practical and epistemic reasoning in argumentation ... · practical reasoning models for practical...

Introduction

Argumentation inAI

Practical reasoning

Models forpracticalreasoning

Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

1

Practical and epistemic reasoning inargumentation dialogues: from theory to

practice

Eric M. Kok John-Jules Ch. Meyer Henry PrakkenGerard A. W. Vreeswijk

Utrecht University

June 4, 2012

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

2

IntroductionArgumentation in AIPractical reasoning

Models for practical reasoningExisting approachesDifficulties

An experimentScenario generationA simple alternative

ResultsExperimental resultsConclusions

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

3

Argumentation

I Argumentation logics(semantics, structure, values/preferences, . . . )

I Argumentation dialogues(persuasion, negotiation, deliberation, . . . )

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

4

Why argue?

Agent that argue are supposed to be

I more efficient

I more effective

But are they, in practise?

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

5

Experimental validation

Validation through:

I Generating scenarios

I Running agents

I Measure the dialogues

I Analyse

Deliberation!

I Multi-agent

I Shared and personal goals

I Epistemic and practical reasoning

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

6

Practical reasoning

Epistemic arguments

I About the truth status of beliefs

I Sceptical reasoning

Practical arguments

I About (proposals to do some) actionI Credulous reasoning (Prakken, 2006)

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

7

Argument scheme

Argument scheme for practical reasoning (Atkinson, 2005)

In the current circumstances RAction A should be performedTo bring about new circumstances SWhich will realise goal GAnd promote value V

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

8

An example

Epistemic reasoning

I ‘steaks made from wagyu cattle are wagyu steaks’

I ‘steaks from wagyu beef are the best steaks’

Practical reasoning

I ‘to enjoy our dinner we should go to the bistro, becausethey serve the best steaks’

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

9

Existing approaches

Atkinson (2005)

I AS encoded as RA−→ S |= G ↑ v

I Strict protocol with pre- and post-conditions

Rahwan and Amgoud (2007)

I Distinct belief and desire undercut

I Desire and consequence conflicts using custom semantics

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

10

Existing approaches (cont.)

Black (2011)

I AS encoded as single-inference abstract argument

I Asserted arguments are evaluated as argumentation theory

Bench-Capon and Prakken (2006); Prakken (2006)

I Accrual of structured arguments

I Desire modality in logical language

I E-p-semantics

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

11

Desirable properties

I Embedded in the logical language

I Playing nice with existing semantics

I Structured argumentation

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

12

Desire modality

Following Bench-Capon and Prakken (2006); Prakken (2006):

I backwards application of rules

DenjoyDinner bestSteak ⇒ enjoyDinner

DbestSteak goToBistro ⇒ bestSteak

DgoToBistro

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

13

Attack between practical arguments


DbestSteak goToBistro ⇒ bestSteak

DgoToBistro

is alternative-attacked by


DbestSteak

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

14

Explosion of arguments

Now consider a (still small) knowledge base:

I Goals: Dg

I Action: o

I Rules: o⇒ p; p⇒ q; q⇒ r; r⇒ g

I q⇒ g;p⇒ g; o⇒ g

I o⇒ q; o⇒ r

Explosion of arguments and argument attacks!

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

15

Explosion of arguments

1. PRO P START

DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS

r3:q<-rFR

DrPPS

r4:r<-oFR

DoPPS

2. CON O ALTERNATIVE

DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS

r9:q<-oFR

DoPPS

3. PRO P ALTERNATIVE

DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS

r3:q<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS

r9:q<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS

r3:q<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS

r8:p<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r7:p<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS


DgFF

r1:g<-pFR

DpPPS

r7:p<-rFR

DrPPS


DgFF

r10:g<-oFR

DoPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS

r3:q<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS

r9:q<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS

r3:q<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS

r8:p<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r7:p<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS


DgFF

r1:g<-pFR

DpPPS

r7:p<-rFR

DrPPS


DgFF

r10:g<-oFR

DoPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS

r3:q<-rFR

DrPPS


DgFF

r10:g<-oFR

DoPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS

r8:p<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r7:p<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS

r8:p<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS


DgFF

r1:g<-pFR

DpPPS

r8:p<-oFR

DoPPS


DgFF

r10:g<-oFR

DoPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r6:g<-rFR

DrPPS

51. PRO P START

DgFF

r5:g<-qFR

DqPPS

r3:q<-rFR

DrPPS

r4:r<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS

r9:q<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS

r3:q<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS

r9:q<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS

r3:q<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS

r8:p<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r7:p<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS


DgFF

r1:g<-pFR

DpPPS

r7:p<-rFR

DrPPS


DgFF

r10:g<-oFR

DoPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS

r3:q<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS

r9:q<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS

r3:q<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS

r8:p<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r7:p<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS


DgFF

r1:g<-pFR

DpPPS

r7:p<-rFR

DrPPS


DgFF

r10:g<-oFR

DoPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS

r3:q<-rFR

DrPPS


DgFF

r10:g<-oFR

DoPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS

r8:p<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r7:p<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS

r8:p<-oFR

DoPPS


DgFF

r1:g<-pFR

DpPPS

r2:p<-qFR

DqPPS


DgFF

r1:g<-pFR

DpPPS

r8:p<-oFR

DoPPS


DgFF

r10:g<-oFR

DoPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r6:g<-rFR

DrPPS

101. PRO P START

DgFF

r6:g<-rFR

DrPPS

r4:r<-oFR

DoPPS


DgFF

r10:g<-oFR

DoPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r10:g<-oFR

DoPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r1:g<-pFR

DpPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r10:g<-oFR

DoPPS


DgFF

r6:g<-rFR

DrPPS


DgFF

r5:g<-qFR

DqPPS


DgFF

r6:g<-rFR

DrPPS

1

Cause: alternatives

I How realistic is this?

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

16

Generation deliberation scenarios

Deliberation scenario characteristics:

I Multi-agent

I Shared and personal knowledge

I Shared and personal goals

I Various possible actions

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

17

Knowledge assignment

Connect an assigned goal g to some possible action a:

o⇒ p;p⇒ q; q⇒ r; r⇒ g

Rule chaining (Kok et al., 2011)

I Repeat for actions and personal and mutual goals

Result: argument explosion...

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

18

Alternative encoding

Allow goals and actions in the topic language

enjoyDinner enjoyDinner ⇒ bestSteak

bestSteak bestSteak ⇒ goToBistro

goToBistro

Upside:

I normal models for structured argumentation apply

I no explosion of (attack amongst) arguments

Downside:

I action as conclusion instead of premise

I no intertwined credulous & epistemic reasoning

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

19

An experiment with deliberation

Full model for experimentation

I Formal model for deliberation dialogues

I Generating scenarios through rule chaining

I Simple arguing and non-arguing strategies

Software simulation

I Play many dialogues

I Measure efficiency and effectiveness

(See http://aspic.cossac.org/ for South and Vreeswijk’s ASPIC Java Inference Components)

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

20

Dialogue efficiency

●

●●

●

●

●

●

●●

●

●●

●●●●●●●●●●●●●●●●●●●●●●●●●

●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●

arguing non−arguing

1020

3040

50

f d

Arguing vs. non-arguing efficiency in number of moves (Kok et al.,

2012)

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

21

Dialogue effectiveness

●●

●

●

●●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●●

●●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●●

●

●

●

●

●

●

●

●●●

●

●

●

●

●●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●●

●

●

●

●●

●

●●

●

●

●

●

●

●

arguing non−arguing

010

2030

4050

v d

Arguing vs. non-arguing effectiveness as combined utility (Kok et al.,

2012)

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

22

Wrap up

Conclusions

I Formal models for practical reasoning not quite ready

I But argumentation has great potential benefits

I’ll be...

I At AAMAS/ArgMAS

I Finishing up this work in my thesis

Introduction

Argumentation inAI

Practical reasoning


Existingapproaches

Difficulties

An experiment

Scenario generation

A simplealternative

Results

Experimentalresults

Conclusions

23

References

I Atkinson, K., Bench-Capon, T. J. M., and McBurney, P. (2005). A dialogue game protocol formulti-agent argument over proposals for action. Au- tonomous Agents and Multi-Agent Systems,11(2):153171.

I Bench-Capon, T. J. M. and Prakken, H. (2006). Justifying actions by accruing arguments. InComputational Models of Argument: Proceedings of COMMA 2006, pages 247258.

I E. Black and K. Bentley. An empirical study of a deliberation dialogue system. In Proceedings of the1st International Workshop on the Theory and Applications of Formal Argumentation, Barcelona,Spain, 2011.

I E. M. Kok, J.-J. C. Meyer, H. van Oostendorp, H. Prakken, and G. A. W. Vreeswijk. A Methodologyfor the Generation of Multi-Agent Argumentation Dialogue Scenarios. In 9th European Workshop onMulti-agent Systems, Maastricht, The Netherlands, 2011.

I E. M. Kok, J.-J. C. Meyer, H. van Oostendorp, H. Prakken, and G. A. W. Vreeswijk. Testing thebenefits of structured argumentation in multi-agent deliberation dialogues. To be presented at the11th International Conference on Autonomous Agents and Multiagent Systems, Valencia, Spain, 2012.

I Prakken, H. (2006). Combining sceptical epistemic reasoning with credulous practical reasoning. InDunne, P. and Bench-Capon, T., editors, Proceedings of COMMA-06, pages 311322. IOS Press.

I Rahwan, I. and Amgoud, L. (2007). An argumentation-based approach for practical reasoning.Lecture Notes in Computer Science, 4766:74.

I E-mail: [email protected]

practical and epistemic reasoning in argumentation ... · practical reasoning models for practical...

Documents