Transcript Shah_Malalur - Computer Science

Autonomous Intelligent
Research Robots
Maulik Shah & Sandeep Malalur
May 01, 2000
What are Autonomous Intelligent
Research Robots?
 Autonomous?
 Lie within the environment
of the system
 Employ Artificial Intelligence
(AI) techniques
 Act on complex and
dynamic environment
between the user and the
system resources.
Autonomous Intelligent
Research Robots? (Contd.)…
 Intelligent?
 Assists user with the applications assigned
as a set of goals and tasks.
 Can make decisions based upon simulation
of needed solutions.
 Can determine choices based upon
experience.
How they work?
 Internet lacks semantic information
 HTML specifies …how to display info without
specifying its meaning
 Internet is a dynamic structure… page
contents are not static
 …hence these robots’ perception are that web
pages are written in HTML, I/P-O/P of clientserver side programs (applets, scripts…) and
interacts with user with a natural language
processing unit (NPL)
How they work?(Contd.)…
 Lie within the environment of the system
 New Agent – autonomous clustering
 Changes goal-oriented behavior based on
neutral clustering to find other agents on the
LAN.
Reference: http://www.sce.carleton.ca/netmanage/docs/AgentsOverview/ao.html
How do they communicate?
 Using languages
 Blackboard: Read and write messages in shared
location.
 Knowledge Query and Manipulation Language
(KQML): Protocol for information and knowledge
exchange.
 Knowledge Interchange Format (KIF)
 COOL: Structured conversation based on KQMLused for co-ordination with other agents.
Softbots
 Software Robots
 Perform tasks on
user’s behalf

Finding information
 Filtering email
 Scheduling meetings
 Effectors
 mv, ftp…
 Sensors
 ls, finger…
Reference: http://www.cis.udel.edu/agents98/
Potential of Robots
 Crawl from one server to another,
compiling lists of URLs to find
information and report back.
 Traverse Web’s hypertext structure by
retrieving a document and recursively
retrieving others which are referenced.
 Maintain a hypertext structure that can
be checked for “dead links”.

Example: CERN HTTPD servers log failed
requests caused by dead links, with the
preference to the page where dead link
occurred.
Potential (Contd.)…
 Performs statistical analysis of
retrieved documents and provides
resource discovery database.
 Operate in parallel, resulting in high
use of available bandwidth.
 Great potential in “Data Mining”
Potential (Contd.)…
 Data Mining
 Process of finding patterns in enormous
amounts of data.
 Requires series of searches.
 Makes decisions based on experience to
perfect complex searches.
Reference: http://pattie.www.media.mit.edu/people/pattie/ECOM/index.htm
Potential (Contd.)…
 Artificial Intelligence
 Develops software capable of processing
information on its own without need of
human intervention.
 Interoperability
 Major initiatives that will make agents
ubiquitous:
OPS (open profiling standard)
 XML (extended markup language)
 JEPI (joint electronic payment initiative)

Applications of Autonomous
Intelligent Robots
 Use in E-Commerce
 Support for Wireless Application Protocol
(WAP) – enabled net devices and wireless
handheld devices using Network Query
Language (NQL).
 Systematically search commercial sites on
the web and capture detailed data (Online
Media Network Intelligent Agent – OMNIAC).
Applications (Contd.)…
 Personal Research Assistants
 User assigns set of rules and preferences to
the agent.
 Agent acts as an assistant by
communicating and understanding user’s
preference and achieves assigned tasks.
 How?
Scans the database and information resources.
 Delivers summaries and information on certain
topics base on requests.
 Examples: Open Sesame and Microsoft’s Bob.

Applications (Contd.)…
 Information Management Assistants
(Resource Discovery)

Behaves similar to the Personal Research
Assistant.
 But they work in complex and dynamic
environment between the user and system
resources.
 Pre-determines data resources.
 Example: Oracle’s ConText.
Applications (Contd.)…
 As Robot Systems in Engineering
Applications

Lie in the engineering and technology field.
 These include space and marine
applications.
 Artificial intelligence (especially agent
architectures, machine learning,
planning, distributed problem-solving),
information retrieval, database and
knowledge-base systems, and
distributed computing.
Applications (Contd.)…
 Internet-based information systems,
adaptive (customizable) software
systems, autonomous mobile and
immobile robots, smart systems (smart
homes, smart automobiles, etc.),
decision support systems, and
intelligent design and manufacturing
systems.
Disadvantages ?
 They are domain and task dependent.
 Solution?
 Different knowledge base for different domains.
 Uniform way of using the knowledge domain.
 Different specialized agents for each domain pair.
 Need for uniform framework.
 Is it possible?
Problems Encountered
 Compatibility with the WWW?
 NO!
 WWW uses client-server orientation,
while agents require peer-to-peer
communication.
 WWW
is oriented around data transport
through networks.
 They require structure reflecting task
level semantics.
Problems (Contd.)…
 Bootstrapping






New agent performs autonomous clustering.
Changes its behavior based on results of
mutual clustering.
Unable to locate existing agents and initiate
conversation.
Slows down server performance.
Communication with other agents existing in
the network is a MAJOR problem.
Remedy…ever developing architecture and
data paths could resolve the problem in the
near future.
Problems (Contd.)…
 Client side robot
 Cannot fix bugs.
 Cannot provide new efficient advantageous
facilities.
 Cannot add knowledge of problem areas.
 Technical issues of vigilance, thrift,
secrecy and user privacy.

Lower level tasks - implemented by sensor
based controllers (embedded within the
overall system architecture).
 Integrates real-time operation & aspects of
AI.
Current Developments and
Challenges
 Making decisions based on information
availability.
 Stability and Performance Issues.
 Interoperability and Communication.
 Collaborative Research Systems.
 Setting up systems which are user
customized.
 Trust and Competence Issues.
 Multi-agent systems which use
heterogeneous architecture.
Developments and Challenges
(Contd.)…
 State of the Art (Information
Infrastructure Context)





Connectivity (e.g. Internet/ WWW)
Growing digital content (size, complexity,
modality)
Computation intensive (High-performance
computers)
Limited Access Modality (Conventional
Channels, plugs)
Distributed Resources
Developments and Challenges
(Contd.) …
 State of the Art (Contd.) …
 Growing heterogeneity
 Economies of scale (speed, bandwidth
issues)
Reference: http://www.sce.carleton.ca/netmanage/docs/AgentsOverview/ao.html
Examples of Autonomous
Intelligent Research Robots
 Information Visualizer
 Experimental system where the user and the
agent system perform communication,
monitor events and look after tasks for
information retrieval.
 It attempts to utilize graphics technology to
lower the cost of finding and accessing
information.
 Harvest
 Resource discovery robotic system which
allows the users a much controlled way of
indexing the Web.
Examples (Contd.)…
 ALIVE
 Entertainment intelligent agent system which
allows users to enter a virtual world and
interact with full-body images and animated
agents.
 Calendar Apprentice (CAP)
 Personal assistant agent system which
learns about user preferences and habits
and manages his calendar operations. The
system learns about the user's scheduling
and preferences from experience and serves
as a personal software assistant.
Examples (Contd.)…
 BullsEye
 IntelliSeek's BullsEye is a search and
retrieval agent that allows you to search
multiple search engines (450+) through a
single interface.
 Integrates many of the tasks power
searchers use under one intuitive interface
and combines targeted meta-searching with
full text analysis (via the Verity 97 toolkit),
and filtering, for more relevant results.
Works behind a firewall. Windows only.
 … and a lot more….
 KnowMan
 Intelligent software for creating Internet
agents. The software comes in complete
product packages and as easily embedded
components.
 Mind It
 Formerly "URL-Minder", keeps track of a
specified URL and e-mails you (or your
readers) when it changes. Can also embed
an e-mail form in your web sites so users
can be notified when your pages change.
 MOMspider
 A web-roaming robot written in Perl 4 that
automatically checks web sites for bad URLs
and indexes sites.
Amazon.com
PersonaLogic.com
Barnes and Noble
(Recommendation Agents)
A Case Study: ShopBot
 Internet agent developed by researchers at
Washington University.
 Still a prototype.
 Goal-oriented and assist human user in
shopping(virtually at different sites) and
presents information extracted from those
sites.
 Exhibits learning by example and off-line
learning.
ShopBot (Contd.)…
 Two phases of operation Learning phase




Learns how to shop at different sites specified by its
creators. (Disadv. Limits flexibility)
Sites must support search forms for ShopBot to learn.
Retrieves info on learning-by-example technique.
Real-time online comparison-shopping phase


Extracts info based on human user query.
Uses experience and learned extraction techniques to
compare prices at different vendor sites.
ShopBot (Contd.)…
 Advantages

Product-independent



Learns description of a particular domain.
No NLP required.
No Natural language interface required.
 Disadvantages
 Domain independent in one domain.
 Just learns to shop…not efficient shopping ability.
 Shop at sites that have search forms.
 End user cant specify example sets.
 ShopBot is used at http://www.jango.excite.com
Conclusion
 Global resource discovery
 Provide data retrieval over LANs and WANs.
 Largely incompatible with the web, but
future developments in network
architecture and data path will resolve
the problem.
 Multi-user domains

Information retrieval and storage (Metadata)
 Internet search (algorithms, indexes, UIs,
spiders)
 Document/file management
 Storage/repositories (text, hypermedia)
Conclusion(Contd.)…
 Used for collaboration, electronic commerce,
finding, gathering, filtering, management,
planning, resource allocation, network
management, diagnostics, as personal
assistants, process workflow etc.
 Goal identification and planning – initial phase.
 NLP techniques are not effectively used in the
Internet agent framework.
References

Les Gasser, 1998. “Agents in Rational Structure of Scientific
Research”, National Science Foundation.
http://www.cis.udel.edu/agents98/LG-agents98-talk/ppframe.htm

Edited by Gray & Caldwell, 1996. "Advanced Robotics & Intelligent
Machines," IEE Control Engineering Series, pp. xvi-xx.

Pattie Maes, 1995. "Artificial life meets entertainment: Lifelike
autonomous agents," Communications of the ACM, vol. 38, no. 11,
pp. 108-114.

Witold Jacak, 1998. "Intelligent Robotic Systems," IFSR
International Series on Systems Science and Engineering, Vol 14,
pp 1-4, 10.

The Agents' Agents
www2.computerworld.com/home/online9697.nsf/All/970630agents
References(Contd.)…

Anonymous, 1994. "The age of the Intelligent Agent," Insurance
Systems Bulletin, Vol. 9, No. 10, pp. 4-5.

Using an Intelligent Agent to Enhance Search Engine
Performance: by James Jansen
http://131.193.153.231/issues/issue2_3/jansen/

Is it an Agent, or just a Program?: A Taxanomy for Autonomous
Agents by: Stan Franklin and Art Graesser, Institute for Intelligent
Systems, University of Memphis.
http://www.msci.memphis.edu/~franklin/AgentProg.html

AARIA: Autonomous Agents at Rock Island Arsenal
http://www.aaria.uc.edu/overview.html
References(Contd.)…
 Agent-Based Engineering, the Web, and Intelligence by:
Charles J. Petrie, Stanford Center for Design Research.
http://cdr.stanford.edu/NextLink/AID.html
 Chronological overview of expected/ predicted
developments.
http://www.broadcatch.com/agent_thesis/h622.htm
 The Agent Technique
http://www.broadcatch.com/agent_thesis/h62.htm
 The User
http://www.broadcatch.com/agent_thesis/h63.htm
References(Contd.)…

Software Agents and the Future of Electronic Commerce
http://pattie.www.media.mit.edu/people/pattie/ECOM/index.htm

Intelligent Systems: Robots, Autonomous Agents, Agent Societies
http://www.cs.byu.edu/info/mikeg/research/Research.html

Hyacinth S. Nwana, “Software Agents: An Overview”
http://www.sce.carleton.ca/netmanage/docs/AgentsOverview/ao.html