Cached from
http://www.thefreelibrary.com/Autonomy+Puts+Industry's+First+Intelligent+Agent+Software+on+the+Web.-a018742794

Autonomy Puts Industry's First Intelligent Agent Software on the Web.


PALO ALTO, Calif.--(BUSINESS WIRE)--Oct. 8, 1996--Autonomy Inc., today announced the industry's first software applications to be built using intelligent agent technology. 

The two products, Autonomy Web Research Agent and Autonomy Press Agent, make use of intelligent agents -- software that goes out into the world and performs tasks on behalf of the user. The technology will change the way people use the World Wide Web and other electronic resources by making information searches faster, easier and more precise and more personal. 

The applications can be downloaded free of charge today from http://www.agentware.com and are fully supported via e-mail. Beta versions have been available at the site since August. 

Both packages are based on Autonomy's personalized intelligent agents, which -- like a smart bloodhound -- are first "trained" by the user, then unleashed to sniff our useful information from a variety of information resources, such as the World Wide Web and corporate intranets. Once trained, Autonomy's agents represent a user's interests on an ongoing basis.

Intelligent agents grow smarter over time, becoming more discriminating as users select the pages of most interest to them. Users also can train agents on specific topics and share or exchange them with other users. 

Autonomy's intelligent agents represent a major advancement over the current state of the art. "Whereas today's agent technology is adapted for very narrow applications, such as playing on-line matchmaker, our agents serve in many roles while maintaining your personal information in the privacy of your own computer," said Richard Gaunt, Autonomy technical director. 

In the future, Autonomy's agents will sort and prioritize electronic mail, notifying the user of important incoming mail and separating "junk e-mail" into a separate listing. They will screen out objectionable content for children, collaborate with other agents, and operate in a client/server network environment. 

Trainable Agents That Get Smarter Over Time 

Autonomy's agents are not only trainable, but get smarter with use -- employing the results of one search in subsequent searches. "For example, an agent trained on the phrase `infidelity in the Kennedy family' retrieves many relevant documents using these key words," said Gaunt. 

"And when the user `retrains' the agent by selecting the most interesting documents, subsequent retrievals become far more precise than any conventional technology can deliver. In this case, the agent now understands that, for example, the user is interested not just in `infidelity,' but `misogyny' and `adultery' as well. 

"A trained agent stands ready to repeat this search whenever you wish, on any collection of network information. The agent can do its work in the foreground while you watch, or in the background while you do something else," Gaunt said. 

Cambridge Neurodynamics developed its intelligent agent technology over a six year period based on research from Cambridge University. Both Gaunt and founder Dr. Michael Lynch are leading authorities in the development of neural network applications. 

These applications mimic the pattern-finding abilities of the human mind to pinpoint inter-relations within large amounts of information. Autonomy's intelligent agents use neural networking to search for patterns of information, rather than specific words or phrases, thereby distinguishing relevant from irrelevant information. 

Train, Unleash and Retrieve 

Autonomy Web Research Agent and the Autonomy Press Agent search the broad Web and publication Web sites, respectively. Both run under Microsoft Windows and work in conjunction with standard browsers like Netscape Navigator and Microsoft Explorer. Both packages give users an unlimited number of agents (represented metaphorically on the screen as a pack of hounds), which are trained with a phrase, concept or question. 

Any agent can then be unleashed simply by dragging its icon onto the Web button. Users "retrain" agents by entering new concepts or by selecting the most relevant pages from Autonomy's built-in library of previously retrieved sites. 

Intelligent agents begin their search using an internal index, as well as accessing some of the Web's most popular sites. Once "on the scent," an agent searches from site to site until it finds information it considers relevant to the original query. While users need not keep track of an agent, its progress is charted on a "ticker tape" display that shows what material the agent is evaluating at the moment. 

A location map represents where on the web the agent has been looking for information and color-codes the relevance of each site. Relevant sites are color-coded and sorted for relevancy. Users can limit the time of a search, set the "curiosity level" governing how deep into a particular site an agent will burrow, and specify the retrieval of pictures as well as text. 

The Web Research Agent delivers a list of Web site links, complete with preview information, that have been selected based on the training criteria. Press Agent works similarly, compiling a personal newspaper from a variety of sources. For sites requiring a subscription, Press Agent can be configured to automatically log in with passwords and registration details. 

About Autonomy Inc. 

Autonomy Inc. is the U.S. operation of Autonomy Corp. PLC, which was founded in Cambridge, U.K., by Dr. Michael Lynch. Spun out of Cambridge Neurodynamics, a world leader in the commercial application of neural networks, fuzzy logic and related technology, Autonomy Inc. develops intelligent agent applications for online environments. Autonomy Inc. is privately held, with U.S. offices in Palo Alto and Roseland, New Jersey. 

CONTACT: Access Communications 

Robin Rootenberg, 415/904-7070 ext. 282 

rrootenberg@accesspr.com 

Gwen Marker, 415/904-7070 ext. 289 

gmarker@accesspr.com

COPYRIGHT 1996 Business Wire