Arguing and Negotiating

1
Arguing and Negotiating in the Presence of Social
Influences
Nishan C. Karunatillake Nicholas R. Jennings Iyad
Rahwan Timothy J. Norman
CEEMAS05 Conference Budapest, Hungary 15th
September 2005
2
Introduction
The Context

• Actions that an individual agent may perform
  within a multi-agent society,
• are influenced via different forms of
  motivations.

• These motivations can be classified into two
  broad categories

• Internal Influences
• internal intrinsic motivations that drives the
  individual agent
• Social Influences
• external motivations effected via the social
  context

3
Introduction
The Problem

• Disparate motivations

• Between certain internal and social influence
• Between two or more social influences

• Incomplete information.

- On certain social influences that could or
should affect them - On certain internal and
social influences affecting their counterparts
4
Introduction
A Solution

• Argumentation-Based Negotiation (ABN) allows
  agents to a means to resolve such conflicts.

• How does it function

• ABN allows agents to negotiate

- Enables agents to achieve mutually acceptable
agreements, thus, - Provides a means to resolve
conflicts due to disparate motivations.

• ABN also allows agents to exchange additional
  meta-information
• (i.e., justifications, critics, and other forms
  of persuasive locutions)

• Enables agents to gain a wider understanding of
  the internal and social influences affecting
  their counterparts, thus,
• - Provides a means to resolve conflicts due to
  incomplete information.

5
Argumentation Model
Overview
6
The Schema
Two agent setting

• To capture social influences
• we use Castelfranchis notion of social
  commitment as our basic building block

• Social Commitment
• is a commitment by one agent x (termed the
  debtor)
• to another y (termed the creditor)
• to perform a stipulated action q

• As a result of such a social commitment
• - Debtor attains an obligation toward the
  creditor to perform the action q.
• Creditor attains the rights to exert influence
  (i.e., demand, require, and question)
• the debtor regarding its performance.

7
The Schema
Within a structured Society

• A society
• we can perceive as a collection of roles
  interconnected via a web of relationships.

• Since most relationships involve the related
  parties carrying out actions for each other ...
• we can view a relationship as
• an encapsulation of social commitments between
  the associated roles

• Thus

- The Debtor role gains the obligation to perform
the specified action - The Creditor role gains
the right to exert influence regarding its
performance.
8
The Schema
Formulation
9
The Arguments
Capturing Social Arguments
10
The Arguments
Socially Influencing Decisions
11
The Arguments
Negotiating Social Influence
12
Language and Protocol
The Language
The language has two basic parts
Act, RoleOf, In, DebtorOf, CreditorOf, ActionOf,
InfluenceOf, and AssocWith Do, Conflict
OPEN-DIALOGUE, CLOSE-DIALOGUE, PROPOSE, ACCEPT,
REJECT, CHALLENGE, ASSERT
13
Language and Protocol
Overview of the Protocol
Protocol Governs the interaction and guides the
agents to resolve their conflicts
14
Language and Protocol
Overview of the Protocol
Protocol Governs the interaction and guides the
agents to resolve their conflicts

• Inspired by the work on computational conflicts
  by Tessier et al.

• Complies well with the with the
  pragma-dialectics model for critical discussion
  proposed by Eemeren and Grootendorst

15
Language and Protocol
Operation of the Protocol

• We define the operation of the protocol as a
  dialogue game protocol

• Locutions Rules the valid set of moves
• Structural Rules the types of moves available
  after a given move
• Commitment Rules the commitments each
  participant incurs with each move

16
Decision Making Algorithms
Overview

• The protocol governs the rules of encounter

• However,
• The Individual agents need the capability to
  decide how they could gain the best possible
  outcome while playing within those rules.

• Decision mechanisms
• Model different type of agent behaviours
• Model different forms of societies

17
Decision Making Algorithms
Algorithms

• We model
• A self-interested agent society in which
• agents do not intentionally attempt to deceive
  one another

• Thus, when making decisions on whether or not to
  perform a certain action they will consider
• Feasibility if it is capable of performing that
  action
• Viability if the benefit of performing that
  action is more than its cost

18
Decision Making Algorithms
Algorithms
19
Resolving Conflicts through ABN
Example
20
Scenario
The Context
Capability All agents within the domain have an
array of capabilities. Action They also have
actions, which requires these capabilities.
Agents interact and negotiate between one another
to find capable counterparts that are willing to
sell their services to achieve their actions.
21
Modelling Social Influence
The Context
22
Modelling Social Influence
The Context
From this representation, we can extract the
rights and the obligations of each agent.
23
Incomplete Knowledge
The Context
24
Incomplete Knowledge
The Context
0
25
Incomplete Knowledge
The Context
0
26
Incomplete Knowledge
The Context
0
27
Agent Interaction
28
Managing Social Influences
The Experiments
1. Demanding Compensation
2. Questioning Non-performance
29
Demanding Compensation
Experiments
30
Agent Interaction
Buyer
Seller
Challenge(h)
Assert(Ho)
31
Demanding Compensation
Experiments
Observation 1 The argumentation strategy
allows agents to manage their social influences
even at high uncertainty levels.
32
Demanding Compensation
Experiments
Observation 2 In cases of perfect information
and complete uncertainty, both strategies perform
equally.
33
Demanding Compensation
Experiments
Observation 3 At all knowledge levels, the
argumentation strategy exchanges fewer messages
than the non-arguing one.
34
Demanding Compensation
Experiments
5 Roles
7 Roles
10 Roles
15 Roles
Observation 4 When there are more social
influences within the system, the performance
benefit of arguing is only significant at high
levels of knowledge incompleteness.
35
Managing Social Influences
The Experiments
1. Demanding Compensation
2. Questioning Non-performance
36
Questioning Non-performance
The Experiments
1. Claim-Penalty Argue
2. Argue in First Rejection
3. Argue in Last Rejection
37
Questioning Non-performance
Experiments
Observation 5 The effectiveness of the various
argumentation strategies are broadly similar.
38
Questioning Non-performance
Experiments
Observation 6 Allowing the agents to challenge
earlier in the dialogue, significantly increases
the efficiency of managing social influences.
39
Conclusions
Summary
Develops a novel ABN approach that allows agents
to argue, negotiate and, thereby, manage their
social influences.
Results
1. Even at high uncertainty levels ours give a
more efficient and a more effective strategy
in managing social influences. 2. Efficiency can
be enhanced by allowing agents to challenge one
another earlier in negotiation 3. Both
under - complete uncertainty and -
when there are abundant social influences
available in the society, the effectiveness
of our approach is not significantly different
from a non-arguing one.