Transcript Document
Agents, Infrastructure,
Applications and Norms
Michael Luck
University of Southampton, UK
Overview
Monday
• Agents for next generation computing
AgentLink Roadmap
Tuesday
• The case for agents
• Agent Infrastructure
Conceptual: SMART
Technical: Paradigma/actSMART
• Agents and Bioinformatics
GeneWeaver
myGrid
Wednesday
• Norms
• Pitfalls
Agent Technology: Enabling
Next Generation Computing
A Roadmap for Agent Based Computing
Michael Luck, University of Southampton, UK
[email protected]
Overview
What are agents?
AgentLink and the Roadmap
Current state-of-the-art
Short, medium and long-term predictions
Technical challenges
Community challenges
Application Opportunities
What is an agent?
A computer system capable of flexible,
autonomous (problem-solving) action,
situated in dynamic, open,
unpredictable and typically multi-agent
domains.
What is an agent?
A computer system capable of flexible,
autonomous (problem-solving) action,
situated in dynamic, open,
unpredictable and typically multi-agent
domains.
control over internal state and over own
behaviour
What is an agent?
A computer system capable of flexible,
autonomous (problem-solving) action,
situated in dynamic, open,
unpredictable and typically multi-agent
domains.
experiences environment through
sensors and acts through effectors
What is an agent?
A computer system capable of flexible,
autonomous (problem-solving) action,
situated in dynamic, open, unpredictable and
typically multi-agent domains.
reactive: respond in timely fashion to
environmental change
proactive: act in anticipation of future goals
Multiple Agents
In most cases, single agent is insufficient
• no such thing as a single agent system (!?)
• multiple agents are the norm, to represent:
natural decentralisation
multiple loci of control
multiple perspectives
competing interests
Agent Interactions
Interaction between agents is inevitable
• to achieve individual objectives, to manage interdependencies
Conceptualised as taking place at knowledgelevel
• which goals, at what time, by whom, what for
Flexible run-time initiation and response
• cf. design-time, hard-wired nature of extant
approaches
AgentLink and the Roadmap
What is AgentLink?
Open network for agent-based
computing.
AgentLink II started in August 2000.
Intended to give European industry a
head start in a crucial new area of IT.
Builds on existing activities from
AgentLink (1998-2000)
AgentLink Goals
Competitive advantage through
promotion of agent systems technology
Improvement in standard, profile,
industrial relevance of research in
agents
Promote excellence of teaching and
training
High quality forum for R&D
What does AgentLink do?
Industry action
• gaining advantage for Euro industry
Research coordination
• excellence & relevance of Euro research
Education & training
• fostering agent skills
Special Interest Groups
• focused interactions
Information infrastructrure
• facilitating AgentLink work
The Roadmap: Aims
A key deliverable of AgentLink II
Derives from work of AgentLink SIGs
Draws on Industry and Research
workpackages
Aimed at policy-makers, funding
agencies, academics, industrialists
Aims to focus future R&D efforts
Special Interest Groups
Agent-Mediated Electronic Commerce
Agent-Based Social Simulation
Methodologies and Software Engineering for
Agent Systems
Intelligent Information Agents
Intelligent and Mobile Agents for Telecoms and
the Internet
Agents that Learn, Adapt and Discover
Logic and Agents
The Roadmap: Process
Core roadmapping team:
• Michael Luck
• Peter McBurney
• Chris Preist
Inputs from SIGs: area roadmaps
Specific reviews
Wide consultation exercise
Collation and integration
State of the art
Views of Agents
To support next generation computing
through facilitating agent technologies
As a metaphor for the design of complex,
distributed computational systems
As a source of technologies
As simulation models of complex realworld systems, such as in biology and
economics
Agents as Design
Agent oriented software engineering
Agent architectures
Mobile agents
Agent infrastructure
Electronic institutions
Agent technologies
Multi-agent planning
Agent communication languages
Coordination mechanisms
Matchmaking architectures
Information agents and basic ontologies
Auction mechanism design
Negotiation strategies
Learning
Links to other disciplines
Philosophy
Logic
Economics
Social sciences
Biology
Application and Deployment
Assistant agents
Multi-agent decision systems
Multi-agent simulation systems
IBM, HP Labs, Siemens, Motorola, BT
Lost Wax, Agent Oriented Software,
Whitestein, Living Systems, iSOCO
The Roadmap Timeline
Dimensions
Sharing of knowledge and goals
Design by same or diverse teams
Languages and interaction protocols
Scale of agents, users, complexity
Design methodologies
Current situation
One design team
Agents sharing common goals
Closed agent systems applied in specific
environment
Ad-hoc designs
Predefined communications protocols
and languages
Scalability only in simulation
Short term to 2005
Fewer common goals
Use of semi-structured agent
communication languages (such as FIPA
ACL)
Top-down design methodologies such
as GAIA
Scalability extended to predetermined
and domain-specific environments
Medium term 2006-2008
Design by different teams
Use of agreed protocols and languages
Standard, agent-specific design methodologies
Open agent systems in specific domains (such
as in bioinformatics and e-commerce)
More general scalability, arbitrary numbers and
diversity of agents in each such domain
Bridging agents translating between domains
Long Term 2009 Design by diverse teams
Truly-open and fully-scalable multi-agent
systems
Across domains
Agents capable of learning appropriate
communications protocols upon entry to a
system
Protocols emerging and evolving through
actual agent interactions.
The Roadmap Timeline
Technological Challenges
Technological Challenges
Increase quality of agent systems to
industrial standard
Provide effective agreed standards to
allow open systems development
Provide infrastructure for open agent
communities
Develop reasoning capabilities for
agents in open environments
Technological Challenges
Develop agent ability to adapt to
changes in environment
Develop agent ability to understand
user requirements
Ensure user confidence and trust in
agents
Industrial Strength Software
Fundamental obstacle to take-up is lack of
mature software methodology
• Coordination, interaction, organisation, society joint goals, plans, norms, protocols, etc
• Libraries of …
agent and organisation models
communication languages and patterns
ontology patterns
CASE tools
AUML is one example
Industrial Strength Software
Agreed Standards
FIPA and OMG
• Agent platform architectures
• Semantic communication and content
languages for messages and protocols
• Interoperability
• Ontology modelling
Public libraries in other areas will be
required
Agreed Standards
Semantic Infrastructure for Open
Communities
Need to understand relation of agents,
databases and information systems
Real world implications of information agents
Benchmarks for performance
Use new web standards for structural and
semantic description
Services that make use of such semantic
representations
Semantic Infrastructure for Open
Communities
Ontologies
• DAML+OIL
• UML
• OWL
Timely covergence of technologies
Generic tool and service support
Shared ontologies
Semantic Web community exploring many
questions
Semantic Infrastructure for Open
Communities
Reasoning in Open Environments
Cannot handle issues inherent in open
multi-agent systems
•
•
•
•
Heterogeneity
Trust and accountability
Failure handling and recovery
Societal change
Domain-specific models of reasoning
Reasoning in Open Environments
Coalition formation
Dynamic establishment of virtual
organisations
Demanded by emerging computational
infrastructure such as
• Grid
• Web Services
• eBusiness workflow systems
Reasoning in Open Environments
Negotiation and argumentation
• Some existing work but currently in infancy
Need to address
•
•
•
•
•
Rigorous testing in realistic environments
Overarching theory or methodology
Efficient argumentation engines
Techniques for user preference specification
Techniques for user creation and dissolution of
virtual organisations
Reasoning in Open Environments
Learning Technologies
Ability to understand user requirements
• Integration of machine learning
• XML profiles
Ability to adapt to changes in environment
• Multi-agent learning is far behind single agent
learning
• Personal information management raises issues of
privacy
Relationship to Semantic Web
Learning Technologies
Trust and Reputation
User confidence
Trust of users in agents
• Issues of autonomy
• Formal methods and verification
Trust of agents in agents
• Norms
• Reputation
• Contracts
Trust and Reputation
Challenges for the Agent
Community
Community Organisation
Leverage underpinning work on similar
problems in Computer Science: Object
technology, software engineering,
distributed systems
Link with related areas in Computer
Science dealing with different problems:
Artificial life, uncertainty in AI,
mathematical modelling
Community Organisation
Extend and deepen links with other
disciplines: Economics, logic,
philosophy, sociology, etc
Encourage industry take-up: Prototypes,
early adopters, case-studies, best
practice, early training
Existing software technology
Build bridges with distributed systems,
software engineering and object technology.
Develop agent tools and technologies on
existing standards.
Engage in related (lower level)
standardisation activities (UDDI, WSDL,
WSFL, XLANG, OMG CORBA).
Clarify relationships between agent theories
and abstract theories of distributed
computation.
Different problems from related
areas
Build bridges to artificial life, robotics,
Uncertainty in AI, logic programming
and traditional mathematical modelling.
Develop agent-based systems using
hybrid approaches.
Develop metrics to assess relative
strengths and weakness of different
approaches.
Prior results from other
disciplines
Maintain and deepen links with
economics, game theory, logic,
philosophy and biology.
Build new connections with sociology,
anthropology, organisation design,
political science, marketing theory and
decision theory.
Encourage agent deployment
Build prototypes spanning organisational
boundaries (potentially conflicting).
Encourage early adopters of agent
technology, especially ones with some risk.
Develop catalogue of early adopter case
studies, both successful and unsuccessful.
Provide analyses of reasons for success and
failure cases.
Encourage agent deployment
Identify best practice for agent oriented
development and deployment.
Support standardisation efforts.
Support early industry training efforts.
Provide migration paths to allow smooth
evolution of agent-based solutions,
from today’s solutions,
Application Opportunities
Application Opportunities
Ambient Intelligence
Bioinformatics and Computational
Biology
Grid Computing
Electronic Business
Simulation
Semantic Web
Ambient Intelligence
Pillar of European Commission’s IST vision
Also developed by Philips in long-term vision
Three parts
• Ubiquitous computing
• Ubiquitous communication
• Intelligent user interfaces
Thousands on mobile and embedded devices
interacting to support user-centred goals and
activity
Ambient Intelligence
Suggests a component-oriented world
populated by agents
•
•
•
•
Autonomy
Distribution
Adaptation
Responsiveness
Demands
• Virtual organisations
• Infrastructure
• Scalability
Bioinformatics
Information explosion in genomics and
proteomics
Distributed resources include databases
and analysis tools
Demands automated information
gathering and inference tools
Open, dynamic and heterogeneous
Examples: Geneweaver, myGrid
Grid Computing
Support for large scale scientific endeavour
More general applications with large scale
information handling, knowledge
management, service provision
Suggests virtual organisations and agents
Future model for service-oriented
environments
Electronic Business
Agents currently used in first stage –
merchant discovery and brokering
Next step is real trading – negotiating deals
and making purchases
Potential impact on the supply chain
Rise in agent-mediated auctions expected
• Agents recommend
• But agents do not yet authorise agreements
Electronic Business
Short term: travel agents, etc
• TAC is a driver
Long term: full supply chain integration
At start of 2001, there were
• 1000 public eMarkets
• 30,000 private exchange
Simulation
Education and training
Scenario exploration
Entertainment
The Two Towers
Thousands of agents simulated using
the MASSIVE system
Realistic behaviour for battle scenes
Initial versions included characters
running away!
Previous use of computational
characters did not use agent
behaviour (eg Titanic).
Current State
Pivotal role in contributing to broader visions
of Ambient Intelligence, Grid Computing,
Semantic Web, etc.
European strength is broad and deep
Still requires integration, needs to avoid
fragmentation, needs effective coordination
Needs to support industry take-up and
innovation
For more information ...
Dr Michael Luck
Department of Electronics and
Computer Science
University of Southampton
Southampton SO17 1BJ
United Kingdom
Feedback sought: please send feedback!
Roadmap: www.agentlink.org/roadmap
The Book
The CD
The Agent Portal
www.agentlink.org