Content-type: text/html ~ Stephen's Web ~ User Agents

Stephen Downes

Knowledge, Learning, Community

Nov 10, 1998

Posted to NewsTrolls November 11, 1999

Logon: downes
Password: ******

Good morning. Here is your news for the day.

Three articles on the Civilization IV multi-gamer edition were posted last night. Each announed the new release date, now set at November 24, 2002. Click for more.

Stocks for companies selling hand-held agents rose on your selected Asian and European exchanges last night. The most active stock was Acti-Hand, which averaged a 3.2 percent gain. Click to review stocks. Click to review Acti-Hand.

A new restaurant will be opening downtown in December. According to their license application at City Hall, they will seat 80 people, serve beer, wine, and spirits, and will feature live entertainment. Click for application. Click to contact the owners.

Your application to study Netware 6.0 at the Novell Institute has been accepted. They request a tuition of $US 245. Click to process transaction. You have three meetings scheduled for today...

Too good to be true?

No. This is the world of user agents, a world in which your computer, on your behalf, listens to the information highway, retrieves information you want to review, and interacts with agencies and organizations on your behalf. It is a world which is already well under construction and will be coming to your desktop early in the next century.

User agents - sometimes called personal agents - are your representatives on the internet. As the ISR Agent Research Program defines them, "Personal agents are personalised interface agents, whose fundamental component is a user model, or personal profile, which they employ when undertaking their task. Profiles allow personal agents to perform tasks according to the needs and preferences of the user, turning the computer into an intelligent personal assistant. With knowledge of the user, personal agents can specifically tailor how they interact with the user."

Personal agents have been constructed to assist users with tasks such as information retrieval, information filtering, email handling and meeting scheduling. For personal agents to provide a worthwhile, effective, service they must learn their user's preferences and habits. Furthermore, the profile constructed by the agent must adequately and accurately reflect the true requirements and needs of the user.

There are two major components of a user agent architecture.

  1. On the one hand, web sites must have the capacity to personalize their interactions with individual users.
  2. And on the other hand, users must employ programs which interact with these websites.

Today, the latter is a dumb instrument, the web browser. It has minimal agent capacities. It can store a set of bookmarks or favourites. It can read and write cookies. It can keep track of your browsing history. And that's about it.

Your browser of tomorrow will do a lot more for you.

Smart Browsers

The concept of intelligent browsing has been with us for a while. Consider, for example, the internet Angel, announced in 1996 (that's 14 dog years ago): "The Angel is a personalized search tool that conducts customized searches, offers recommended Web sites of interest based on a user's personal profile, and gives users the ability to store and retrieve Web sites via their 'Angel Marks,' a specialized bookmarking management application." The Angel, sadly, no longer appears to be of this Earth, however, numerous products offer personalized searching, recommendations, pre-caching, and similar user agent services. See, for example, Alexa, a browser plug-in which rates sites, offers recommendations, and tries to sell you a set of enclyclopedias.

But the technology is just getting started. Intelligent searching is about to get very intelligent. For example, in their paper, The Info Agent: an Interface for Supporting Users in Intelligent Retrieval, authors Daniela D'Aloisi and Vittorio Giannini describe an agent with the following characteristics:

  • Distributedness: a complex task is shared by several agents that cooperate to accomplish it. That happens in a user-transparent manner.
  • Standard and transparent communication: the agents exchange messages by using the KQML protocol.
  • Activeness: the agents exhibit an active and cooperative behavior since they try to understand the user's needs starting from the past and present states of affairs also when there is no explicit user's request.
  • Cooperation: the agents use the user's behavior to modify their actions and to decide what to do next.
  • Scalability: new agents able to perform different types of functionality can be incrementally added to the system.

But why stop at searching? A list of links is still just a list of links; what you want is an informative presentation of the data, a report, say, or summary. Or perhaps you want to contact people working in the area you are researching, so you can follow up with personal interviews or simply lurk in their online community. An intelligent agent should not only find you the data, it should plug you into the environment.

Enter co-NEXuS, an intelligent search agent which does all this and more. To get an idea of what it does, see the picture below, from their site:

Here's what they say about their product: "An Agent Like Device ... Co-NEXuS has been developed as an Agent Like Device. It offers its users an intelligent, dynamic environment that uses consecutive searches to devise a personal user profile with which it can improve the quality of information it provides. Points of departure for this device were user-friendliness of the technology, relevance of the information to be retrieved and motivation for creative use and communication between users."

Suppose you need to fly to China. You could spend a few hours with the travel agent checking schedules, seat openings, and hotel reservations. Or you could let your computer do all that for you while you brush up on your Manderan. Enter agencies such as Amadeus: "Amadeus is a neutral provider of travel information. Your travel plans will be automatically handled by an Amadeus Travel Partner or Travel Agency."

When you are travelling around the world you won't want to leave your agent behind. After all, you may need to take a side trip to Hong Kong, order a book, and keep up on your quote. And what if you were to get lost?

No problem. Prism Lab's Barry Smyth proposes a technology called IMPACT (Initiative for Mobile Programmable Agent-based Computing Technologies (IMPACT), which addresses this need. The heart of this proposal is the Intelligent Mobile Personalised Systems (IMPS):

  • Situated Assistants provided as Agent based applications on the latest hand held Computing Technology (Personal Data Assistants, PDAs)
  • Customised Dynamic Information Delivery based upon two important parameters.
  • User Profile: based upon user preferences, interests
  • Location Information: based on Global Positioning Systems (GPS) or Infrared (IR) Technologies.
  • Development of experience-based learning techniques such as case-based reasoning.

Of course, you'll want to find that great job which sends you to China or Hong Kong. Why not troll for openings while enjoying high tea at Marita? Swoop, your personal job search agent, "returns matching jobs to your Profile In Box and/or sends them to you via e-mail as they are found. That way you save time without missing any important information."

Websites Which Learn

The prototype website which learns was probably Firefly. Users accessing this site created a personal profile, which they could modify over time, listing their interests and preferences in such areas as music or film. Users also joined special interest communities, thus accessing bulletin boards and chat rooms filled with other users interested in the same topic.

Firefly, and other sites like it, is based on the idea that websites can learn about their users, producing demographics, promoting products, and customizing the interface. As Firefly states, "Using Firefly's products and services, businesses benefit by:

  • Developing customer relationships more efficiently
  • Personalizing sales and communications to increase profitability
  • Retaining and strengthening brand and business relationships"

To be a website which learns, a website must have three major components:

  1. A greeter or ambassador, which interacts with the user and collects data
  2. An adaptive database or knowledge engine which stores and interprets the data
  3. A reporter which constructs user profiles, demographics, and other useful information

For example, take a look at Learn Sesame, by Bowne Internet Solutions, the structure of which is pictured above. Learn Sesame "enables a Web site, network, or application to deliver content, user interface, advertising, and product recommendations that are customized and targeted to each individual user. The software builds a dynamic interest profile of each user that integrates with other best-of-breed dynamic page servers, indexing engines, and ad servers."

Such websites are common and widespead; there are too many to list here. But you have probably accessed some of the more famous: My Yahoo, My Netscape, CRAYON, and others.

Customized websites are nice, and for many users, convenient. However, there is a weakness in this model: if each site collects its own data, users are constantly logging on, creating personal profiles, and entering data. Who among us does not already have separate user IDs for Yahoo, Netscape, and a variety of other web sites?

The solution to this problem is to delegate the task of creating user agents to a third party, one which will mediate between the user and the website. As the Foundation for Intelligent Physical Agents wrote last July, "functionalities will often be private to user agents that are to show personalized behavior. However, it may often be beneficial to delegate these personalization tasks to a third party. This has several positive consequences:

  • User agents can be relieved from performing possibly complex user model learning, representation, maintenance and reasoning tasks.
  • User models can be based on the interactions between users and more than one agent, which may lead to richer user models.
  • User models can be available to more than one agent, which avoids repeated acquisition of the same information and hence may decrease obtrusiveness of the human-agent interface.

What this will generally mean is that some program, acting on behalf of the user, will interact with the website. Thus we see a natural convergence between the intelligent search agents, described in the first part of this article, and websites which learn, described just above.

What we want is something like a smart cookie, something which will exchange personal information with various websites, but which will do so on our terms. Something, perhaps, like Cupcakes, recently introduced by I33 Technology Corporation. In their words,

"Cupcakes offer the ability to customize content, sales offers and advertising messages for each user based on his/her Cupcake, an editable profile that resides on his/her computer. Unlike any other technology, Cupcakes allow users to decide what the world knows about them. The client-side Cupcake puts information in the hands of users, allowing them to store that information, edit it, and decide who sees it and when. Unlike cookies, the user controls the data."

Cupcakes look a lot like cookie technology, and I33 Technology is open about the similarities. But there are differences, which they highlight in a table:

Cookies Other Registration
Data about the user is stored: in the user's cookie file on the server in the user's cupcake file
User decides what information gets stored: No No Yes!
User can edit stored information: No No Yes!
Information can be shared by multiple sites: No No Yes!
Server can "secretly" store data about the user: Yes Yes No!

Cupcakes are in beta testing now; for a nibble, visit their site at

Another approach is Firefly's concept of the Passport. A Passport is "an individual's trusted identity and personal profile. It's like a credit card which stores information about who people are, what people like, and what they don't. By using a Firefly Passport at different sites which use Firefly Products, people can have access to personalized content, community, and services while maintaining their privacy."

Significantly, the Firefly passport can be used to personalize your interactions with more than one site. Hence, for example, My Yahoo will recognize your Firefly passport and, with your permission, use your Firefly information to create your Yahoo user profile.

This development has allowed Firefly to develop a network of companies using Passports, including Barnes and Noble, ZD Net, and College Beat.

So get ready. Your browser of the future will employ cupcakes or passports, intelligent search agents, and a variety of other tools in order to act as your personal agent on the internet.

Stephen Downes Stephen Downes, Casselman, Canada

Copyright 2024
Last Updated: Jul 19, 2024 4:27 p.m.

Canadian Flag Creative Commons License.