Agents

Agents

The Communication

The Communication

Aspects

Aspects

NZDIS Team
November 2000
University of Otago

Outline

Outline



Introduction to Agent Communication



Agent Messaging



Agent Communication Language (ACL)



Ontologies



Conversations



Standards bodies



Comparison to other technologies

Introduction

Introduction



Agents and Objects



What is messaging?



What is Communication Language
(ACL)?



Why not to use API instead of ACL?

Agents and Objects

Agents and Objects



Better and more tight encapsulation
of the behaviour and control



More goal/task oriented design and
modelling approach



More process and behaviour-centric
programming model



Independence from the underlying
implementation machinery

Agent Messaging

Agent Messaging



The concept of a message



Message delivery

– push vs. pull models
– peer-to-peer
– tuple spaces
– content based addressing



The message structure

– outer layer (routing, encoding, type)
– inner layer (content, ontology)

Pull vs. Push models

Pull vs. Push models

Push

Pull

sender

receiver



Delivery notification



Delivery failure recovery



Participant crash recovery

Peer-to-peer messaging

Peer-to-peer messaging



Drawbacks

– Explicit addressing schema
– The need for facilitators
– Not fully decoupled participants



Solutions

– Tuple spaces
– Content based addressing

Tuple Space

Tuple Space



Fully decoupled senders and receivers



No bootstrapping problem



Easy backup and recovery



But single point of failure

Space

receiver

senders

Content-based addressing

Content-based addressing



Combines some of the elements from pure peer-to-peer and tuple space
models



No need for explicit addressing



More flexible routing than in peer-to-peer model



The need for facilitators

sender

receivers

Routing

Agent

Agent

Communication

Communication

Language

Language



Language Elements

– Communicative Acts
– Actions and Propositions
– Logical Operators an predicate logic



Communicative acts

– denote the actions related to communication;

this is based on the speech act theory
(Searle, et al) and related utterance theories

Example of

Example of

Communicative Acts

Communicative Acts



Inform

– informs the recipient about a proposition

which the senders believes is true



Query-if

– asks the recipient if a given proposition

is true or not



Request

– requests the recipient to execute a given

action

Message Example

Message Example



text-based format for:

– human readability
– cross-process compatibility (the

participant implementation language is
irrelevant)



ongoing work on binary formats
and specific API for a given
programming language

Example Inform Message

Example Inform Message

(inform
:sender AgentA
:receiver AgentB
:content (and
(is_a AgentCommunication presentation)
(is_in_progress AgentCommunication))
:ontology PresentationOntology
)

The need to agree on meaning of
objects: presentation,
and predicates: is_a, is_in_progress

Ontology

Ontology



Conceptualization of the domain of
discourse



Defines the meaning and relations
between all the concepts (objects,
actions, predicates, relations,
associations, etc) in the domain of
discourse

Example Ontology

Example Ontology



presentation

– class of objects which denote the

presentation. Subclass of a class: action.
Can contain associations to concepts:
venue, date, time, subject.



is_a

– relationship between an object and its

class



is_in_progress

– state of the object of class action, which

denotes that the action is in progress

Ontology (cont.)

Ontology (cont.)



The need for:

– formal ontology specification
– formal ontology language
– ontology interchange
– reasoning and inference capabilities
– easy management and human readability



Solution:

– Unified Modelling Language and Object

Constraint Language (UML+OCL)

Protocols and

Protocols and

Conversations

Conversations



Protocol

– Complex and dynamic interaction schemas
– Interoperability issues for the standard

interactions schemas

– Reducing the complexity of the full ACL by

constraining the possible interactions



Conversation

– ongoing sequence of communicative acts,

conforming to one or several protocols

Modelling Conversations

Modelling Conversations

Terminated

Process request

Preprocess request

Terminated

Refuse

Agree

Done

Prepare request

Sender (initiator)

Receiver

Modelling Conversations

Modelling Conversations



Robust and easy modelling formalism



Graphical representation of static and
dynamic structures



complex and concurrent structures
support



direct mapping to the implementation



possibility of reusing the structures as
building blocks of complex conversations



easy monitoring of the progress of a
conversation

Standards Organizations

Standards Organizations



ACL standards:

– Foundation of Intelligent Physical Agents

(FIPA-ACL)

– Knowledge Query and Manipulation Language

(KQML)



Ontology

– FIPA
– Open Knowledge Base Connectivity (OKBC) and

Ontolingua (Stanford University)

– Knowledge Interchange Format (KIF)

FOR MORE INFO...

FIPA - http://www.fipa.org

FIPA - architecture

FIPA - architecture



Peer-to-peer messaging model



Specified delivery model



Specified standard services: ACC, DF,
AMS,



Semantic Language (SL) as a content
language



Work in progress on formal ontology
specifications

Messaging technologies

Messaging technologies



Peer-to-peer

– Java Messaging Service, CORBA

Messaging Services

– Remote Procedure Calls (RPC), Simple

Object Access Protocol (SOAP)



Tuple spaces

– TSpaces from IBM
– JavaSpaces from Sun Microsystems

ACL vs. API

ACL vs. API



Full featured language

– high flexibility, robustness
– expressive power
– encapsulation of domain specific

concepts inside ontology

– implementation/efficiency

considerations



Mapping the language to the API