Agents with Personality: Negotiating Agents and Marital Stability

1
Agents with PersonalityNegotiating Agents and
Marital Stability

• Vicki H. Allan
• (Sponsored by CURI)
• Sylvia Niehuis

2
A Software Agent in its Environment
AGENT Database of Goals Personality Traits
(Neo) Couple Interactions(Gottman)
Sensor Input Comments Emotions Inquiries
Action output Comments Emotions Inquiries Body
Language
ENVIRONMENT Partner
3
Agents

• Need for computer systems to act in our best
  interests
• The issues addressed in multiagent systems have
  profound implications for our understanding of
  ourselves. Wooldridge
• The mere fact of writing equations forces you to
  be clear about your hypotheses. John Gottman
  (recent Chronicle of Higher Education)

4
Emile IPDSynthetic Agents Military Scenario
(UCLA)

• Mission rehearsal
• training scenario or education
• interactive story whose outcome depends on
  decisions and actions of participants
• Opening scene accident, child injured. Mother
  waits for lieutenant to arrive.
• Arrive goal authority present is satisfied.
• Squads ordered to move out mothers goal is
  threatened.

5
Affective Reasoning Project DePaul

• Elliot and Rick raced bikes around their block.
  Rick won as usual
• Rick Hey I won the race again
• Rick bullies Elliot as usual
• Rick Hey Elliot. I can make you do anything I
  want.
• Elliot Okay. Lets race again.
• They raced again and this time Ricks chain came
  off and he lost.
• Rick My chain came off.
• Elliot Thats too bad.
• Rick Come back here and help me before I pound
  you.
• Elliot I wont do it. You cannot make me.
• Rick Darn. I am going to get you.
• Elliot I dont care. Ill get you too.
• Rick I am going home
• Rick thinks imagine that, he fought back for
  once.

6
Alpha Wolves(MIT)

• Human participants have high level of control
  over wolves not sacrifice realistic autonomous
  personality of wolf
• Participant growls into microphone to direct his
  pup to growl at white pup.
• Next time they meet remember interaction
• Might perform action as directed, but in a style
  that demonstrates its own impressions
• Action comes bundled with a trigger context

7

• Emotional agentsA computable science of emotions
• Data personality surveys, NEO-IPIP
• Gottman attachment, influence functions
• Surveyascertain goals, importance, emotional
  reaction to achievement/failure to meet goals.
• Virtual actors
• Respond morphing faces, text, and speech
• manifest temperament/control of emotions

8
We program agents to act like couple

• Give results of survey information so alerted
• Replay interaction
• Explore alternatives winner may actually lose
• See futility detached spouse not my fault
• Fast forward to predict changes in 10 years
• facilitate conversation about concerns
• Tools analyze problem
• May use with therapist

9
Program Agents with Personality and Emotion

• Bob and Alice are considering marriage.
• Evaluate their personalities (neo IPIP) via
  on-line perl-cgi scripts
• Create algebra of emotion
• Agent Bob and Agent Alice
• Give Agents a problem and view how they negotiate.

10
My wife and I had words, but I never got to use
mine. -Fibber McGee
11
Marital Research

• How a couple differs is not so important (as
  there will always be differences).
• What is important is how they deal with those
  differences.
• I have engine to produce emotions from goals.
  I need Sylvias help with
• selecting goals/subgoals for couple
• assigning emotions to goals/failure
• selecting appropriate reactions/actions
• evaluating believability of agent system

12
Interaction Patterns

• speaker/listener (take roles)
• hostile-detached
• validating
• conflict-avoiding
• criticism
• defensiveness
• contempt
• stonewalling (listener withdrawal emotionally and
  perhaps physically)
• kitchen sink (prior complaints brought up)

13
Which emotions to model?

• Joy
• Fear/anxiety
• Like/dislike
• anger
• shame/remorse
• startle

• interest
• sadness
• disgust/contempt
• shyness
• love

• Emotions are important in giving Disney
  characters
• the illusion of life.
• Believability vs realism may be better to
• use simplified, exaggerated characters.

14
How to Combine Emotions

• Winner take all ignore all but the highest
  intensity emotion
• Additive
• Logarithmic log(2emotion1 2emotion2)
• Focus kicking example
• React to immediate occurrence but add in general
  emotional state.

15
How emotions are created

• trigger emotion structure in response to a failed
  goal or goal that becomes less likely.
• also triggered if goal succeeds or becomes more
  likely
• Raw emotion is created, but must be manipulated
  by action component.

16
System details

• Data base of goals/subgoals. Tagged with
  importance and emotions.
• Use couple surveys (NEO, Relate) to
• activate goals
• assign emotions to goals
• increase likelihood of certain styles of reaction
  or behavior patterns

17

• Event happens
• emotions are generated
• deficiency list is generated
• Select appropriate action based on
• unmet goals
• deficiency list
• emotions
• personality
• history relationship to other person
• reading of other person
• importance of goal

18
Conclusion

• Access from website
• Visual tool
• Interactive tool to help couple
• Applies results of surveys
• Explore alternatives
• Analyze concerns
• Helps researcher to be clear about hypothesis