Affective Computing: Agents With Emotion

1
Affective Computing Agents With Emotion
Victor C. Hung University of Central Florida
Orlando, FL EEL6938 Special Topics in
Autonomous Agents March 29, 2007
2
Agenda

• Introduction
• Highlighted Projects
• Affective Cognitive Learning Decision Making
• Questions

3
Introduction

• Affective Computing relates to, arises from, or
  deliberately influences emotion or other
  affective phenomena
• Engineering, computer science with psychology,
  cognitive science, neuroscience, sociology,
  education, psychophysiology, ethics
• Emotion is fundamental to human experience
• Cognition
• Perception
• Learning
• Communication
• Rational decision-making

4
Introduction

• Technologists have largely ignored emotion
• Affect has been misunderstood
• Hard to measure
• MIT Media Lab Affective Computing
• http//affect.media.mit.edu
• Develop new technologies and theories
• Understanding affect and its role in human
  experience
• Restore a proper balance between emotion and
  cognition in the design of technologies for
  addressing human needs

5
Introduction

• Issues in affective computing
• Communication of affective-cognitive states to
  machines
• Techniques to assess frustration, stress, and
  mood indirectly
• Make computers can be more emotionally
  intelligent
• Personal technologies for improving
  self-awareness of affective states
• Emotions influences personal health
• Ethics

6
Highlighted Projects

• Affective-Cognitive Framework for Machine
  Learning and Decision-Making
• Emotions role in learning and decision making
• Digital Story Explication as it Relates to
  Emotional Needs and Learning
• Emotional interaction in child learning
• ESP - The Emotional-Social Intelligence
  Prosthesis
• Aid for the emotionally-impaired

7
Highlighted Projects

• Fostering Affect Awareness and Regulation in
  Learning
• Combat frustration during the learning process
• Machine Learning and Pattern Recognition with
  Multiple Modalities
• Emotional sensor data fusion
• Ripley A Conversational Robot
• Human-robot interaction platform through language
  and visual perception modalities

8
Affective-Cognitive Learning Decision Making

• (2006) Ahn and Picards Affective-Cognitive
  Learning and Decision Making The Role of
  Emotions, The 18th European Meeting on
  Cybernetics and Systems Research
• Framework for learning and decision making
• Inspired by neural basis of motivations and the
  role of emotions in human behavior
• Affective biases
• Loss aversion
• Effect of mood on decision making

9
Affective-Cognitive Learning Decision Making

• Affective biases
• Two-armed bandit

10
Affective-Cognitive Learning Decision Making

• Loss aversion
• Prefer avoiding losses than acquiring gains

11
Affective-Cognitive Learning Decision Making

• Effect of mood on decision making

ANGER Optimism about the future
HAPPINESS Optimism about the present
Pessimism about the future FEAR
Pessimism about the present SADNESS
12
Affective-Cognitive Learning Decision Making

• A motivational value (reward)-based learning
  theory
• Extrinsic value from the cognitive (deliberative
  and analytic) systems
• Intrinsic value from multiple affective systems
  such as Seeking (Wanting), Fear, Rage, and other
  circuits
• Probabilistic models
• Cognition (cognitive state transition)
• Multiple affect circuits (Seeking, Joy, Anger,
  Fear, ...)
• Decision making model
• Previous knowledge can be incorporated for
  expecting the consequences of decisions (or
  computing the cognitive value)

13
Affective-Cognitive Learning Decision Making
14
Affective-Cognitive Learning Decision Making

• The Decision-Making Model
• Cognitive state (c)
• Affective state (a)
• Decision (d)

15
Affective-Cognitive Learning Decision Making

• Affective seeking value
• Valence decided by the mean of the filtered
  values for the reward samples
• Arousal uncertainty of the reward sample
  distribution (modeled as standard deviation)
• Complete decision-making expression
• Non-affect agent has only the cognitive component

16
Affective-Cognitive Learning Decision Making

• Affective agent vs. Non-affect agent

17
Affective-Cognitive Learning Decision Making

• Influence of an outlier on the cognitive values
  and the valence values

18
Affective-Cognitive Learning Decision Making

• Affective component less sensitive to outliers
  than cognitive component
• Affective Cooling Agreement between two
  components
• More likely to follow the decision by the
  cognitive component (Exploitation)
• Value of the induced inverse temperature
  parameter increases
• Humans using cognition in decision-making
• Affective Heating Conflict between two
  components
• Less likely to follow the decision by the
  cognitive component (Exploration)
• Value of the induced inverse temperature
  parameter decreases
• Humans depending on emotion in decision-making

19
Affective-Cognitive Learning Decision Making

• 10-armed bandit tasks

20
Affective-Cognitive Learning Decision Making

• Too much or too little affect impairs learning
• Excessive learns faster, but not good for
  long-term
• Insufficient better for long-term, but slow

21
Affective-Cognitive Learning Decision Making

• Results and Conclusions
• Framework enhancements
• Model other affect circuits
• Incidental influences on decision making
• Use of prior knowledge for expecting cognitive
  outcomes ?
• Affective bias
• Helps automatically regulate exploration and
  exploitation
• Speed up learning without sacrificing decision
  quality
• This framework might mimic well-studied human
  behavior
• Risk aversion
• Effects of mood on decision making
• Self-control

22
Questions?