Liar Detection Within Agent Communication
G. M ULLER and L. V ERCOUTER
muller, vercouter @emse.fr. ´ ´ MAS – G2I – ENS des Mines de Saint-Etienne
Liar Detection Within Agent Communication – p.1/23
Context Context: Open systems, Peer-to-peer presentations; Problem: vulnerability to malevolent agents; Our objectives: Introduce trust and reputation, Detect liars to update reputation values.
Liar Detection Within Agent Communication – p.2/23
Context
Local Beliefs
Communication with other agents
Agent b
Liar Detection Within Agent Communication – p.3/23
Context
Local Beliefs
Accept or Refuse Communication with other agents
Trust Decision
Reputation of other agents Agent b
Liar Detection Within Agent Communication – p.3/23
Graphical Outline
Local Beliefs
Communication with other agents
Lie Detection
Trust Decision
Reputation of other agents Agent b
Liar Detection Within Agent Communication – p.3/23
Outline
Formalisms we use; Definition of lie; Lie detection; Lie Detection and Trust Decision.
Liar Detection Within Agent Communication – p.4/23
Logical Operators
Doxatic: =“agent
[Firozabadi 98]
believes ”;
Action: [Elgesem 93]:
=“agent succeeded to bring about ”, [Santos 97]: =“agent tried to bring about ”, Deontic [von Wright 51]: =“it ought-to-be that ”.
Liar Detection Within Agent Communication – p.5/23
denotes a speech act; to ;
: th speech act from
Speech Acts
: Feasibility Precondition;
FP: RE:
: Rational Effect;
[Searle 69, Austin 62, FIPA 02]
Liar Detection Within Agent Communication – p.6/23
Combining Logics and Speech Acts
and
reached;
;
sends
:
sends
:
We can write:
.
Liar Detection Within Agent Communication – p.7/23
Outline
Formalisms we use; Definition of lie; Lie detection; Lie Detection and Trust Decision.
Liar Detection Within Agent Communication – p.8/23
only if
Sincerity condition of FIPA ACLs: It is assumed that an agent sends holds.
Lie Definition Objectives
We explicit this sincerity condition to: define the concept of lie; introduce a norm to forbid lies; give agents means to detect liars.
Liar Detection Within Agent Communication – p.9/23
Lie and Deception Definitions
FP are not satisfied
to agent
emits
and succeeds
and a deception from agent as a successful lie:
emits
We define a lie from agent to agent as the emission of an act when do not holds:
FP are not satisfied
Liar Detection Within Agent Communication – p.10/23
FIPA’s sincerity condition
emits
FP hold
We can rewrite FIPA’s sincerity condition:
Violation of this sincerity condition:
Liar Detection Within Agent Communication – p.11/23
Outline
Formalisms we use; Definition of lie; Lie detection; Lie Detection and Trust Decision.
Liar Detection Within Agent Communication – p.12/23
Lie Detection
We propose two detection processes: Observation-Driven Detection Detectors are introduced and can observe some messages; Target-Driven Detection Any agent can detect an inconsistency and ask for “proofs”.
Liar Detection Within Agent Communication – p.13/23
Lie Detection – Assumptions
Our assumptions are: Agents share FIPA specs; Agents have consistent beliefs; Agents’ beliefs are not time-dependant; Messages are digitally signed;
Liar Detection Within Agent Communication – p.13/23
Observation-Driven Detection
B
T
A
Liar Detection Within Agent Communication – p.14/23
T
B
Observation-Driven Detection
O
A
Liar Detection Within Agent Communication – p.14/23
T
B
Observation-Driven Detection
O
A
Liar Detection Within Agent Communication – p.14/23
Observation-Driven Detection
From the Observer’s point of view:
Liar Detection Within Agent Communication – p.14/23
Observation-Driven Detection
From the Observer’s point of view:
Liar Detection Within Agent Communication – p.14/23
Observation-Driven Detection
From the Observer’s point of view:
Liar Detection Within Agent Communication – p.14/23
Observation-Driven Detection
If
From the Observer’s point of view:
The assumption of sincerity does not hold; The observer detects the lie; The observer becomes an evaluator.
Liar Detection Within Agent Communication – p.14/23
Target-Driven Detection
!
"
T
A
Liar Detection Within Agent Communication – p.15/23
Observer
!
"
A
Target-Driven Detection T
D
Liar Detection Within Agent Communication – p.15/23
"
(Evaluator)
A
!
(Evaluator)
Target-Driven Detection T
D
Liar Detection Within Agent Communication – p.15/23
Target-Driven Detection
!
T
"
Evaluator
T: Liar
D
Observer
A
Liar Detection Within Agent Communication – p.15/23
"
Target-Driven Detection A T
D
Evaluator
T: Liar
Liar Detection Within Agent Communication – p.15/23
Graphical Outline
Local Beliefs Consistency check and recovery Communication
t,b
sa k
with other agents Interaction with external observers and/or evaluators
t,b
Lie Detection
sa k
Trust Decision
Update
Reputation of other agents Agent b
Liar Detection Within Agent Communication – p.16/23
Outline
Formalisms we use; Definition of lie; Lie detection; Lie Detection and Trust Decision.
Liar Detection Within Agent Communication – p.17/23
Using Reputations to Decide to Trus Trust decision
Witness Unknown Reputation or not discriminant
Unknown or not discriminant
Neighbour Reputation
Unknown or not discriminant
Confidence
Distrust decision
information
Individual Reputation
Liar Detection Within Agent Communication – p.18/23
accepted
Trust Decision:
Accepting/Rejecting Messages
Liar Detection Within Agent Communication – p.19/23
accepted
Trust Decision:
Accepting/Rejecting Messages
rejected
Distrust Decision:
Liar Detection Within Agent Communication – p.19/23
accepted
Trust Decision:
Accepting/Rejecting Messages
rejected
Distrust Decision:
Protection against undetected lies.
Liar Detection Within Agent Communication – p.19/23
Summary
Local Beliefs Consistency check and recovery Communication
t,b
sa k
with other agents Interaction with external observers and/or evaluators
Accept or t,b refuse sa k t,b
Lie Detection
sa k
Update
Trust Decision
Use
Reputation of other agents Agent b
Liar Detection Within Agent Communication – p.20/23
Conclusion – Current Work Formalization of lie in Agent Communication; 2 decentralized processes of Detection; Integration in a reputation framework; From FIPA-ACL to Social Semantics: Check consistency of commitments, Temporal formulae;
Liar Detection Within Agent Communication – p.21/23
Bibliography – Fraud [Austin 62] J. L. Austin. How to do things with words. Oxford University Press, 1962. [Elgesem 93] D. Elgesem. Action Theory And Modal Logic. PhD thesis, Dept. of Philosophy, University of Oslo, 1993. [FIPA 02] FIPA. Fipa communicative act library specification, December 2002. [Firozabadi 98] B. S. Firozabadi, Y.-H. Tan, and R. M. Lee. Formal definitions of fraud. In DEON’98, 1998. [Lomuscio 03] A. Lomuscio and M. Sergot. A formalisation of violation, error recovery, and enforcement in the bit transmission problem. Journal of Applied Logic, 1, 2003. [Santos 97] F. Santos, J. Carmo, and A. Jones. Action concepts for describing organised interaction. In R. A. Sprague, editor, 13th Annual HICSS. [Searle 69] J. R. Searle. Speech Acts: an essay in the philosophy of language. Cambridge University Press, 1969. [Simmons 95] M. R. Simmons. Recognizing the elements of fraud, 1995. http://users.aol.com/marksimms/mrweb/fraudwww.htm. [von Wright 51] G.H. von Wright. Deontic logic. In Mind, volume 60, pages 1–15, 1951.
Liar Detection Within Agent Communication – p.22/23
Bibliography – Trust [Conte 02] R. Conte and M. Paolucci. Reputation in Artificial Societies. Social Beliefs for Social Order. Kluwer Academic Publishers, 2002. [Fornara 03] N. Fornara and M. Colombetti. Defining interaction protocols using a commitment-based agent communication language. In Proceedings of the AAMAS’03 Conference, pages 520–527, 2003. [McKnight 01] D.H. McKnight and N.L. Chervany. Trust in Cyber-societies, chapter Trust and Distrust Definitions: One Bite at a Time, pages 27–54. Springler-Verlag Berlin Heidelberg, 2001. [Rouchier 00] J. Rouchier. La confiance à travers l’échange. Accès aux pâturages au Nord-Cameroun et échanges non-marchands : des simulations dans des systèmes multi-agents. PhD thesis, Université d’Orleans, 2000. [Singh 03] M. P. Singh. Agent communication languages: Rethinking the principles. In Marc-Philippe Huget, editor, Communication in Multiagent Systems, volume 2650 of Lecture Notes in Computer Science, pages 37–50. Springer, 2003. [Sabater 02] J. Sabater and C. Sierra. Social regret, a reputation model based on social relations. SIGecom Exchanges. ACM, 3.1:44–56, 2002. http://www.acm.org/sigecom/exchanges/volume_3_(02)/3.1-Sabater.pdf.
Liar Detection Within Agent Communication – p.23/23
