Emile : Planning with Emotional Reasoning

1
Emile Planning with Emotional Reasoning

• Presented by
• Ramesh Viswanath
• Based on Emile Marshalling Passions in
  Training and Education
• By Jonathan Gratch

2
Introduction

• Named after Emile, treatise by Rousseau, the
  French thinker/philosopher
• Computers viewed as inanimate generally
• Human reaction to real life situation isnt
  predicated on logic
• Model to appraise emotional significance of events

3
Planning and Emile

• Plan based approach to representing emotions
• Handling emotions like fear and hope involve
  future expectations from plans
• Planning algorithms have tools to detect
  incompatibility between two plans. We can extend
  this to enure we never have two contradicting
  emotions.
• Emotional state will change as a reflection of
  the state of the planning process.

4
Emile Processing Stage 1

• Represent the plans and manipulate ideal
  representation to choose correct action to
  further the goal
• Emile uses a set of STRIPS operators, plus
  constraints ( temporal, binding , protection).

5
Example
6
Emile Processing Stage 2

• Appraisal evaluation of emotional significance
  of an event
• Construal Frames Used to assess relationship
  between event and agents disposition ( ie goals,
  preferences et al )
• Derive Emotion eliciting conditions ( like
  desirability, expectation status, evaluation )

7
Example of a Construal Frame
8
Stage 2 Appraisal in Emile

• Appraises the state of the plan in memory
• Level of indirection from standard, as events
  modify the state of plans in memory
• Places large number of domain specific construal
  frames with small number of domain independent
  rules to derive emotion eliciting conditions
• Emotions are defined as relations over these
  conditions.

9
Example Extended
10
Emile Processing Stage 3

• Intensity of the appraisals of emotions
• Use two factors to determine this
• Probability of goal attainment Using a simple
  model, and defining probability of correct
  execution of action, and probability of reaching
  an unplanned goal, they propogate the probability
  as a joint ( an approximation )
• Emotional Importance based on reward for
  reaching a goal/subgoal ( intrinsic ) and how it
  furthers other goals/subgoals (extrinsic ). They
  use a first order approximation again to summate
  the values.
• They use the following rules to derive the
  emotion intensity.

11
Estimation of Probability and Importance
12
Final Emotional Intensity
13
Emile Processing Stage 4

• Integration of various appraisals of the current
  state with the overal picture.
• They update the probability assessments of stage
  2.
• Linear update with smoothness factor, a

14
Emile Processing Stage 5

• The agents current emotional state should
  influence his behaviour.
• In Emile, the agents learn of other agents
  activities through perception of their actions,
  though it should eventually guess the plans of
  others.
• Emotional state affects how it interprets actions
  of other agents, and its own action selection.

15
Illustration Emotion reflected as facial
gestures
16
Conclusion

• Simplistic model
• Need for well defined relevance models and goals
• Planning approach fits in well with greater body
  of work.