Software Agents: An Overview and Designing Behaviors for Information Agents

Download Report

Transcript Software Agents: An Overview and Designing Behaviors for Information Agents

Software Agents: An Overview
by Hyacinth S. Nwana
and
Designing Behaviors for Information
Agents
by Keith Decker, Anandeep Pannu, Katia Sycara and
Mike Williamson
Presenters: Wendy Nikiforuk, Rui Lopes, Brad
Jones, and Chris Kliewer
February 10, 1999
Software Agents - Outline
•
•
•
•
Introduction to Agents & Papers - Chris
Typologies from Paper One - Brad
A Framework for Information Agents - Rui
Conclusions and the Future - Wendy
Designing Behavior For
Information Agents
• Frameworks for constructing Agents
• Behavior of basic Information Agents
• WARREN
Software Agents: An Overview
• 2 strands of Agent research
• Strand 1 1977 to 1996
– Deliberative Agents
– Macro Issues
– Research and Development
• Strand 2 1990 to 1996
– Diversification of agent types
What is an Agent?
•
•
•
•
No clear consensus on a definition
The term has been over used
Many physical forms
A component of SW or HW capable of
accomplishing tasks for its user.
Creating the Classes
•
•
•
•
Mobility
Deliberative or Reactive
Roles
Primary Attributes
– Autonomy
– Learning
– Cooperation
• Secondary Attributes
A Typology Of Agents
•
•
•
•
•
•
•
•
Collaborative
Interface
Mobile
Reactive
Hybrid
Heterogeneous Systems
Smart
Information / Internet
Collaborative Agents
• Emphasize autonomy and cooperation.
• Whole is greater than sum of the parts.
• promises
– flexible solutions to complex problems
• problems
– based on deliberative thinking paradigm
– communication and stability issues
– unclear implementation
Interface Agents
• Emphasize autonomy and learning.
• promises
– automation of mundane or regular tasks
– essentially an ‘avatar’
• problems
– Is learning mechanism valid, competent,
upgradable, defined?
– needed or desired?
Mobile Agents
• Agent is a non-static entity.
• promises
– better / more efficient use of resources
– easily coordinated and flexible asynchronous
system architecture
• problems
– few “real world” examples
– typical distributed computing problems
(transportation, security, performance, etc.)
Reactive Agents
• No internal, symbolic environmental model.
• Relatively simple & use emergent behavior.
• promises
– robust, fault tolerant, flexible, and adaptable
• problems
– unclear development methodology
– potential scalability and performance issues
Hybrid Agents
• Combination of other agent philosophies.
• Combination is better than singular type.
• promises
– combines ‘best’ of agent philosophies
– provides focused applicability of agent
• problems
– unspecified theories underlying hybrid systems
– ad-hoc design
Heterogeneous Agent Systems
• System of different agent types.
• Focused on interoperability between agents.
• promises
– provide flexible solutions to complex problems
– provides new way approach to old problems
• problems
– communication - what language, how, etc.
– requires an standard framework
Information Agents
• Information source in support of other
agents in RETSINA framework
• Framework encapsulates much of the
reusable functionality
• Not a simple API
Functional Overview
• Three conceptual functional parts
– Current Activity And Request Information
– Local Information Database
– Problem Solving Plan Library
Reusable Behaviors
• Approaches to Accomplishing a Goal
• Information Agent Behaviors
– Advertising
– Message Polling
– Information Monitoring
– Query Answering
– Cloning
Agent Architecture
• Building Blocks for Agent Behaviors
– Planning
• higher level tasks broken down into lower level
primitive actions
– Scheduling
• dynamically decides which primitive action gets run
next
Agent Architecture 2
– Execution Monitoring
• prepares, monitors and completes agent’s next
intended action
– Local Agent Infobase
• local data store defined by an ontology, a set of
attributes, a language, and a schema
Odds and Ends
• Multi-Source Information Agents
– One agent assumes responsibility for many
others
• WARREN
– Six? information agents
•
•
•
•
two stock ticker agents
news agent
current and historical sales information agent
company annual report agent
What Agents Are Not
• Expert Systems
• Modules in distributed Computing
– rarely smart
– low level messaging
– run at symbol level
Societal Issues
• For success in the future, there are several
societal issues which must be handled
–
–
–
–
–
–
Privacy
Responsibility
Legal
Ethical
Etiquette
Restricting agents
Conclusions
• Agents can work independently, but more
powerful when they work together.
• Truly smart or intelligent agents to not exist
• Fear of agents
• Evolutionary not Revolutionary
• Can exploit diverse and distributed
knowledge
Conclusions
• Agents are not a passing fad
–
–
–
–
‘agent’ not ‘intelligent agent’
have papers reviewed by a colleague
do not oversell the domain
be critical of the progress