Intelligent Agents
Download
Report
Transcript Intelligent Agents
Intelligent Agents: Applications
MgSc 497
Chapter 19
4-22-98
Intelligent Agents: Some
Definitions
computer program which helps humans
with routine computer tasks
advanced AI programs that function in
specific environments to complete tasks
persistent software entity dedicated to a
specific purpose
computational systems which inhabit an
environment and sense and act
autonomously in this environment
Intelligent Agents: “Softbots”
intelligent software robots--concept grew
out of robotics
moving from the simulated, symbolic
changes in artificial environments to the
complexities of interacting with the physical
world
software architecture which integrated
reactions to the world with goal-based
planning
Software Agents
grew out of a desire to develop smaller and
more self-reliant code
programs running separately from the host
environment, moving across networks to
deliver information
JAVA and ActiveX are programming
languages in which software could be
developed in one place and ported to
another
What Kind of Agents Exist in AI?
physical agents (robots) are located in
places where humans find it difficult to go
–
outer space, depths of the ocean, hazardous
situations
information agents travel the network or
interactive cyberspace worlds
What is an Autonomous
Intelligent Agent?
Capable of acting on its own on the behalf
of the user in pursuit of an agenda
will it pass the Turing test?
is it software or something else?
is it just a Web agent?
Components of an Agent
owner
–
author
–
user name, process name, master agent name
development owner, person or process
account
–
point of origin for the agent
Components of an Agent
goal
–
clear statement of successful task completion
for the agent
subject description
–
details of the goal’s attributes
creation and duration
background
intelligent system
Simple vs. Intelligent Agents
a simple agent is given a task to complete
and all the rules it needs to complete the
task
an intelligent agent has the added capability
of learning and the ability to exhibit
autonomy (i.e., can take independent action)
Internet-based Software Agents
simple e-mail agents
web-browing assistants
–
ex: Letizia
FAQ agents
intelligent search agents (metasearch
engines, spiders, crawlers)
network management and monitoring
e-commerce agents
Characteristics of Intelligent
Agents: Autonomy
goal-oriented
–
collaborative
–
accepts high level requests from humans and
decides how and when to fill the request
can modify requests, clarify requests or even
refuse requests
flexible
–
modifies behavior when it encounters obstacles
and in response to its environment
Characteristics of Intelligent
Agents
operates in the background
designed to accomplish a single task
communication with other agents
automates repetitive tasks
supports conditional processing
capable of learning
Characteristics of Intelligent
Agents:
reactive to changes in environment
proactiveness
temporal continuity (running continuously)
personality (realistic human interface)
mobility (can transport itself across
Sounds like the
systems)
perfect assistant!
Intelligent Agents: Applications
air traffic control
air craft mission analysis
control of telecommunications and network
systems
transactions management in banks and
insurance companies
supervision and control of manufacturing
environments
Intelligent Agents: Applications
provision and monitoring of medical care
monitoring and control of industrial
processes
on-line fault diagnosis and malfunction
handling
Intelligent Agents: Proposed
Capabilities
reasoning and learning
sensors
interfacing with humans
inter-agent communication
DARPA’s Intelligent Integration
of Information Program
provide easy access to information in a form
needed by end users and high level
applications
intelligently reviewing, filtering, extracting,
integrating, and abstracting information
from the growing volume of data
this data is dynamically changing,
inconsistent, incomplete, difficult to search
DARPA’s Intelligent Integration
of Information Program (I*3)
data sources include knowledge bases,
databases of all kinds, Web documents, text
documents, graphics, videos, images
layer of information integration to buffer the
user from this data
describes the content, format, location, and
semantic meaning of the data
DARPA’s Intelligent Integration
of Information Program
visit the site: http://dc.isx.com/I3/
I*3 Applications
air campaign planning tool
–
reduces planning time from 48 hours to 10
minutes
advanced logistics program
–
for moving resources in an efficient manner
battlefield awareness
engineering design
Other Agent Projects
Stanford’s ABE (Agent-Based Engineering
Group)
–
–
how agent-based software can be applied to
engineering
design documentation, CAD applications
Lockheed Cosmos System
Madefast
–
collaborative design
JAVA Agent Template
fully functional template, written in JAVA
used for constructing software agents which
have a peer-to-peer network of other agents
distributed over the Internet
JAT agents have a static existence, living on
one host, but they can immigrate
agents use KQML (Knowledge Query and
Manipulation Language)
JAVA Agent Template
can be executed as standalone applications
as applets using appletviewers
will not work with Netscape Navigator
Other Agent Projects
Andersen Consulting, IBM, Lockheed
http://bf.cstar.ac.com/bf/
http://www.agents-inc.com/
http://www.cs.washington.edu/homes/glinde
n/TravelSoftBot/
ATAInstructions.html
Other Agent Projects
intelligent browsers
tracks user browser patterns and suggest
shortcuts
IBM Web Browsing Intelligence
–
checks favorite pages and updates as needed
LiveAgent
–
–
lets users record their mouseclicks and
keystrokes as they navigate the Web
the recording becomes an agent
Other Agent Projects
Firefly
–
intelligent music recommendation service
ShopBot
Distributed Intelligent Agents
interface agents are single agents with
simple knowledge and problem solving
capabilities who perform information
filtering and retrieval
softbots perform a wide range of userdelegated information finding tasks
mulitagent systems offer another alternative
to access, filter, and integrate information
extends the concept of agent
Distributed Intelligent Agents
an agent’s user could be human or another
agent
agents must be able to communicate with
each other
open society of reusable agents who selforganize and cooperate in response to task
requirements
Distributed Intelligent Agents:
Architecture
planning module uses goals and produces a
plan that satisfies the goals
–
uses a hierarchical planning method
communication and coordination modules
accepts and interprets messages from other
agents
–
–
e-mail is often used for communication
event notification services
Distributed Intelligent Agents:
Architecture
scheduling module schedules each of the
plan steps
–
a variety of scheduling heuristics can be used
execution monitoring process handles agent
reactivity considerations
–
–
when an actions fails, an exception-handling
process takes over
ex: another agent doesn’t respond
Distributed Intelligent Agents:
Architecture
belief and facts data structures contain facts
and knowledge
–
ex: user profile
ex: an information agent is monitoring the
Security APL Quote Server to watch the
prices o IBM stock
–
it will periodically compare the price to a given
threshold at which it believes the user wants to
sell
Distributed Intelligent Agents:
Applications
organizational decision-making
–
–
–
e-mail filtering
calendar management
filtering news
more complex tasks
–
–
–
distributed, collaborative meeting schedules
finding information on the Internet
distributing relevant announcements about
software, grants, etc.
Distributed Intelligent Agents:
Applications
financial portfolio management
–
–
–
multiple sources of information
multiple sources of expertise
user profile, status of user portfolio
Portfolio Management Agent
Portfolio Manager
Agent
Fundamental
Analysis
Agent
Technical Analysis
Agent
Breaking News
Agent
Analyst Tracking
Agent
Earnings Analyst
Agent
Ticker
Tracker
Market
Tracker
SEC Filings
Tracker
Economic
Indicator
Tracker
Distributed Intelligent Agents:
Applications
one more application: energy management
in Spain
Iberdrola, electric utility
transport network contains 25,000 data
points
296 lines, 294 transformers, 939 breakers
Distributed Intelligent Agents:
Applications
seven hetereogeneous agents on five
different machines for an alarm-analysis
expert system with service restoration
heterogeneous nature of the electrical
network lent itself to distributed agent
architecture
management of the network becomes much
more complex during emergencies
Distributed Intelligent Agents:
Applications
agent 1: breakers and relays supervisor
agent 2: alarms analysis agent
agent 3: blackout area identifier
agent 4: service restoration
agent 5: user interface
agents 6 and 7: control system interface
More on Agents
http://aif.wuwien.ac.at/usr/geyers/archive/iagents/vo/u1/
unit/unit.html
http://lcs.www.media.mit.edu/groups/agents
/
http://lcs.www.media.mit.edu/groups/agents
/projects/
http://trafficopter.www.media.mit.edu/proje
cts/trafficopter/
Managerial Issues
cost justification
security
privacy
industrial intelligence and ethics
agent learning
agent accuracy and liability