Eva Hudlicka

To feel or not to feel: the role of affect in human-computer interaction

The past decade has witnessed an unprecedented growth in user interface and human-computer interaction (HCI) technologies and methods. The synergy of technological and methodological progress on the one hand, and changing user expectations on the other, are contributing to a redefinition of the requirements for effective and desirable human-computer interaction. A key component of these emerging requirements, and of effective HCI in general, is the ability of these emerging systems to address user affect. The objective of this special issue is to provide an introduction to the emerging research area of affective HCI, some of the available methods and techniques, and representative systems and applications.

Affective game engines: motivation and requirements

The tremendous advances in gaming technologies over the past decade have focused primarily on the physical realism of the game environment and game characters, and the complexity and performance of game simulations and networking. However, current games are still lacking in the affective realism of the game characters, and the social complexity and realism of their interactions. To achieve the next leap in the level of engagement and effectiveness, particularly in the arena of serious games, gaming research needs to focus on enhancing the social and affective complexity and realism of the game characters, their interaction, and the game narrative as a whole. To achieve these goals, games and game development tools will need to provide functionality to support the recognition of user and game character emotions, real-time adaptation and appropriate responses to these emotions, and more realistic expression of emotions in game characters and user avatars. To support these functionalities, the games will need to construct affective models of the players, and include computational models of emotion within the game characters. In this paper, we discuss these functionalities, and suggest a set of requirements for an affective game engine, capable of supporting the development of more affectively realistic, engaging, and effective games. The discussion is organized around the functional requirements and the computational tasks necessary to support them. We emphasize the importance of selecting appropriate semantic primitives, and discuss how existing methods and techniques in affective computing and computational affective modeling contribute to the development of affective game engines and game development tools.

Computational Modeling of Emotion: Toward Improving the Inter- and Intradisciplinary Exchange

The past years have seen increasing cooperation between psychology and computer science in the field of computational modeling of emotion. However, to realize its potential, the exchange between the two disciplines, as well as the intradisciplinary coordination, should be further improved. We make three proposals for how this could be achieved. The proposals refer to: 1) systematizing and classifying the assumptions of psychological emotion theories; 2) formalizing emotion theories in implementation-independent formal languages (set theory, agent logics); and 3) modeling emotions using general cognitive architectures (such as Soar and ACT-R), general agent architectures (such as the BDI architecture) or general-purpose affective agent architectures. These proposals share two overarching themes. The first is a proposal for modularization: deconstruct emotion theories into basic assumptions; modularize architectures. The second is a proposal for unification and standardization: Translate different emotion theories into a common informal conceptual system or a formal language, or implement them in a common architecture.

This time with feeling: Integrated model of trait and state effects on cognition and behavior

Both stable personality characteristics (traits) and transient emotions (states) influence cognition and behavior. In this paper, we describe a methodology for modeling these effects in terms of a set of parameters that control processing within a symbolic cognitive architecture. The underlying thesis of the approach is that the combined effects of these individual differences can be modeled by varying the architecture parameters that control both processing and the structure of knowledge within the architecture modules. We describe the architecture, provide operational definitions of representative trait and state influences in terms of the controlling parameters, and demonstrate how observed trait/state phenomena are modeled in the context of the current demonstration scenario: a peacekeeping training simulation.

Social interaction with robots and agents: Where do we stand, where do we go?

Robots and agents are becoming increasingly prominent in everyday life, taking on a variety of roles, including helpers, coaches, and even social companions. A core requirement for these social agents is the ability to establish and maintain long-term trusting and engaging relationship with their human users. Much research has already been done on the prerequisites for these types of social agents and robots, in affective computing, social computing and affective HCI. A number of disciplines within psychology and the social sciences are also relevant, contributing theories, data and methods relevant for the emerging areas of social robotics, and social computing in general. However, the complexity of the task of designing these social agents, and the diversity of the relevant disciplines, can be overwhelming. This paper presents a summary of a special session at ACII 2009 whose purpose was to provide an overview of the state-of-the-art in social agents and robots, and to explore some of the fundamental questions regarding their development, and the evaluation of their effectiveness.

CHALLENGES IN DEVELOPING COMPUTATIONAL MODELS OF EMOTION AND CONSCIOUSNESS

There is a long-standing debate regarding the nature of the relationship between emotions and consciousness. Majority of existing computational models of emotions largely avoid the issue, and generally do not explicitly address distinctions between the conscious and the unconscious components of emotions. This paper highlights the importance of developing an adequately differentiated vocabulary describing the mental states of interest, and their features and components, for the development of computational models of the relationships between emotions and consciousness. We discuss current psychological theories of emotion, highlighting specific points in the affective processes where links exist with consciousness, and possible roles played by each. We discuss examples of models that are beginning to address components of the interface between emotions and consciousness: models of social emotions, requiring explicit representations of the self; models of affective biases on attention; models of emotion and metacognition; and models of emotion regulation. We conclude with a discussion of some of the challenges associated with modeling mental states whose core distinguishing characteristic is an awareness of affective feelings, and highlight the importance of integrating the diverse approaches to emotion research and modeling currently existing within psychology and neuroscience.

Response: is affective computing an oxymoron?

The responses are organized by indicating specific point being addressed, and indicating the corresponding section in the Hollnagel paper. Regarding nature of thinking (Introduction) But what is thinking? And what is a thinking machine and how would we recognize one? Does the Turing test still apply? Does thinking in fact imply the presence of consciousness, whatever THAT may be, or some other ineffable characteristic? A notorious problem for artificial intelligence (AI) is that once something is working (admittedly, very partially, in the case of thinking but indisputably so in some cases—e.g., expert systems, chess, pattern recognition), it is no longer AI. Perhaps AI is then not so much a set of particular problems, but rather the set of problems that has not yet been solved (and associated methods). If thinking is representing and manipulating information, then certainly many of our current machines can be described as thinking machines (chess, examples from this issue—Braezel’s emotional robot, etc.) Or perhaps what is missing from current machines in the minds of some researchers is precisely an affective—personality component, that would then make them truly thinking machines. Is perhaps ‘feeling’ a sine qua non of ‘thinking’? Certainly emotions appear to be critical for decisionmaking (see Hudlicka article in this issue). Regarding whether or not computing can be affective (Introduction) I agree with this, partially. If by affective we mean ‘able to feel emotions as humans (animals)’ then yes, computers most likely cannot be affective, because the essence of affective experience is the subjective, idiosyncratic, embodied ‘feeling’. Here one quickly enters the slippery slope of discussions about emergence of consciousness and may even enter the dangerous grounds of discussion of ambitious, ‘feeling’ robots attempting to control their human creators. I do not think that the vast majority of serious affective computing researchers are too concerned about

Approaches for modeling individuals within organizational simulations

The human behavior modeling community has traditionally been divided into those addressing individual behavior models, and those addressing organizational and team models. And yet it is clear that these extremes do not reflect the complex reality of the mutually-constraining interactions between an individual and his/her organizational environment. In this paper we argue that realistic models of organizations may require not only models of individual decision-makers, but also explicit models of a variety of individual differences influencing their decision-making and behavior (e.g., cognitive styles, personality traits, and affective states). Following a brief review of individual differences and cognitive architectures research, we describe two alternative approaches to modeling the individual within an organizational simulation: a cognitive architecture and a profile-based social network. We illustrate each approach with concrete examples from existing prototypes.

From Habits to Standards: Towards Systematic Design of Emotion Models and Affective Architectures

Emotion modeling has been an active area of research for almost two decades now. Yet in spite of the growing and diverse body of work, designing and developing emotion models remains an art, with few standards and systematic guidelines available to guide the design process, and to validate the resulting models. In this introduction I first summarize some of the existing work attempting to establish more systematic approaches to affective modeling, and highlight the specific contributions to this effort discussed in the papers in this volume. I then propose an analytical computational framework that delineates the core affective processes, emotion generation and emotion effects, and defines the abstract computational tasks necessary to implement these. This framework provides both a common vocabulary for describing the computational requirements for affective modeling, and proposes the building blocks necessary for implementing emotion models. As such, it can serve both as a foundation for developing more systematic guidelines for model design, and as a basis for developing modeling tools. I conclude with a summary and a discussion of some open questions and challenges.

Modeling Cultural and Personality Biases in Decision-Making

Cultural, personality and affective biases in decision-making are well documented. This chapter describes a method for modeling multiple decision biases resulting from cultural effects, personality traits and affective states, within the context of a symbolic cognitive-affective agent architecture: the MAMID methodology and architecture. The approach emphasizes the role of affect in decision-biases, as the primary mediating factor of a wide range of biasing effects, and lends itself to exploring alternative mechanisms mediating a wide range of decision biases. The approach provides a uniform framework for modeling both content and processing biases, in terms of parameter vectors that control processing within the architecture modules. The effects of these biases are encoded in specific values of architecture parameters, which then influence the processing of the distinct architecture modules, including the architecture topology itself. The associated simulation environment enables the modeling of a wide variety of decision-makers, in terms of distinct personality and cultural profiles, and consequent affective profile and affect-induced decision-biases. The key contribution, and distinguishing feature, of the MAMID modeling approach is the parameter space it provides for representing the interacting effects of multiple types and sources of biases, and the potential of this approach for modeling the fundamental mechanisms that mediate decision-biases.