Get documents about "Interoperability in Agentspace proposal of agent interface to
Document Sample
Interoperability in Agentspace:
proposal of agent interface to environment
Stanisław Ambroszkiewicz
IPI PAN, Warsaw,
Poland
Supported by ESPRIT project CRIT2
September 2000
1
FOCUS OF OUR RESARCH:
Agent organizations in Cyberspace
Autonomous, heterogeneous agents are
supposed to form organizations!
The key issues:
•Infrastructure
• Interactions: mobility, communication,
services, ...
•Semantic interoperability
• Understanding: negotiation, cooperation, …
2
Interoperability
agent communities A1, A2, A3, … in a world (AGENTSPACE)
a community Ai consists of homogeneous agents perceiving the
world, interacting and speaking their own language Li
interoperability between heterogeneous communities:
heterogeneous agents can interact and understand each other
AGENTSPACE:
L1 L2
A1 A2 A3
L3
3
Interoperability continued
Fix agent community A1
What does it mean: perceiving the world, interacting, speaking
language L1, and understanding each other ???
perfect (interaction and semantic) interoperability inside A1:
common interaction infrastructure is NOT necessary
explicit meaning (ontology) of L1 is not needed !
L1
A1
4
Interoperability continued
interoperability between heterogeneous communities:
heterogeneous agents can interact and understand each other
interaction interoperability: common interaction
infrastructure;
semantic interoperability, explicit semantics is necessary:
language for ontology interchange + interpretation of O2
into O3 and vice versa
meaning of concepts is reduced to common generic
representation of the world structure
OIL or generic
L2,O2 representation of
the world
A2
Common A3
L3, O3
interaction
5
infrastructure
Generic MAP architecture as
Interaction Infrastructure
PEGAZ - our MAP for agents,
services, and agent
organizations development
Mobile Agent Platform - a uniform view of Cyberspace
service
place place place
Java Virtual Machine
Internet/Intranet/WAN/LAN (TCP,UDP)
6
LINUX SunOS MS Win NT Win 95/98
AGENT ARCHITECTURE
Library of Decision
routines mechanism
Interface Environment:
Routine
Learning Goal Action:
execution communicate
situations
communicate
Perception
KNOW event
Communic
-ation
Agent memory, i.e. local states 7
agent or service
Proposal of agent interface to
environment
The goal: means to achieve semantic interoperability
Agent interface:
Responsible for interactions. Based on
MASIF / FIPA standard. It assures
Functionality interaction interoperability: migration,
layer communication, using services, etc.
Representation of the world (Agentspace)
Representation structure. Local event structure is the
layer basis for the representation. Agents
perceive the environment in the same way
Communication
Language
Language Communication Language: simple query
layer language for homogenous agents and
Meaning Interchange Language (MIL to be
constructed) for heterogeneous agents
8
The idea behind
The reason for common interface:
nothing in common = no understanding, i.e. no semantic
interoperability
Functionality layer: MASIF / FIPA standard are not
sufficient; still work in progress
Our proposal:
Representation layer as formal representation of the
structure determined by the functionality level : generic
MAP environment & mechanisms for acquiring knowledge
from perception (already done!)
Language layer - MIL (Meaning Interchange Language) -
work in progress
9
ENVIRONMENT of a Mobile Agent
Platform
place service place place
10
FORMAL SPECIFICATION OF
GENERIC MAP ENVIRONMENT
take / give
Primitive entities: places,
agents, services, resources
Primitive actions: migrate, use service
use service, take (give)
resource, communicate communicate
with agent (service)
Events: each event
represents an action
execution - local
interaction of agents,
services and places
involved in the execution
place place
migrate
11
FORMAL SPECIFICATION OF
GENERIC MAP ENVIRONMENT
Partial order of events
...
service service
12
AGENT MEMORY
concept vi = ( name , carrier , meaning )
variable vi = (identifier, type, modification way)
Agent’s memory: collection of variables V1 , V2, … , Vk
(abstraction of database) standing for concepts.
13
ACQUIRING KNOWLEDGE 1:
FROM PERCEPTION
Update variable: at event:
where the current
location is stored place place
move
where names of
services are stored
where the info on
contents is saved
14
ACQUIRING KNOWLEDGE 2:
FROM COMMUNICATION
Homogeneous agents:
Agents’ databases are of the same type, i.e., V1 , V2, … , Vk (of
blue agent) and V1,V2, ... ,Vk (of green agent) , and Vi and Vi are
of the same type for i=1,2,...,k
Query Query: what is your Vi ?
Answer: my Vi=10.
Answer Agent: revision of Vi;
preserve Vi, change Vi to
10 or compute new value
Vi is used in the same way by the homogeneous agents !!!
15
meaning of concept is defined inside a community of homogeneous agents; sharing the same ontolog
CONCEPT MEANING is the way the
concept is used (Wittgenstein)
Agent’s Agent’s Environment:
database interface: Perception
event a homogeneous agent
Representation layer
Functionality layer
Language layer
update
Vi
Vi
revise
agent Communication
meaning of Vi = the way Vi is used in updating and revision
updating and revision mechanisms are the same for all
16
homogeneous agents !!! The same interfaces !!!
Meaning Interchange Language:
MIL
Ontology of society A1
Concepts: v1; vi
Reduction of
concept meaning Language layer Concept formation
generic meaning
Representation layer generic meaning
of v1 (in RDFS ?)
of vi (in RDFS ?)
? Functionality layer
?
from / to
17
a heterogeneous agent
Conclusion
What has been done:
Specification of the representation layer;
What not:
The rest of the interface;
Future work:
Construction of MIL, generic rules (primitive
constructors) for concept formation;
18
SEMANTIC INTEROPERABILITY:
agents can identify, communicate and
understand each other
to understand each other they must agree on the
meaning of concepts they use!
KEY ISSUE: meaning, semantics (ontology)
represented in a machine-readable way
Natural language: Can we represent the meaning of
concepts (we use) explicitly, i.e. in a machine-
readable way?
A fundamental problem !!!
Kant , Husserl, Wittgenstein, …
23
Formal approaches to semantics
and semantic interoperability
Tarskian semantics: semantics of a theory is given
by interpretation in a model. The model is another
theory !!!
Ontolingua: Gruber, Guarino et al.: meaning of
concept is constrained by logical axioms.
OKBC - Open Knowledge Base Connectivity,
exchange standard for ontologies chosen by FIPA
XML and RDF web standards for information
exchange
OIL - Ontology Interchange Language, a European
project
DAML project - DARPA Agent Markup Language: a
semantic language that ties the information on page
to machine-readable semantics 24
Related docs
