or motivational) level of behaviour, where different motivations compete with each other for the control of the behaviour of the organism, and a tactical (or cognitive) level, where the organism executes the activities aimed at reaching the goal decided at the strategic level. To illustrate and operationalise this distinction we describe three sets of simulations with artificial organisms that evolve in an environment in which in order to survive they need either to eat and drink, or to eat and avoid being killed by a predator. The simulations address some simple aspects linked to the strategic level of behaviour, i.e. the role played by the environment in determining what are the motivations driving an organism and what is the strength of each of these motivations. Other phenomena investigated are the usefulness for the organism's brain to receive information from its own body (e.g., in the form of hunger or thirst), how inter-individual differences among individual organisms may concern both the strategic and the tactical level of behaviour, and how the unsolved competition between very strong motivations can lead to pathological states such as depression.
-
11 Sequencing embodied gestures in speech
- + Show details - Hide details
-
p.
301
–332
(32)
The embodied character of cognitive motor systems that has greatly influenced the understanding of their constitution and function is reflected in many recent models. Embodiment conditions that system behaviours must take appropriate account of energy expenditure and metabolic costs that are unavoidable in a physically realised medium. We here consider that optimisation, presumed to result from both phylogenetic and ontogenetic processes, can be used to con strain the space of potential degrees of freedom of a system, ensuring that the resulting action is efficient and smooth. To understand the emerging adaptations, it is necessary to factor in the properties of the physical and physiological substrate that anchor the system's goal-oriented performance. However, the embodied nature of speech production has been disregarded by most phonological research to date. This leads to a failure in providing a coherent phonological grounding of a wide range of phenomena extensively documented by experimental phoneticians, in particular those associated with the relative timing of gestures (also called gestural phasing) in connected speech and its variability as found in different manners of speech. Existing phonological theories of sequencing rely on essentially external system-wide rules and principles or explicit dynamical constraints governing phasing to account for various suprasegmental properties and prosodic parameters of an utterance. We introduce here a new and highly abstract modelling platform developed to investigate the embodied character of speech. The physically instantiated, second-order dynamic nature of the system allows us to define and exactly evaluate various cost functions, which we hypothesise to play a role in efficient gestural sequencing. We investigate the general dynamical properties of the system, and identify a set of its high-level, intentional parameters linked to the cost functions associated with its goal-oriented performance. We show that the phenomena accompanying gestural sequencing, coarticulation, fluency and prosodic modulation emerge as consequences of a non-trivially formulated efficiency constraint, thus providing a principled phonological account of phonetic reality.
-
12 A neural network model for the prediction of musical emotions
- + Show details - Hide details
-
p.
333
–370
(38)
This chapter presents a novel methodology to analyse the dynamics of emotional responses to music in terms of computational representations of perceptual processes (psychoacoustic features) and self-perception of physiological activation (peripheral feedback). The approach consists of a computational investigation of musical emotions based on spatio-temporal neural networks sensitive to structural aspects of music. We present two computational studies based on connectionist network models that predict human subjective feelings of emotion. The first study uses six basic psychoacoustic dimensions extracted from the music pieces as predictors of the emotional response. The second computational study evaluates the additional contribution of physiological arousal to the subjective feeling of emotion. Both studies are backed up by experimental data. A detailed analysis of the simulation models' results demonstrates that a significant part of the listener's affective response can be predicted from a set of psychoacoustic features of sound tempo, loudness, multiplicity (texture), power spectrum centroid (mean pitch), sharpness (timbre) and mean STFT flux (pitch variation) and one physiological cue, heart rate. This work provides a new methodology to the field of music and emotion research based on combinations of computational and experimental work, which aid the analysis of emotional responses to music, while offering a platform for the abstract representation of those complex relationships.
-
13 A conceptual model of investor behavior
- + Show details - Hide details
-
p.
371
–396
(26)
Behavioral finance is a subdiscipline of finance that uses insights from cognitive and social psychology to enrich our knowledge of how investors make their financial decisions. Agent-based artificial financial markets are bottom-up models of financial markets that start from the micro level of individual investor behavior and map it into the macro level of aggregate market phenomena. It has been recognized in the literature, yet not fully explored, that such agent-based models are very suitable tool to generate or test various behavioral hypotheses. To pursue this research idea, first we develop a conceptual model of individual investor that consists of a cognitive model of the investor and a description of the investment environment. In the modeling tradition of cognitive science and intelligent systems, the investor is seen as learning, adapting, and evolving entity that perceives the environment, processes information, acts upon it, and updates its internal states. This conceptual model can be used to build stylized representations of (classes of) individual investors, and further studied within the paradigm of agent-based artificial financial markets.
-
14 Decision making under risk in swarm intelligence techniques
- + Show details - Hide details
-
p.
397
–414
(18)
The term Swarm intelligence is associated with the decision processes that determine the overall group behavior of a collection of individuals or sets of simple agents cooperating to achieve some purposeful objective or common goal. However, the usual logical decision processes used in the literature to model individual agent behavior are generally found to be inadequate when the phenomena of uncertainty and risk are factored into process; and these models are usually incapable of fully emulating actual human decision-making behaviors under risk and uncertainty. This chapter proposes a significant modification to agent reasoning processes employed so far in conventional swarm intelligence techniques. We show that by endowing each agent with some descriptive behavior from the field of psychology named Prospect Theory (PT), we can improve considerably the efficiency of global searching procedures. The efficacy of the technique is illustrated by numerical results obtained from applications to two classical problems quoted in the literature.
-
15 Towards a cognitive model of interaction with notations
- + Show details - Hide details
-
p.
415
–434
(20)
The cognitive dimensions framework provides a mechanism by which we may better understand the factors which influence human interaction with notational systems. Primarily used to inform notational design, or to evaluate programming languages, interaction environments and specification notations, the focus of the framework is in highlighting the things that make using notations for a given task 'hard' or 'easy'. We believe that the dimensions, which have their roots in cognitive psychology, may offer the foundation for a cognitive model of interaction with notations. This work examines the existing body of literature that seeks to understand the cognitive features of human interaction with notations, noting the parallels with the cognitive dimensions framework. We suggest that we may derive a cognitive model of interaction with notations through examination of the dimensions using cognition as a lens. The work provides a contribution to the cognitive dimensions framework itself, highlighting the distinction between cognitive effort and constraints arising at the notational level. It also suggests that interaction with symbols and notations may be seen as a type of embodiment, and we believe that development of a cognitive model of interaction with notations may be of use in better understanding than embodiment.
-
16 How to choose adjudication function in distributive cognition?
- + Show details - Hide details
-
p.
435
–464
(30)
Modular redundancy, in which a function is performed by multiple, redundant agents machines or humans requires an 'adjudication' stage to produce a single 'opinion' out of the multiple ones produced. Commonly used adjudication functions are, e.g. majority voting and averaging of the multiple results, but many alternatives are possible and in actual use. The adjudication problem has also emerged as a key part in distributed cognition systems where the decision making process is inferred from a set of multiple agents. The choice of an adjudication function is of paramount importance for the overall dependability of a redundant system. This problem has been addressed separately by scholars in different disciplines, e.g. in the normative study of group decision making and in computer design. Each discipline has typically addressed a specific set of scenarios, with its own assumptions, which we believe would be of interest in other disciplines as well. All can benefit from taking into account results developed in other disciplines, and seeing their separate sets of assumptions and results in a common framework. This chapter explains the adjudication problem, investigates the different criteria that help the designer in choosing an appropriate adjudication function and proves some interesting properties that characterize an optimal adjudication function, given a specific utility function and failure hypothesis for the system in which it is to be used.
-
17 Diversity in cognitive models
- + Show details - Hide details
-
p.
465
–482
(18)
This final chapter looks back and forward. It looks back over the other chapters, to trace a 'thread' that runs through the work as a whole, which can suggest new and interesting avenues for future work. A very obvious theme argued in several chapters is that cognition is developed by interaction with the environment, especially of an embodied kind. Rather taken for granted, but no less important, is the diversity of these (embodied) interactions between the agent and its domain of application. It is diversity as such which this chapter traces as the thread that runs through the entire work, to point a way forward. Diversity may be understood by reference to Dooyeweerd's philosophical exploration of distinct spheres of meaning and law which, as aspects, enable us to differentiate types of interaction and embodiment. Plotting the chapters by these aspects reveals interesting patterns, which allow a number of suggestions to be made for future research, including focussing on under-represented aspects, providing agents with good quality meta-rules that express the distinct rationality of each aspect, and contributing to meeting a number of challenges. It concludes by suggesting that, in addition to contributions this work makes to the field of cognitive systems and the paradigm of embodied interaction, it can also contribute an understanding of diversity itself.
-
Back Matter
- + Show details - Hide details
-
p.
483
(1)