Emotion in Cognition and Action: WP7 Exemplars

1. Emotion in Cognition and Action:WP7 Exemplars WP4 Workshop Santorini, September 18-21, 2004 Lola Cañamero University of Hertfordshire, UK WP7: Emotion in Cognition and Action

2. The Task of WP7 • WP7 task (Exemplar): • Shared understanding of mechanisms underlying the involvement of emotion in cognition and action (decision making) • Comparative approaches to emotion-oriented agent architectures: assumptions, guidelines, and integration challenges • Working groups within this task: • WG1: Emotion in “lower-lever” cognition and action • UH, GERG, OFAI (KCL?) • WG2: Emotion in “higher-level” cognition and action • Uni. Bari, France Telecom RD, QUB, EPFL, GERG • WG3: Bridging gap between “lower-” and “higher-level” cognition & action • OFAI, USAL, INESC-ID, IST • WG4: Emotion in Social Cognition and Interaction • UH, Paris8, DFKI, MIRALab, OFAI, DIST WP7: Emotion in Cognition and Action

3. WG1: Emotion in “lower-level” C&A • “fast pathway”; Embodied AI approach (Perception-Action models) • Mechanisms underlying emotional modulation of cognition & action: • Neuromodulation of cognition and action (of basic robotic architectures) • Incremental approach -> increasing complexity and flexibility • Bottom-up approach -> how emotions “emerge” • Critical analysis from psychology, neuroscience, philosophy • Principled design of “emotivated” architecture for action selection • “Modulated” robot architecture (UH), analysis in terms of appraisal • Feedback from appraisal theory to add relevant elements; e.g. what triggers “neuromodulations”? nature of stimuli, what is relevant, how to process them • Focus on perception -> from signals to signs of emotion in robots? (WP4) • Integration of paradigms (cybernetics, dynamical systems, ethology, neuromodulation, appraisal) WP7: Emotion in Cognition and Action

4. WG2: Emotion in “higher-level” C&A • “slow pathway”; Symbolic AI approach (cognitive BDI&E models) • Evaluation of cognitive models of emotion activation • BDI models of event-based emotions (OCC) that might be activated in a decision-making process (e.g. car driving, Haptek agent player) • Evaluating them in terms of sensitivity to individual parameters and of uncertainty in the probability of emotion activation in a given condition • Extension to ‘HCI emotions’ (anxiety, frustration, satisfaction, etc); + WP3 • Recognition and interpretation of emotional states • Integrating recognition of emotion (recognition methods; WP4) + (cognitive models for) interpretation of its causes • Recognize emotional state of users in natural language interaction (corpus of ‘natural’ dialogs collected through Wizard of Oz tool; WP5, WP9) • Build model of interlocutor to interpret their possible causes (incrementally) WP7: Emotion in Cognition and Action

5. WG3: Bridging the gap • The gap: • Behavioral architectures (embodied agents): • Good for low-level, pre-attentive accounts of emotion, fast reactions • BUT reactive -> subject to problems of local decision-making • Symbolic AI architectures (rule-based systems): • Good for high-level appraisal-based accounts of emotion (planning, rich expectations and action sequences into the future) • BUT problems with sensor- and environment-directed (multi-task) processing, brittle and unreceptive to changes in the world • Task of WG: integrate both types of architectures using a model of affect • Proposed work: motivation-driven continuous planner (hybrid) architecture, extension of TABASCO • NOT another hybrid, multi-layer architecture but critical analysis of integration challenges WP7: Emotion in Cognition and Action

6. WG4: Social cognition and interaction Two goals: • Critical understanding of (selected) key mechanisms and key issues to take into account social aspects in emotion-oriented agent architectures • Guidelines for integration of emotion-oriented architecture (WP7) & expressive behaviors (WP6) • Moods and appraisal in social interaction: DFKI • Appraisal-based model that generates both emotions and moods based on a configurable, subjective appraisal component, in multiple animated conversational characters • Compute social aspects for interpersonal relationships, longer lasting affective state: overall mood (‘morale’ vector) of group, social integrity, ‘mood groups’ • Emotion in social dynamics (agent-based simulations) • Mutual support role of emotions and social norms: OFAI • Emergence of groups with different dynamics and structure: UH WP7: Emotion in Cognition and Action

7. WG4: Social cognition and interaction • Automatic and low-level mechanisms in social interaction: • Perception-Action model to study the role of affect in imitation of movements, emotional contagion (synchrony, mimicry) using robots: UH • Artistic framework: synchronization and coordination of movements in dance (humans and robots): DIST • Social attention and memory: Paris8 • General environmental attention mechanism for early social interaction; e.g. visual attention towards faces, other features (change, movement?) WP4 • Memory: autobiographical, social • Generation of socially meaningful emotional behavior: MIRALab • Standards for translating a cognitive process into body behavior • Definition of basic “units” that control body behavior • Take advantage of animation systems that result in realistic body animations WP7: Emotion in Cognition and Action

8. Interaction with WP4 • WP4 to WP7: • Signs of emotion in different modalities for different types of architectures and platforms • WG1: signals to signs of emotions for robots (e.g. robotic “biosensors”) • WG2: emotion recognition methods for symbolic processing • WG3: signs of emotion for “fast” and “slow” processing simultaneously • WG4: focus on movement; faces? • WP7 to WP4: • Specifications on constraints imposed by different architectures and platforms for the interpretation / processing of signs of emotion • Generalization of Markov analysis from ethology & dynamical systems theory to data analysis (non-linearity problems); complementing it with information theory => Rene WP7: Emotion in Cognition and Action

9. Guidelines for exemplars • Informed by critical analysis of our existing models / systems / tools • Guided by analysis of needs (think ahead, goals, “visions”) • Exemplars / Exemplar: “modules” under common “umbrella” • Theoretical + pragmatic criteria, e.g. meaningful publications • Integration effort: • Within “modules”: across areas / problems / disciplines • Across “modules”: self-critical + think positive • Our scope, our limitations, what we intend to and can / cannot achieve • What others can do that I can’t, insights / know-how I could gain • What I could contribute to other “modules” and approaches • Links with other WPs WP7: Emotion in Cognition and Action

10. What? Goals & Tasks of WP 7 • Goals: • Achieve shared understanding of basic issues and basic research topics • Set the grounds to advance the state of the art • Disseminate the achieved research results and conclusions • Tasks: • Task 1: Integration of existing know-how • Task 2: Plans to advance the state of the art • Task 3: Dissemination of gained insight WP7: Emotion in Cognition and Action

11. Integration effort? • Shed light towards development of sound computational models that: • Enhance behavior & interactions of emotion-oriented systems • Feed back to emotion theorists (synthetic approach) • Integrated effort (≠ IP!!), shared critical reflection about: • Key conceptual issues • Assumptions and dependencies • Open research problems • Key research and application scenarios • Critical analysis of existing approaches, systems and tools -> • Analysis of needs and directions for future research and applications • Evaluation methods, scenarios, tools • Recommendations for good practice, usability requirements • Potential for cross-fertilization among disciplines WP7: Emotion in Cognition and Action