kwc.org Photos Spare Cycles MythBusters

Talk: Useful Roles of Emotions in Artificial Agents

A Case Study from Artificial Life Matthias Scheutz, Notre Dame

Hard to model given that even psychologists can define well.

Overview

  • emotions and AI
  • possible roles of emotions in agent architectures
  • a methodological approach to the study of the utility of emotions
  • experiments with emotional agents
  • concluding thoughts/results

Why emotions

Other people's reasons: * game industry wants them * because biological creatures have them * because complex agents are not possible without them

Speaker's reasons: * because agents may not be able to interact effectively with humans otherwise. have to be aware of human needs/emotional states * interested in cognitive science. understand emotions better * because emotional control mechanisms might be more efficient than other control mechanisms

Sprint voice operator doesn't take into account users' affective state when they are interacting with customer service.

Subfields for emotions

  • 'believable' agents
  • instruction and tutoring
  • user-interfaces
  • 'emotional' robots
  • computational models of human emotion
  • artificial life creatures

12 possible roles of emotions in agent architectures

  1. Action selection
  2. Adaptation: change behavior in short or long term
  3. Social regulation: display emotions to be observed, others react to them
  4. Sensory integration: based on internal states, integrated sensory stimuli in different way
  5. Alarm mechanisms: e.g. fight-or-flight reaction
  6. Motivation: create motives as part of emotional coping
  7. Goal management: repriortize on emotional states
  8. Learning
  9. Attentional focus: selection of data based on emotional reaction
  10. Memory control: bias formation memories based on emotion
  11. Strategic processing
  12. Self model

Focusing on first three for this talk*

Task and environment

Artificial life setting with foraging and survival tasks. Need to compete for resources to survive and procreate. Agents need energy for movement, processing, and procreation. Resources are scarce: agents can choose to fight for resource or retreat (emotional control).

The Basic Conflict Model

  • every agent has an action tendency for fight or retreat in conflict
  • agents display their action tendency
  • agents also have a conflict tendency that determines their disposition

The Utility of anger and fear * anger and fear states * Performance based on number of survivors after certain number of cycles

Three architectural dimensions

  • fixed vs. variable conflict tendency
  • non-adaptive vs. adaptive action tendencies
  • asocial vs. social

Eight agent kinds based on the three dimensions

Experiment

Parameters * action and conflict tendencies * conflict resolution policy * number and initial position of agents * number and initial position of resources

Emotional agents did better than all other agents except for adaptive social agents

Post a comment


tags.

related entries.

what is this?

This page contains a single entry from kwc blog posted on July 27, 2004 4:24 PM.

The previous post was Talk: Identifying Terrorist Activity with AI Plan Recognition Technology.

The next post is Comic-Con '04 Backposting.

Current entries can be found on the main page.