ICT619 Intelligent
Systems
Topic 8: Intelligent Agents
Intelligent Agents








What is an intelligent agent?
Why intelligent agents?
What intelligent agents can do for us
Characteristics of a good agent
Types of agents
Building intelligent agents
Intelligent agents in E-Commerce
Intelligent agent design - state-of-the-art
and future
ICT619
2
What is an intelligent agent?
Underlying concept  An autonomous computational entity designed to perform a
specific task, without direct initiation and continuous monitoring on
part of the user
 Emerged in the last 15 years or so
 Distinct from conventional programs, in that it is automatic
Additional properties:
 Some level of intelligence (based on any AI technology from fixed
rules to learning engines) for decisions and/or adaptation to
environmental change
 Acts reactively, but also proactively
 Social ability - communicates with user, system, other agents as
required
 Might cooperate with other agents to carry out complex tasks
 Agents might move from one system to another to access remote
resources and/or meet other agents
ICT619
3
What is an intelligent agent?
(cont’d)
 Intelligent agents (also called “software agents”) do not
necessarily possess all these possible features
 Wide range of variation in capabilities:
 Some perform tasks individually while others are
cooperative
 Some are mobile- able to move across a network,
others are not
 Most communicate via coded messages or even
natural language, some don't communicate at all
 Multiple agents work in groups or swarms to solve
problems collectively, some work as individual units
 Not all agents learn and adapt themselves
ICT619
 Robots are physically embodied agents
4
Why intelligent agents?
 More and more everyday tasks becoming computer-based
 An increasing number of untrained users using computers
 Current human-computer interfaces require users to initiate all
tasks and monitor them - manually
 Intelligent agents engage in a cooperative process with the user to
leverage the effectiveness and efficiency of human-computer
interaction
 Staggering growth in information availability
 Intelligent agents can be a tool for relieving the user of this
information overload
 Intelligent agents can act as personal assistants to the user to
manage information
 Might one day take over routine tasks in personal management
such as appointments, meetings and travel arrangements
ICT619
5
What intelligent agents can do for
us




Carry out tasks on the user’s behalf
Train or teach the user
Help different users collaborate
Monitor events and procedures
 Specifically, intelligent agents can help us with
 Information retrieval
 Information filtering
 Mail management
 Recreational activities – selection of
books, music, holidays
 Booking of meetings, hotels, tickets
ICT619
6
What intelligent agents can do for
us (cont’d)
Information filtering agent
 One type is the selection of articles from a continuous
stream to suit particular user needs
 User can create “news agents” and train them by giving
positive or negative feedback for articles recommended
 The use of key words alone can be restrictive
 Underlying semantics must be extracted for more
effectiveness
 Eg VPOP Technologies' Newshub - an automated,
agent-based web news feeder service, which delivers
customised updates of stories from major news outlets
every 15 minutes
ICT619
7
What intelligent agents can do for
us (cont’d)
Electronic mail agent
 Assist users with electronic mail
 Learn to prioritize, delete, forward, sort and archive
mail messages on behalf of the user
 May use intelligent system techniques like case-based
reasoning
 Can associate a level of confidence with its action or
suggestion
 Use of “do-it” and “tell-me” thresholds set by user
 May involve multi-agent collaboration
ICT619
8
What intelligent agents can do for
us (cont’d)
Selection agents for entertainment
 Conversational agents show potential for
becoming popular and commercially
successful eg Cybelle, ALICE
Hi, I am Cybelle.
What is your name?
 Use “social filtering” – correlation between different
users to make recommendations on books, CDs, films
etc.
 So, if user A liked items X and Y, and user B liked item
X and Z, then item Z may be recommended for user A
 Amazon.com has been using this system for years ->
ICT619
9
ICT619
10
What intelligent agents can do for
us (cont’d)
Some other current and emerging applications of
intelligent agents:
 air traffic control
 air craft mission analysis
 control of telecommunications and network systems
 provision and monitoring of medical care
 monitoring and control of industrial processes
 on-line fault diagnosis and malfunction handling
 supervision and control of manufacturing environments
 transactions management in banks and insurance
companies
 E-commerce, tourism
ICT619
11
Characteristics of a good agent
Action
 Agent must be able to take some action and not just
provide advice
 Present state of web technology limits capability of
Internet agents
- still no standard interface for agents, but agent
communication languages such as ACL and KQML
might win out
 As the Internet becomes more agent-friendly, more
capable agents will emerge
Autonomy
 An agent can be much more useful if it can act
autonomously
 The right level of autonomy
for a task must be found
ICT619
12
Characteristics of a good agent
(cont.)
Communication
 Must communicate well with the user
 Should understand user’s goals, preferences and constraints
 Useful communication requires shared knowledge on
 language of communication
 problem domain
Example Problem: Web search engines
 accept key words and phrases (some knowledge of the
language)
but
 understand nothing about the documents they retrieve (no
domain knowledge)
 Solution: provision of a machine-readable ontology
- a definition of a body of knowledge including its
components and their relationships
ICT619
13
Characteristics of a good agent
(cont.)
Adaptation
 Can gain user confidence by learning user preferences
 ML techniques such as ANNS, GAs or CBR can be
used
 Adapting to user preferences can be also achieved by
using data mining techniques such as clustering
 Agent forms clusters of users with similar features
 User's needs can then be anticipated by placing the
user in one of these clusters and analysing the cluster
 Social problem solving method, similar to Amazon
recommendations
ICT619
14
Types of agents
 Based on operational characteristics and
functional objectives:
 Collaborative agents
 Work together to
- integrate information and
- negotiate with other agents to resolve conflict
- Provide solutions to inherently distributed problems,
e.g., air traffic control
 Reactive agents
 Act by stimulus-response to the current state of
the environment
 Each reactive agent is simple and interacts
with others in a basicICT619
way
15
Types of agents (cont’d)
Interface agents
 Provide user support and assistance
 Cooperate with user in accomplishing some task in an
application.
 Interface agents learn:




by observing and imitating the user
through receiving feedback from the user
by receiving explicit instructions
by asking other agents for advice (from peers)
 Examples:
 Personal assistants performing information filtering,
email management.
ICT619
16
Types of agents (cont.)
Mobile agents
 Programs that migrate from one machine to another.
 Execute in a platform-independent execution environment, like
Java applets running on a Java virtual machine
 Practical but non-functional advantages:
 Reduced communication cost
 Asynchronous computing (when you are not connected)
ICT619
17
Types of agents (cont.)
Two types of mobile agents:
 One-hop mobile agents (migrates to one other
place)
 Multi-hop mobile agents (roam the network
from place to place)
Example applications:
 Distributed information retrieval
 Telecommunication network routing
ICT619
18
Types of agents (cont.)
Information agents
 Manage information
 Manipulate or collate information from many distributed
sources.
 Can be mobile or static.
 Examples:
 BargainFinder compares prices among Internet stores for
CDs
 Jasper works on behalf of a user or community of users and
stores, retrieves and informs other agents of useful
information on the WWW
ICT619
19
Types of agents (cont.)
Multiple agent systems
 Consist of collections, or swarms, of simple agents that
interact with each other and the problem environment
 Can be mobile or static, same or different agents
 Complex patterns of behaviour emerge from collective
interaction
 Examples:
 Swarm of bees finds an optimal location for the hive
 xxxx
ICT619
20
ICT619
21
Building intelligent agents
Two main problems to overcome:



Competence
 How do we build agents with the knowledge needed to decide
 when to help the user
 what to help the user with, and
 how to help the user?
Trust
 How to guarantee user comfort (and protection!) in
delegating tasks to the agent
Approaches to building agents
1. User-programmed agents - write specialised scripts
2. Knowledge-based agents
3. Machine-learning approach
ICT619
22
Building intelligent agents (cont’d)
 The main problem with user-programmed approach
- requires high level of user competency
- user must be able to
 Recognise opportunity for employing an agent
 Take initiative to create an agent
 Impart specific knowledge to agent by codifying it in a
special language
 Maintain agent’s knowledge by updating rule base with
time
 The issue of trust is then reduced to users’ trust in
their own programming skills
ICT619
23
Building intelligent agents (cont.)
In the knowledge-based approach,
 The agent is supplied with knowledge about
the application and user
 At run-time, agent uses the knowledge to
recognise user’s plans and find opportunities
to contribute to them
 Example of knowledge-based agent: the
UCEgo - designed to help users solve
problems in using the UNIX operating system.
ICT619
24
Building intelligent agents (cont.)
Problems with knowledge-based approach  Both competence and trust are issues of concern
 The problem of competence relates to the competence
of the knowledge engineer
 Knowledge-base is fixed and cannot be customised to
specific user needs
 User’s trust is affected as agent is programmed by
someone else
ICT619
25
Building agents – the machine
learning approach


Metaphor of a personal office assistant
Agents start with minimum knowledge and learn
from:
1.
2.
3.
4.


Observation and imitation of user
User feedback – direct, indirect
Training by user
Other agents
User can build up model of agent decision making –
more trust
Agent capable of explanation
ICT619
26
Development of an agent through
learning
ICT619
27
Building agents – the machine
learning approach
Advantages:

Less work from end-user and developer

Agent customises to user/organisation
habits/preferences

Helps distribute know-how and competence
among different users
Some examples:

Agent for e-mail handling

Agent for meeting scheduling

Agent for electronic news filtering

Agent for recommending books, music
ICT619
28
Intelligent agents in E-commerce
 Rapid growth continues in e-commerce
 Information about products and vendors is easily
accessible
 But transactions are still mostly not automated
 Six fundamental stages of the buying process:






Need identification
Product brokering
Merchant brokering
Negotiation
Purchase and delivery
Product service and evaluation
ICT619
29
Intelligent agents in E-Commerce
(cont’d)
 In the need-identification stage, agents can help in
purchases that are repetitive or predictable
 Continuously running agents can monitor a set of
sensors or data streams and take actions when certain
pre-specified conditions apply
 Agents can use rule-based systems or data mining
techniques to discover patterns in customer behaviour
to help customers find products
ICT619
30
Intelligent agents in E-commerce
(cont.)
 In the merchant brokering stage, on-line
shopping agents can look up prices for a
chosen product for a number of merchants
 Many business-to-business transactions are
canvassed
 In a web auction, customers are required to
manage their own negotiation strategies
 Intelligent agents can help with this
ICT619
31
Examples of on-line shopping
framework with agent mediation
PERSONA
Logic
Firefly Bargain Auction
Finder
Bot
*
*
Jango Auction
Bot
[email protected]
Need
identification
Product
brokering
Merchant
brokering
*
*
*
*
Negotiation
*
*
*
*
Payment &
delivery
Service &
Evaluation
ICT619
32
Examples of on-line shopping
framework with agent mediation
ICT619
33
Examples of on-line shopping
framework with agent mediation
ICT619
34
Examples of on-line shopping framework
with agent mediation (cont’d)
 Software agents are helping buyers and sellers cope
with information overload and expedite the online
buying process
 Agents are creating new markets (eg, low-cost
consumer goods) and reducing transaction costs
 Use of agents in e-commerce still at an early stage
 Visit
http://agents.umbc.edu/Applications_and_Software/Ap
plications/Electronic_Commerce/index.shtml
for more
ICT619
35
Intelligent agent design - state-ofthe-art and future
 Few agents are available with all the desired
characteristics
 Agent technology still in experimental stage
 Autonomy and mobility already achievable
 Example: Java applets which execute independently
across networks
 But autonomy limited so far in practical use due to the
agent-unfriendliness of the current web technology
ICT619
36
Intelligent agent design - state-ofthe-art and future (cont’d)
 A major limiting factor is lack of ontologies
essential for effective communication
 Building and maintaining ontologies remains a
major challenge
 Some of the proposed capabilities to be
developed in future intelligent agents include:
 Learning as well as reasoning, which are
characteristics of machine intelligence
 Interacting with the external environment through
sensors
ICT619
37
REFERENCES
 Chin, D., Intelligent Interfaces as Agents. In Intelligent User
Interfaces, J. Sullivan and S. Tyler(eds), ACM Press, New
York, 1991.
 Hendler, J., Making Sense out of Agents, IEEE Intelligent
Systems, March/April 1999, pp.32-37.
 Hendler, J., Is There an intelligent Agent in Your Future?
http//www.nature.com/nature/webmatters/agents/agents.html
 Maes, P., Agents that Reduce Work and Information Overload,
Communications of the ACM, Volume 37 , Issue 7 (July
1994), pp. 30-40.
 Maes, P., Agents that Buy and Sell, Communications of the
ACM, Volume 42 , Issue 3 (March 1999), pp. 81-91.
 Sheth, B. and Maes, P. Evolving Agents for Personalized
Information Filtering. In Proceedings of the Ninth Conf. on
Artificial Intelligence for Applications. IEEE Computer Society
Press, 1993
 UMBC Agent News http://agents.umbc.edu/agentnews/current/
 http://www.agentland.com/
ICT619
38
Descargar

ICT619 Intelligent Systems