Virtual Organizations as Normative Multiagent Systems

1
Virtual Organizations as Normative Multiagent
Systems

• Guido Boella
• Università di Torino, guido_at_di.unito.it
• Joris Hulstijn
• Vrije Universiteit, Amsterdam, jhulstijn_at_feweb.vu.
  nl
• Leendert van der Torre
• CWI, Amsterdam, torre_at_cwi.nl

2
Virtual Organizations

• Virtual Organizations individuals and
  institutions that need to coordinate resources
  and services across institutional boundaries
  (Foster et al)
• Infrastructures e.g. GRID, CSCW, KM,
• Users form a virtual community, with shared norms
  and objectives.
• Align community norms with infrastructure rules?
• Client-server global policies, but no local
  control
• Peer-to-peer local control, but no global
  policies
• Need a conceptual model of norms at different
  levels of control.

3
Normative Multiagent Systems

• Normative system set of norms (obligations) with
  an enforcement mechanism.
• Multiagent system set of autonomous agents with
  beliefs, goals, actions ...
• Model interaction between agents by recursive
  modeling
• Model normative system N as any other agent
• Boella Van der Torre KR04,AAMAS04

4
Overview

• (i). Example (ii). NMAS (iii). Conclusions
• How can the behavior of an individual agent in a
  virtual organization be described?
• How can agents change a virtual organization?
• How can agents in a virtual organization
  establish normative relations or contracts with
  each other?
• How can we deal with norms that operate at
  different levels of control?

5
(i). Example
6
Distributed Access Control

• Global norms, but local access control.
• owners have the right to entitle access to a
  resource
• storage providers can grant or withhold access
• Firozabadi and Sergot 2002
• So a2 must
• check ID a1,
• check entitlement a1
• weigh obligations against own goals

7
Observations

• Dynamic agents can enter, leave and alter the
  normative system.
• Interactive agents can agree on a contract (set
  of mutual obligations), enforced by N.
• Obligations are effective only when accompanied
  by an enforcement mechanism.
• Violation detection and sanctioning can be
  delegated to other agents.
• Roles subjects, defenders, normative system

8
(ii). NMAS

9
Individual Agent

• Focus on goal generation
• Use sets of production rules P ? Q to represent
  beliefs and goals, with a priority order lt.
• Belief rules information about current state
• Goal rules information about ideal future state

10
Example

• 1. Belief at party
• 2. Goal at party ? drink beer
• 3. Goal drink beer ? smoke cigarette
• 4. Goal smoke cigarette
• Priority 1 gt 2 , 4 gt 3
• Outcome

at party, drink beer, smoke cigarette ,
11
Recursive Modeling
agent A deliberates about optimal decision
considers optimal decision of agent B
agent B deliberates about optimal decision
considers optimal decision of agent A
agent A deliberates about optimal
decision considers optimal decision of agent B

• Profile (set of P ? Q rules) depends on role.
• Used for trust and deception.

12
Constitutive Rules

• Establish institutional facts by constitutive
  rules Searle 1995.
• E.g. counts as an entitlement to
  lunch,
• at the HICSS conference.
• P counts as Q in institutional context C
  whenever C ? P ? Q is a belief of N

13
Norms

• Obligation of A to N to bring about P in context
  C,
• under sanction S iff
• 1. Goal of N C ? P
• 2. Goal of N (C ? P) ? Viol(A, P)
• 3. Goal of N Viol(A, P)
• 4. Goal of N Viol(A, P) ? S
• 5. Goal of N S
• 6. Goal of A S
• 7. Goal of A P
• A 6 gt 7, N 2 gt 3, 4 gt 5

My wish is your command?
14
Dynamics

• The fact that normative rules 1-6 hold, is
  itself an institutional fact, i.e. a belief of N.
  
• A performative speech act counts as the creation
  of an institutional fact in context C, provided
  
• preparatory conditions hold, and
• sincerity, propositional and essential conditions
  hold.
• Owner a entitles b access to d, means either
• 1. create an obligation for all to grant b access
  to d, or
• 2. create a credential, used with a general
  access obligation.

15
(iii). Conclusions
16
Four kinds of Structures

• A regards N
• - e.g. decide to violate or not
• N regards As behavior
• - e.g. decide how to enforce
• A1 regards A2 given N
• - e.g. decide whether to trust
• N1 regards N2, given As behavior given N
• - e.g. decide to delegate or not

17
Conclusions

• Virtual organizations as normative multiagent
  systems.
• Individual agents are modeled goal generation,
  based on beliefs, goals and priorities.
• Dynamics can be captured by constitutive rules.
• Using recursive modeling and interaction, complex
  normative relations can be broken down into four
  types AN, NA, AA (N) and NN (A).
• Norms at different levels of control, can be
  dealt with by delegation to embedded normative
  multiagent systems, leading to different roles
  subjects, defenders and the normative system.

18
References

• G. Boella and L. van der Torre. Regulative and
  constitutive norms in normative multiagent
  systems. KR04.
• G. Boella and L. van der Torre. Attributing
  mental attitudes to normative systems. AAMAS04.
• J. R. Searle. The Construction of Social Reality.
  The Free Press, New York, 1995.
• Ronald M. Lee. Bureaucracies as deontic systems.
  ACM Transactions on Information Systems, 6(2)87
  108, 1988.
• Jones, A.J.I. Sergot, M.J. On the
  characterisation of law and computer systems In
  Deontic Logic in Computer Science, Wiley 1993,
  275 -- 307.
• L. Kagal, T. Finin, A. Joshi Trust-Based Security
  in Pervasive Computing Environments,
  Communication of the IEEE, 34 (12), 154 157,
  2001

19
Distributed Access Control (2)

• Goal of N req(a3,d) ? cred(a1,a3,d) ?
  acc(a2,a3,d)
• Goal of N req(a3,d) ? cred(a1,a3,d) ?
  acc(a2,a3,d) ? Viol(a2, acc(a2,a3,d))
• Goal of N Viol(a2, acc(a2,a3,d))
• Goal of N Viol(a2, acc(a2,a3,d)) ? ban(a2)
• Goal of N ban(a2)
• Goal of a2 ban(a2)
• Goal of a2 acc(a2,a3,d)
• for a2 6 gt 7
• for N 2 gt 3, 4 gt 5

20
Example

• Obligation of a to n not to overfish in spring,
• under sanction of paying a fine.
• Goal of n spring ? overfish
• Goal of n spring ? overfish ? Viol(overfish,
  a)
• Goal of n Viol(overfish, a)
• Goal of n spring ? Viol(overfish, a) ? fine
• Goal of n fine
• Goal of a fine
• Belief of a,n spring
• Goal of a overfish

Works only in case Desire 8 lt Desire 6