Multi party multi issue multi strategy negotiation for multi modal virtual agents
This presentation is the property of its rightful owner.
Sponsored Links
1 / 22

Multi-party, Multi-issue, Multi-strategy Negotiation for Multi-modal Virtual Agents PowerPoint PPT Presentation


  • 138 Views
  • Uploaded on
  • Presentation posted in: General

Multi-party, Multi-issue, Multi-strategy Negotiation for Multi-modal Virtual Agents. S antosh S ugur 10/25/2011. Dr. David R Traum. Research Assistant Professor, USC Institute for Creative Technologies, USC. PhD CS , U Rochester Natural Language Processing. Stacy Marsella.

Download Presentation

Multi-party, Multi-issue, Multi-strategy Negotiation for Multi-modal Virtual Agents

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

Presentation Transcript


Multi party multi issue multi strategy negotiation for multi modal virtual agents

Multi-party, Multi-issue, Multi-strategy Negotiation for Multi-modal Virtual Agents

Santosh Sugur

10/25/2011


Multi party multi issue multi strategy negotiation for multi modal virtual agents

Dr. David R Traum

  • Research Assistant Professor, USC

  • Institute for Creative Technologies, USC.

  • PhD CS , U Rochester

  • Natural Language Processing.


Multi party multi issue multi strategy negotiation for multi modal virtual agents

Stacy Marsella

  • Research Associate Professor Computer Science, USC

  • Co-Director USC Emotion Group

  • Associate Director, Institute for Creative Technologies – Social Simulation

  • B.A. in Economics from Harvard University

  • Ph.D. in Computer Science from Rutgers University. Focus on AI planning, human problem solving and cognitive science


Multi party multi issue multi strategy negotiation for multi modal virtual agents

Jonathan Gratch

  • Research Associate Professor, USC

  • Associate Director, Institute for Creative Technologies - Virtual Human Research

  • Co-Director, USC Computational Emotion Group

  • Ph.D. in Computer Science, University of Illinois, June 1995

  • Joined ICT in August 1995


Multi party multi issue multi strategy negotiation for multi modal virtual agents

Jina Lee

  • Phd Student under Prof. Stacy Marsella

Arno Hartholt

  • Project leader of the Integrated Virtual Humans group and Central ICT Art Group.

  • BS, MS, University of Twente, Netherlands

  • Joined ICT in 2005


Negotiations never cut what you can untie

Negotiations“Never cut what you can untie”

  • Using agents to train humans in negotiations

  • Communicate over multiple modalities – speech, gestures, body postures

  • Dynamic negotiations – change positions and strategies to achieve those goals.


How can a human trainee do well

How can a human trainee do well?

  • Solve problems

  • Gain Trust

    • Familiarity

    • Credibility

    • Solidarity

  • Manage Interactions


Multi party dialogue model

Multi-party Dialogue model

  • Maintain a snapshot of Dialogue State – called Information State.

  • Information state is updated by dialogue moves based on certain update rules.

  • Information state and dialogue moves are partitioned into layers each dealing with slightly different aspects.


Layers in the dialogue model

Layers in the dialogue model

  • The contact layer concerns whether and how other individuals can be accessible for communication. Modalities include visual, voice (shout, normal, whisper), and radio.

  • The attention layer concerns the object or process that agents attend to. Contact is a prerequisite for attention.

  • The Conversation layer models the separate dialogue episodes that go on during an interaction.


Layers in the dialogue model contd

Layers in the dialogue model contd.

  • The participants may be active speakers, addressees, or over hearers.

  • The turn indicates the (active) participant with the right to communicate

  • The initiative indicates the participant who is controlling the direction of the conversation

  • The grounding component of a conversation tracks how information is added to the common ground of the participants.


Additions to dialogue model

Additions to dialogue model

  • Listening (input)

    • Improved Gaze model

    • Increase in non verbal feed back during listening.

  • Talking (output)

    • Verbal and non verbal behavior also takes into account where the negotiator’s views are coming from / larger understanding of the issue – helps in negotiation decisions.

    • Dynamically update motivations for strategies based on where the negotiations are at.


Multi party n egotiation strategies

Multi Party Negotiation Strategies

  • Apart from disagreements on topics, people may simply dislike each other which is not considered.

  • Find Issue

  • Avoid

  • Attack

  • Negotiate

  • Advocate

  • Success

  • Failure


  • Factors involved in strategy selection

    Factors involved in strategy selection

    • Topic

      • is there a topic? – no - find issue()

      • Is it what I want? – no - avoid()

      • Can I work with topic? Yes – negotiate()

    • Control

      • Can I control the discussion

      • I need to control if I have to avoid the topic

    • Utility

      • High utility? Advocate()

      • Absolute utility of issue

      • Relative utility compared to other options


    Factors involved in strategy selection1

    Factors involved in strategy selection

    • Potential

      • How good or bad can the utility get, should I attack or negotiate

    • Trust

      • Do I trust the other interlocutors

    • Commitments

      • Positive commitments are a required for a success strategy

      • Negative commitments result in failure


    Multi issue utilities

    Multi-issue Utilities

    • Agent actions (dialogues) cause agents to achieve their goals.

    • Agents are aware of the goals of other agents

    • Also aware of actions that can achieve or thwart those goals.

    • In negotiations try to weigh different action roadmaps to achieving goals.

    • Choose a plan (roadmap) based on positive utility and the probability that the plan will be executed.

    • Don’t choose a plan if it has negative utility and flaws that will block its execution


    Expected and potential benefit

    Expected and Potential Benefit

    • Expected benefit is a calculation based on commitments and trust.

    • Potential Benefit is the benefit assuming that the plan will succeed.

    • A plan with high potential benefits help to negotiate by advocating that course of action.


    Choosing negotiation strategies based on factors

    Choosing negotiation strategies based on factors


    Different possible actions for a chosen strategy

    Different possible actions for a chosen strategy

    • Find Issue

      • Propose a topic

      • Request a topic

      • Initiative parameter dictated which action to take

    • Avoid strategy

      • Change topic to high utility

      • Talk about non-issues

      • Disengage from meeting if there is some control


    Different possible actions for a chosen strategy1

    Different possible actions for a chosen strategy

    • Attack strategy

      • Ad hominem attack

      • Point out flaws in issue

    • Negotiate

      • Attack

      • Propose solutions

    • Advocate

      • Talk about high utility outcomes

      • Address flaws

      • Offer / solicit commitments


    Failed negotiation example

    Failed negotiation example


    Successful negotiation example

    Successful negotiation example


    Discussion points

    Discussion points

    • What could be some of the things that could be used in a human to human negotiation, that cannot be used with virtual agents.

      • Eg. Making a connection based on similar past experiences between interlocutors and using that to your advantage.

    • Can such a training setup replace real negotiation experiences?


  • Login