A Logic-Based Approach To Reasoning With Beliefs About Trust

Download Report

Transcript A Logic-Based Approach To Reasoning With Beliefs About Trust

A Logic-Based Approach to
Reasoning with Beliefs about
Trust
ARSPA 2004
Fariba Sadri1 and Francesca Toni1,2
1: Department of Computing, Imperial College, UK
2: Dipartimento di Informatica, Universita di Pisa, Italy
ARSPA04
Sadri, Toni
1
Main Features of Our Work
• Security via handling of trust
• Adopting an existing general-purpose, logic-based
framework to model trust
• Using abductive logic programming both for
knowledge representation and reasoning
• Allowing both static and dynamic knowledge
about trust: the dynamic knowledge allows agents’
belief in trustworthiness of other agents to evolve
through interactions with them
• Using beliefs about trust in communication and
negotiation policies
ARSPA04
Sadri, Toni
2
Abductive Logic Programs
<P, A, I>
• P is a logic program: set of rules of the form
Head  Body
• A is a set of abducible predicates:
in agents’ case A can consist of Actions and
Observations
• I is a set of integrity constraints: set of ifthen rules of the form
Conditions  Conclusions
ARSPA04
Sadri, Toni
3
Abductive Answers
Given <P, A, I> and a query Q an answer to
Q is (E, ) such that:
• EA is a set of ground abducible atoms,
and
• PE entails Q, and
• P E satisfies I
ARSPA04
Sadri, Toni
4
Abductive Proof Procedures
• Defined to compute abductive answers for given
queries
• Several have been proposed - we use CIFF:
– Endriss, Mancarella, Sadri, Terreni, Toni,
The CIFF proof procedure for abductive logic
programming with constraints,
Proc. Jelia 2004
– Endriss, Mancarella, Sadri, Terreni, Toni,
Abductive logic programming with CIFF:
implementation and applications,
Proc. CILC 2004
ARSPA04
Sadri, Toni
5
Example: (part of) KB of agent a
P: have(R, T)  initially(R),
not [given_away(R,T1)), T1<T]
have(R, T)  obtained(R,T2), T2<T,
not [given_away(R,T1)), T2<T1<T]
+ auxiliary definitions for given_away and obtained
tell(X, a, “give me R”, T), have(R,T) 
tell(a, X, “ok, I’ll give you R”,T'), T'<T+5
A: tell(a,X,S,T), tell(X,a,S,T)
I:
ARSPA04
Sadri, Toni
6
Example: Abductive answers
observation
Q: tell(b, a, “give me a camera”, 10)
I triggered
evaluated in P
have(camera,10) 
abduced
tell(a, b, “ok, I’ll give you a camera”, T'), T'<15
ARSPA04
Sadri, Toni
7
Representing Trust Policies:
Static Trust
trust(maria, anna, T)
trust(maria, dracula, T)  T>6, T<24
trust(maria, john, T)  false
ARSPA04
Sadri, Toni
8
Representing Trust Policies:
Dynamic Trust
trust(maria, X,T) 
tell(X,maria,”ok, I’ll give you R by T1”,T'),
do(X, deliver(R, maria, T2)), T2T1<T
trust(maria, X, T) friend(maria,X,T),
honest(X,T)
friend(maria,X,T) 
do(X,lend_money(maria), T'), T'<T
honest(X,T), in_prison(X,T'), T'  T false
ARSPA04
Sadri, Toni
9
Representing Trust Policies:
Context Dependent Trust
trust(maria,X,T,Task)  expert(X, Task, T)
trust(maria,X,T,Task)  has_a_goal(X,G,T),
helps(Task,G)
trust(maria, X,T, Task) 
tell(Y, maria, recommended(X, Task), T1),
T1T, trust(maria,Y,T)
ARSPA04
Sadri, Toni
10
Representing Trust Policies:
Role-Based Trust
trust(maria,X,T,advice(Issue)) 
authorised(X,give_advice(Issue),T)
authorised(X,give_advice(booking(Hotel),T)
receptionist(X,Hotel,T)
ARSPA04
Sadri, Toni
11
Using Beliefs About Trust
In private communication policies:
Determining how to respond to
queries/requests from other agents
Deciding who to contact for one’s information
or other resource needs
Answering other agents’ queries about trust
ARSPA04
Sadri, Toni
12
Example: Using Beliefs About Trust
Request for resources over time windows:
If you want the resource back give it to people
you trust:
tell(X, a, “give me R from T1 to T2”), T),
have(R,T), not need(a,R,T1,T2),
need(a,R,T3,T4), T3>T2, trust(a,X,T) 
tell(a, X, “ok, I’ll give you R from T1 to T2,
but I want R back before T3”,T'), T'<T+5
ARSPA04
Sadri, Toni
13
Further General Information
• Research developed within EU project SOCS :
http://lia.deis.unibo.it/Research/Projects/SOCS/guests/LIAIndex.html
• This model of trust could be employed directly by
KGP agents:
– Kakas, Mancarella, Sadri, Stathis, Toni,
The KGP model of agency,
ECAI 2004
– Stathis, Kakas, Lu, Demetriou, Endriss, Bracciali,
PROSOCS: a platform for programming software
agents in computational logic,
Proc 4th International Symposium “From agent theory
to agent implementation”, 2004
ARSPA04
Sadri, Toni
14
Future Work
• Resolving conflicting information
• Incorporating security
• Experimenting with scaled up, more
realistic examples and scenarios
ARSPA04
Sadri, Toni
15