Emotion as a Layered Control System

1
Emotion as a Layered Control System

• Nancy Alvarado

2
What Does Emotion Do for Humans?

• It is part of what it means to be human.
• It makes life worth living by giving value to
  experiences.
• It permits us to respond flexibly to our
  environment, avoiding bad, approaching good.
• It coordinates brain, body behavioral
  responses.
• It guides inter-personal relationships.
• Emotion is necessary to cognitive development
  because adult-child interaction is facilitated
  (Kismet).

3
Emotion Consciousness

• Emotion is not strictly a phenomenon of
  consciousness, although its self-report is.
• Implicit vs explicit emotions
• Effects on motivation, memory, decision, other
  cognition can be set aside when emotion is
  consciously experienced.
• Moods vs emotions
• Long-lasting vs short duration
• Non-conscious vs conscious
• Unattributed vs attributed cause

4
Access vs Phenomenal Consciousness

• Without phenomenal consciousness, emotion has no
  effectiveness as a motivator of human behavior.
• We can prove that phenomenal consciousness exists
  by observing its impact on behavior.
• Access consciousness appears to be optional.
• But perceptual accounts focus mostly on access
  consciousness not phenomenal. Emotion works the
  opposite way.

5
Computer Functions of Emotion

• Coordinate internal system-wide responses to an
  environmental stimulus.
• Bias values of parameters.
• Identify goal-relevant outcomes and flexibly
  select responses to them.
• Prioritize among competing processes.
• Evaluate and communicate internal states to
  users.

6
Affect-Guided Cognition

• Barnes Thagards DECO system emotion-guided
  decision making
• An affective version of Newells SOAR
  architecture, used in military simulations
  Jones, Chown Henninger
• Moffat Frijda WILL, personality-based
  autonomous agents
• Kismets behavioral choice system

7
Slomans Distinction

• Shallow implementations
• Behaviors that simulate affect
• Simple links between triggers and behaviors
• Use of emotion term labels for variables
• Deeper implementations
• Emotion is an integral part of a theory of mind
• Emotion influences cognition at multiple levels
• Guided by functionality

8
Zombies (Robots)

• Some claim that it is the ability to feel emotion
  that makes us uniquely human.
• Rodney Brooks, representationalist, disagrees

http//www.aaai.org/AITopics/html/show.html http/
/news.bbc.co.uk/olmedia/cta/progs/02/hardtalk/broo
ks19aug.ram http//alicebot.org/
9
Kismet
Cynthia Breazeal (MIT) and her sociable robot
affect motivates and guides social learning.
Breazeal, C. (2002). Designing Sociable Machines,
MIT Press
10
Joshua Blue (IBM)

• Goal to develop common sense reasoning and
  human-style semantic processing in a computer
  system.
• Method permit an embodied system with sensors
  and effectors to develop its own meaning-system
  through experience in a rich environment.
• This is what people do.

11
How Does Joshua Work?

• Joshua is a deep implementation, using Slomans
  terms.
• Emotion (valence/arousal) is embedded and affects
  every aspect of processing.
• Top-down effects come from cognitive modules
  operating upon nodes in the semantic network.
• Joshua has facial expressions to facilitate
  social interaction and external regulation.

12
Implementation

• A spreading-activation semantic network with an
  advanced knowledge representation creates
  associations based on experience.
• Links that are used gain strength, unused links
  are pruned, as with neurons.
• Links that rise above a threshold are considered
  to enter consciousness and become the focus of a
  set of higher-level cognitive processes.

13
Emotion is Crucial

• Every node has an emotional valence assigned when
  it is created, based on the systems overall
  emotional state.
• The system-wide state constantly changes to
  reflect the states of the individual nodes.
• A second parameter, arousal, also determines the
  strength of valence and the spreading activation
  among nodes.
• Consciousness threshold varies with arousal.

14
Layered Control

• Emotion performs different functions at different
  levels in the system
• Valence arousal bias which nodes become active
  and rise to the consciousness threshold.
• Appraisals of experience can also change emotion,
  results of cognition affect emotion.
• In consciousness, meaning is assigned to certain
  emotional states a perceptual approach,
  permitting override of emotion.
• External influence comes from social interaction.

15
Some Shallow Implementations

• Limited emotional functionality has been designed
  into a variety of systems with special purposes.
• Few theorists have focused on emotion in the
  architecture of mind, but there exist many models
  of specific aspects of emotion
• Appraisal theory
• Basic emotions theory
• Approach/avoidance models

16
MIT Learning Companion
Pupil Detection Using the IBM BlueEyes
Camera Kapoor, Mota Picard (2001). Towards a
Learning Companion that Recognizes Affect, AAAI
Fall Symposium 2001, North Falmouth, MA
17
Relational Agents

• Tim Bickmore has developed a virtual agent that
  acts as an exercise coach to encourage physical
  training. Here he is shown talking with Rea, a
  virtual real estate agent.

18
Sensing Driver Affect
Detecting Driver Stress MIT Media Lab Healey
Picard (2000). Smart Car Detecting Driver
Stress. Proceedings of the 15th International
Conference on Pattern Recognition, Barcelona,
Spain.
19
CMUs OZ Project (Bates/Reilly)
Otto Iris are animated characters that express
their own feelings in interactive games
Zoesis Studios, http//www.ottoandiris.com/
20
Robot Improv
Two robots perform a short play based on an
elementary acting exerciseThe actors decide on
their next action and line of dialog based on
their current goals and emotional state and the
other actor's last actions. There is no
pre-determined script, only sets of available
actions and dialog for the actors to choose from.
Each play is improvised at run-time.
Bruce, Knight Nourbakhsh. Robot Improv Using
Drama to Create Believable Agents. The Robotics
Institute, Carnegie Mellon University
21
Navigating Environments
Rodrigo Ventura and colleagues have created
soccer-playing robots that learn to respond to
environmental cues
Sadio, Tavares, Ventura Custodio (2001). An
emotion-based agent architecture application with
real robots. AAAI Fall Symposium, N. Falmouth, MA.