• Share
  • Email
  • Embed
  • Like
  • Save
  • Private Content
Boella Roles
 

Boella Roles

on

  • 606 views

 

Statistics

Views

Total Views
606
Views on SlideShare
606
Embed Views
0

Actions

Likes
0
Downloads
3
Comments
0

0 Embeds 0

No embeds

Accessibility

Upload Details

Uploaded via as Microsoft PowerPoint

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

Boella Roles Boella Roles Presentation Transcript

  • The Agent Oriented Ontology of Social Reality: Are Roles Agents? Guido Boella Dipartimento di Informatica – Torino joint work with Leendert van der Torre CWI – Amsterdam & TUDelft
  • NorMAS
    • Project: Normative MultiAgent Systems
    • Research question:
      • how to build for MAS a social reality composed of groups, normative systems, organizations and legal institutions
    • MAS composed of autonomous agents : behavior is directed by their beliefs and goals and maximize preferences
  • Normative Multiagent Systems
    • “ Sets of agents whose interactions are norm-governed; the norms prescribe how the agents ideally should and should not behave. [...]
    • Importantly, the norms allow for the possibility that actual behavior may at times deviate from the ideal, i.e., that violations of obligations, or of agents’ rights, may occur.”
    • (Jones & Carmo 2001)
  • Construction of Social Reality
    • Regulative norms:
      • Obligations , prohibitions, permissions, …
    • Searle’s constitutive norms :
      • Something counts as something else.
      • Legal classification of reality in term of institutional facts
      • Regulate the creation of institutional facts:
        • E.g., property, marriage, money, …
  • Construction of social reality
    • Searle ’s construction of social reality “ Some rules regulate antecedently existing forms of behaviour. For example, the rules of polite table behaviour regulate eating, but eating exists independently of these rules. Some rules, on the other hand, do not merely regulate an antecedently existing activity called playing chess; they, as it were, create the possibility of or define that activity. The activity of playing chess is constituted by action in accordance with these rules. Chess has no existence apart from these rules. The institutions of marriage, money, and promising are like the institutions of baseball and chess in that they are systems of such constitutive rules or conventions.
  • Which is the right metaphor?
    • Lakoff: Role of metaphor in cognition to conceptualize reality which is not bodily grounded.
    • An ontology of social reality should disclose the metaphorical mapping we use to understand social reality
    • We propose the agent metaphor for understanding social reality
  • The intentional stance
    • Dennet : attitudes like belief and desire are folk psychology concepts that can be fruitfully used in explanations of rational human behavior. For an explanation of behavior it does not matter whether one actually possesses these mental attitudes : we describe the behavior of an affectionate cat or an unwilling screw in terms of mental attitudes. Dennet calls treating a person o r artifact as a rational agent the ‘intentional stance’.
  • The importance of us
    • “ The possibility of ascribing goals, beliefs, and actions to collectives relies on the idea that collectives can be taken to resemble persons . […] both factual and normative beliefs can be ascribed (somewhat metaphorically) to groups, both formal and informal , structured and unstructured. ” Tuomela, 1995
  • Hint of formalizazion
    • Input/output logic
    • Decisions
  • Conditional attitudes
    • Beliefs and goal are conditional mental attitudes represented by rules
    • Different inferential properties : e.g.
      • beliefs are transitive
      • goals are not
  • Input/Output Logic (Makinson & van der Torre)
    • Let R  Rul: a,..,d ->x or (a,…d,x)
    • Out i (R) is closure under set of rules
      • Out 1 :SI Out 2 :SI,OR
      • Out 3 :SI,CT Out 4 :SI,OR,CT
    a ->x a,b ->x SI a,b ->x a,  b->x OR a ->x a ->b a,b->x CT a ->x
    • Out i + : Out i and ID
    a ->a ID
  • Example
    • Out 4 :SI,OR,CT
    x ->y a,c ->y SI a,b ->x a,  b->x OR a ->x a,x->y CT a ->y
    • Use Out i for goals,Out i + for beliefs
    SI
  • Beliefs and goals
    • Belief rules are used to compute consequences (a set of propositions) of decisions (sets of decision variables)
    • The consequences of beliefs are used to compute which goals remain unsatisfied
    • Agents try to minimize the goals left unsatisfied by decisions
  • First roles…
    • Which is the right metaphor for understanding roles?
    • What do we learn from that for social reality?
    • How do roles fit in the overall agent metaphor for social reality?
  • Role
    • role - 1606, from Fr. rôle lit. " roll (of paper) on which an actor's part is written " from O.Fr. rolle from L. rotula "small wheel," dim. of rota "wheel."
    • From ETYMOLOGY DICTIONARY
  • Expecting behavior
    • Sociology : Roles specify expected behavior
    • Thus the question becomes:
      • Expected by whom or what?
      • Behavior of whom or what? (the actor )
      • How to describe the expected behavior?
  • Expected behavior
    • Expectations have a deontic character : expectation about ideal behavior
    • Who has the expectation must be able to judge whether the actual behavior fits the expected one
    • Who has the expectation can describe behavior
    • Thus, expectations belong to agents
  • Expectation and coordination
    • Also expectations w/o a deontic character: predictions of behavior of others.
    • Castelfranchi: anticipatory coordination is basic cognitive ability
    • Gmytrasievicz and Durfee: recursive modelling
    • Predicting the behavior of other agents is necessary to evaluate an agent’s own decisions when agents interfere in the environment
  • Anticipatory coordination Agent1 B’ G’  Agent2 B G Agent1 B’ G’  ’ AGENT2 has a PROFILE of AGENT1 TO PREDICT IT
  • Behavior of the actor
    • The actor has a behavior
    • The actor has to understand the description of the expected behavior and implement it
    • The actor has to reason about what is expected from him and the consequences of its behavior
    • Thus, the actor is an agent
  • Behavior of the agent
    • The expected behavior is the result of a decision of an agent
    • Thus a role is the description of the behavior of an agent
    • Agents (and their behavior) are described and implemented by attributing them mental attitudes like beliefs and goals
  • Our claim: roles are agents
    • Roles are (descriptions of) agents
    • Roles defined in terms of beliefs and goals
    • Roles require the mechanism for anticipating the agent’s behavior : not the actor actual behavior but the one which is expected from the actor
  • Example 1: international trade
    • The buyer will arrange the payment as soon as the seller can prove that he shipped the goods .
    • The bill of lading is issued by the carrier in return for the goods that he received from the seller.
    • According to Article 10 of the CIMTG the Bill of Lading reliably indicates that the goods have been shipped in international trade procedures.
  • Example 1: international trade
    • Note that this article has normative element. Whether the Bill of Lading is evidence does not depend so much on whether a person is psychologically convinced by it, but the law simply stipulates that the buyer should consider this document as sufficient evidence.
  • “ should consider ”?
    • Autonomous agents cannot be forced to believe or want something.
    • Either the law is wrong or it is referring to something else.
    • In fact, the law is describing the expected behavior of an agent: it is describing a role .
  • “ should consider ”?
    • Wrong approach: Tan and Gordijn: should consider Bill (b) as proof of shipment (s) = O(B(b  s))
    • Beliefs of agents cannot be accessible , O(B(b  s)) is an obligation whose violation cannot be monitored .
  • “ should consider ”?
    • thus “should consider” means “ act as if ”
    • The role is attributed the goal to pay (p) if there is the proof of shipment s: s -> p
    • Note that s is not a fact, it is rather an institutional fact (Searle): the legal classification of reality expected from the role: b counts as s
  • “ should consider ”?
    • “ expected from the role” means: the overt behavior is judged according to the beliefs attributed to the role
    • if b and the role is attributed the belief b  s and the goal s -> p
    • If there is the bill of lading b, not paying ( ¬ p) is a violation of the expectation, since the role should have satisfied the goal p
  • Interpretation of behavior
    • The actual behavior of an agent is compared not with the expected one
    • Not only decisions/actions are compared with the expected ones
    • Actions have meaning in “context” (e.g., obligations are conditional)
    • Also as “context” of an action is used the “context” believed by the role
  • Example 1: international trade
    • There is the bill of lading b
    • The agent has to decide what to do
    • He does not believe the good is shipped ( ¬ s) so he is goal s -> p is not enabled
    • He knows that he will be judged according to the buyer role’s expectation ( b  s and s -> p)
    • The role is expected to pay from its point of view: the goal s -> p is enabled since s
    • The agent pays since he will be judged for his role
  • Anticipatory coordination Agent1 B’ G’  Agent2 B G Agent1 B’ G’  AGENT1 has a PROFILE of ROLE TO DESCRIBE EXPECTED BEHAVIOR  ’ ROLE2 B’’ G’’
  • Counterargument
    • The bill of lading counts as a proof for the trading organization/normative system, not only for the role
    • Hence, it is not shown that it is necessary that roles have beliefs
    • Wait…
  • Example 2: contract
    • Contract is defined as a statement/exchange of intentions .
    • Can be statements be always reliable?
    • Again, what the contract does is to create roles , descriptions of the behavior expected by the parties
  • Example 3: dialogue
    • Searle’s speech act are not useful in MAS: mental state is not accessible
    • Walton&Krabbe: Speech acts commit the speaker
    • What if he is lying?
    • Speech acts describe the beliefs and goals of the role of speaker
  • Example 3: dialogue
    • Assertion “roles are agents” is modeled not in terms of beliefs of the agent but of the role he plays.
    • Speech acts during the dialogue change the state of the role (its beliefs and its goals)
    • The agent is expected to respect the role he builds with his speech acts
  • Dialog games Agent1 B G  Speaker-ROLE B’ G’ Speaker-ROLE B’’ G’’
  • Preliminary summary
    • The agent metaphor can be applied to explain roles
    • What about the other social entities ?
    • If roles are descriptions of (expected) behavior, they depend on entities which are able to give and understand descriptions of behavior in terms of beliefs and goals
  • Roles and social entities
    • Roles are founded concepts (Guarino et al. 04): they depend on other entities
    • These entities are able to describe behavior in terms of beliefs and goals
    • Entities which include roles are agents , too (i.e., can be described as agents)
  • Social entities are agents
    • Are normative systems, organizations and groups agents ?
    • and if they are, which role have roles in them?
  • Normative systems are agents
    • N ormative multiagent systems a re dynamic social orders.
    • Castelfranchi : a social order is a pattern of interactions among interfering agents ``such that it allows the satisfaction of the interests of some agent A''. ( a shared goal, a value )
    • S ocial delegation : an institution, on behalf of the other agents , has to achieve some goal which is part of the plans of all members of the group or institution.
  • Normative systems are agents
    • Social control : ``an incessant local (micro) activity of its units ’’ aimed at restoring the regularities prescribed by norms
    • A dynamic social order requires a continuous activity for ensuring that the normative system's goals are achieved.
    • The a pplication of sanctions in response to violations cannot be taken for granted : it is a result of the decision of an agent: the normative agent
  • Collective acceptance f i G V B d collective attribution f G B f G B f G B social reality normative system/ organization d d d reality agents MAS
  • Your wish is my command
    • Obligations formalized as desires or goals of normative system NS seen as an agent
      • Goals of normative system which describe ideal behavior of system
      • Other goals specify how social control is achieved (Hart’s instrumental norms )
    • Specification of games in which agents take the (autonomous!) normative system into account.
      • an agent considers whether its actions will lead to a sanction
  • Obligations O a,NS (x,s|Y)
    • Y -> x is goal of NS (thus:  out(G NS ))
    • Y,  x ->V(a,n) is goal of NS
    • Y,V(a,n)->s is goal of NS
    • Y->  s is goal of agent a
    • V(a,n) = violation by agent a of norm n
    • s = sanction
  • Games
    • δ a  Lit(X a ) decision a
    • c a = Out(B a , δ a  δ b ) consequencesa
    • Optimal decision minim. unreached goals
      • U a ={y ->x  G a | y  c a and x  c a }
    • Agent a optimizes, given that b optimizes
  • Example of game - e j c - V(a,n) s … NS - V(a,n’) s’ … NS’ agent a NS
  • And beliefs?
    • Agents have beliefs : either they are used in the metaphorical mapping or the agent metaphor is wrong
    • But…
  • Construction of Social Reality
    • C onstitutive norms are beliefs of the normative agent
      • Institutional facts are just legal categories
      • Constitutive rules r egulate the creation of institutional facts: E.g., property, marriage, money,
      • Constitutive rules are the connections between propositions believed by the normative agent: paper -> money
  • x Counts As y in C
    • C,x -> y is belief of NS (thus  out + (B NS ))
    • Choose logic for counts-as conditional
      • Not just a causal conditional!
      • Important discussion on chaining:
        • x counts as y, y counts as z => x counts as z
  • Institutional facts as abstractions
    • “ A fenced field counts as property”: f -> p
    • “ Trespassing property is forbidden”: O(t|p)
    • Property is a legal category which exists only in the beliefs of the NS
    • Its function is to provide a context for the obligation (enabled basing on beliefs of NS)
    • Institutional facts are abstractions for defining obligations on a legal classification of reality
  • Delegation of power f i G V B d collective attribution f G B f G B f G B social reality normative system/ organization d d d reality agents MAS a -> s
  • Behavior is sanctioned NS B G O  Agent B G NS B G O s legal classification of reality Obligations for agents Goals of the normative system
  • Modifying NS
    • Propositional variables C changing NS
      • add/remove belief rules
      • add/remove goal rules
    • Norms defined as goals and beliefs can be changed
    • C are made true only as specified by belief rules of NS : constitutive norms
    • e.g. contract -> add(G, x -> y) in B NS
  • Behavior changes the NS NS B G O  Agent B’’ G’’ NS’ B’ G’ O’ s’
  • Social ontology
  • Social ontology AGENTS
  • Social ontology AGENT DESCRIPTION
  • Social ontology MENTAL ATTITUDES
  • Social ontology PROPOSITIONS
  • Social ontology MENTAL ATTITUDES DESCRIBED BY RULES
  • MOST IMPORTANT AGENTS ATTRIBUTE DESCRIPTIONS TO OTHER AGENTS
  • Normative ontology NORMATIVE AGENT
  • Normative ontology Institutional facts
  • Normative ontology Regulative norms Constitutive norms
  • Legal institutions Regulative norms Norm modification
  • Roles are agents
    • Roles are described by the normative systems as agents via the AD relation :
      • NS attributes them beliefs
      • NS attributes them goals
    • Roles are played by other agents
  • ROLES ARE AGENTS AGENTS ATTRIBUTE DESCRIPTIONS TO ROLES AS AGENTS
  • Groups are agents
    • a. When they take a decision, they consider also the goals of the group and they try to maximize their fulfillment. Hence, they are committed to the joint activity.
    • b. When they take a decision, they include in their decision actions which contribute to the efforts of the partners. Hence, they are committed to mutual support.
    • c. When they take a decision, they recursively model the decisions of their partners and their effects under the assumption that the partners are cooperative, too. Hence, they are mutually responsive to each other.
  • Organizations are agents
    • Organizations (Burocracies in Ouchi’s terminology) are modelled as normative systems structured into functional areas and roles.
    • Both functional areas and roles are modelled as agents:
    • They are the “ parts ” of the organization
  • Social ontology
  •  
  • ROLE PLAYING RELATION
  • Normative system and roles
    • A normative system has roles (trias politica)
      • policemen
      • judge
      • legislator
  • Organizations are agents f i G V B d collective attribution f G B f G B f G B social reality organization d d d reality agents MAS
  • Organizations are agents f i G B d f i G V B d f i G B d collective attribution f i G B d f G B f G B f G B social reality attribution organization roles role play d d d reality agents MAS
  • Power and representation
    • Delegation of powers represented by constitutive rules in organizations: “the signature of the CEO counts as a decision of the organization”
    • But what is the signature of a role? It depends on the player of the role: when an agent enters a role, its actions counts as actions of the role : it’s a belief of the role
  • f i G D B a f i G D V B d Role Organization AD
  • f i G D B a f i G D V B d Role Organization AD a -> s
  • f i G D B a f i G D V B d Role Organization AD f i G D B b Agent a -> s role play
  • f i G D B a f i G D V B d Role Organization AD f i G D B b Agent a -> s b -> a role play
  • Role have instances
    • Roles have a state which changes (AD): they must have instances .
    • Social entities have instances, too
    • The difference is that all social entities do not act in the real world.
    • They have no associated decision variables (at a certain level of detail)
  • Duties and roles
    • When an agent enters a role (he becomes an actor) the expected behavior of the role becomes the obligation of the agent
    • Joost Breuker: norms associated to roles, but responsibility is individual
  • Duties and roles
    • Eg.: if s -> p is a goal of a role R, the agent A playing role R becomes the bearer of the obligation O A (s | p)
    • No transfer of obligations!
    • These obligations are created by the institutional action which changes the role playing relation PL
  • Permissions
    • “ An employee is permitted to use the fax machine of the firm”
    • Beliefs of roles express also the means which the agents uses (planning rules):
      • m -> f is a belief of the role employee,
      • T -> f is one of its goals
    • The means specified by the role to achieve a goal become the permissions : P(m)
  • Duties of roles or of actors?
    • Football rules: expulsion for a fault of a player (the actor of a role)
    • Is the norms directed to the actor or to the role ?
    • Who is affected by the sanction ? Maybe the actor does not care but the role (and indirectly the team) does.
  • Steimann’s properties
    • A role comes with its own properties and behaviour.
    • Roles are descriptions of agents: they specify beliefs and goals.
    • Obligations, permissions and powers refer to roles.
  • Steimann’s properties
    • 2. Roles depend on relationships The “context” of a role is always a social entity like a normative system, an organization or a functional area. This entity is seen as an agent able to attribute mental states to other agents. Thus to create roles.
  • Steimann’s properties
    • 3. An object may play different roles simultaneously
    • The role playing relation has no contraints. Since roles are not types we have no multiple inheritance problems
  • Steimann’s properties
    • 4. An object may play the same role several times, simultaneously
    • Roles have instances which are related to agents by the role playing relation.
  • Steimann’s properties
    • 5. An object may acquire and abandon roles dynamically
    • The role playing relation can be changed as any other component of a normative system
  • Steimann’s properties
    • 6. The sequence in which roles may be acquired and relinquished can be subject to restrictions
    • The role playing relation, as an institutional fact, is changed by a power of an agent. Which are the possible powers is established by the normative system: thus they can be restricted by constitutive rules
  • Steimann’s properties
    • 7. Objects of unrelated types can play the same role
    • The role playing relation is restricted to “agents”, but no further restrictions are necessary
  • Steimann’s properties
    • 8. Roles can play roles
    • In principle, a role can be played by any agent, then also by another role
  • Steimann’s properties
    • 9. A role can be transferred from one object to another
    • The role instance can be linked to another agent via the role playing relation
    • (but obligations must be replaced)
  • Steimann’s properties
    • 10. The state of an object can be role-specific: “role played by an object should be viewed as a separate instance of the object”
    • Roles have instances with their beliefs and goals
  • Steimann’s properties
    • 11. Features of an object can be role-specific:
    • “ an object responds according to the role in which it is being addressed”
    • Powers are defined in terms of actions of roles.
  • Steimann’s properties
    • 12. Roles restrict access:
    • “ When addressed in a certain role, features of the object itself (or of other roles of the object) remain invisible”
    • The agent can affect the normative system only if its actions counts as other actions in a role
  • Steimann’s properties
    • 13. Different roles may share structure and behaviour
    • Roles, as descriptions of behavior, can be structured in a hierarchy
  • Steimann’s properties
    • 14. An object and its roles share identity:
    • ``a role is a mask that an object can wear'‘
    • The organization the role belongs to sees the agent’s actions through the role he plays
  • Steimann’s properties
    • 15. An object and its roles have different identities (the so-called counting problem)
    • Since roles have instances we can simply count role-instances
  • Conclusions
    • Role are agents…
  • Bibliography
    • A Game Theoretic Approach to Contracts in Multiagent Systems . IEEE Transactions on SMC-c , to appear
    • Roles are agents.In Procs. of ICEC , 2004, Delft, 2004
    • Contracts as legal institutions in organizations of autonomous agents. In Procs. of AAMAS'04 , New York, 2004
    • Delta: The social delegation cycle . In Procs. of DEON'04 Workshop , Madeira, 2004
    • The distribution of obligations by negotiation In Procs. of ECAI'04 , Valencia, 2004
    • Groups as agents with mental attitudes. In Procs. of AAMAS'04 , New York, 2004.
    • Permission and authorization in policies for virtual communities of agents. In Procs. of P2P Workshop at AAMAS’04, NY, 2004
    • Regulative and constitutive norms in normative multiagent systems. In Procs. KR'04 , Whistler, 2004
    • A formalization of coalition structures in MAS. In Procs. of ECAI'04
    • Attributing mental attitudes to normative systems. In Procs. of AAMAS'03, Melbourne 2003
    • Local policies for the control of virtual communities . In Procs. of IEEE/WIC WI’03, Halifax 2003
    • Permissions and obligations in hierarchical normative systems. In Procs. of ICAIL’03, Edinburgh 2003
  •  
  •