On the benefits of argumentation for negotiation preliminary version

1
On the benefits of argumentation for negotiation
preliminary version

• Adil Hussain and Francesca Toni
• Department of Computing,
• Imperial College London

2
Introduction

• Setting
• Resource re-allocation between agents
• One-to-one negotiation dialogues
• Aim To demonstrate the benefits of
  argumentation-based agent negotiation (increased
  effectiveness)
• Present work
• Agent model fully worked out and generative
• Two types of negotiation dialogue
• Related Work

3
Outline

• Preliminaries components of our framework
• Representation of the agents mind
• Negotiation Policies
• Simple policy
• Reason-based policy
• Implementation
• Summary and directions for future work

4
Preliminaries (1 of 3)

• Agent System
• Agents, Resources, Goals, Beliefs
• Resource Reallocation Problem (r.r.p.)

z
y
x
c
b
a
5
Preliminaries (2 of 3)

• Dialogue Move tell( X, Y, Subject )
• Content Language
• request(give(Z)) because Reasons
• accept(give(Z))
• refuse(give(Z)) because Reasons
• Dialogue Instance
• Dialogue Sequence

6
Preliminaries (3 of 3)

• 1a. tell( x, y, request(give(b)) because _ )
• 1b. tell( y, x, accept(give(b)) )
• 2a. tell( y, x, request(give(c)) because _ )
• 2b. tell( x, y, refuse(give(c)) because _ )
• 3a. tell( y, z, request(give(c)) because _ )
• 3b. tell( z, y, accept(give(c)) )

z
y
x
c
b
a
7
Representation of Agents (1 of 2)

• Assumption-based Argumentation (ABA) framework (
  L, R, A, )
• L, e.g. isAgent(Y), ()has(Y,Z), ()needs(Y,Z),
  etc.
• R (l0 ? l1, , ln),
• e.g. goalUnachievable(X)
• ? needs(X,Z), has(Y,Z), needs(Y,Z), Y ? X
• A, e.g. asm(has(X,Z))
• , e.g. asm(has(X,Z)) has(X,Z), has(Y,Z) X
  ? Y
• L, A, and (partially) R shared amongst agents

8
Representation of Agents (2 of 2)

• Beliefs modified by adding/removing to/from R
• Agents make assumptions as soon as these
  assumptions are admissible

9
Negotiation Policies (1 of 2)

• Both policies translate roughly as follows
• until needed resources have been obtained repeat
• select a missing needed resource r to obtain
• until r is obtained repeat
• if goal is achievable (GO),
• select a suitable agent y for asking r (SA)
• if there is no such y, end in failure
• initiate dialogue attempting to obtain r from y
• end in success
• Policies defined as sets of communication rules
  ( In, P, Out, X )

10
Negotiation Policies (2 of 2)

• Example Communication Rule
• In tell( Y, X, (request(give(Z)) because
  Reasons )
• P thisAgent(X), has(X,Z), needs(X,Z)
• Out tell( X, Y, (accept(give(Z))) )
• X - has(X,Z)

11
Simple policy Example 1

• 1a. tell( x, y, request(give(b)) because )
• 1b. tell( y, x, accept(give(b)) )
• 2a. tell( x, y, request(give(c)) because )
• 2b. tell( y, x, refuse(give(c)) because )
• 3a. tell( x, z, request(give(c)) because )
• 3b. tell( z, x, accept(give(c)) )

z
x
y
c
a
b
12
Simple policy Example 2

• 1a. tell( x, y, request(give(b)) because )
• 1b. tell( y, x, refuse(give(b)) because )
• 2a. tell( x, z, request(give(b)) because )
• 2b. tell( z, x, refuse(give(b)) because )

z
x
y
c
a
b
13
Reason-based policy - Example

• 1a. tell( x, y, request(give(b))
• because needs(x,b), has(x,b) )
• 1b. tell( y, x, refuse(give(b))
• because has(y,b), needs(y,b) )

z
x
y
c
a
b
14
Properties of the policies

• Consistency of beliefs
• Exhaustiveness of responding conditions
• Termination of dialogue instances
• Non-duplicate requests
• Termination of dialogue sequences
• Completeness
• Efficiency

15
Implementing the policies

• Java Agent DEvelopment Framework (Jade)
• Credulous and Sceptical Argumentation Prolog
  Implementation (CaSAPI)
• PrologBeans

16
Future Work

• Goals achievable by alternative plans (each
  requiring different sets of resources)
• Conflicting goals (possibly with preferences)
• Persuasion
• ABA, Truthfulness, Resources, etc.

17
Conclusion

• Presented a framework that allows for generative
  agent policies to be defined
• Defined and Contrasted two negotiation policies
  for resource re-allocation
• Discussed properties of the agent policies and
  resulting dialogues
• Outlined directions for future work

18
Thanks for listening

• Any questions?