FEATURES:Intelligent Agents

Is this software after your job?

In Kate Bornstein and Caitlin Sullivan's cyber novel Nearly Roadkill, the troubled police chief has a virtual assistant who talks to him via his computer. Based visibly on a vampish, film noir femme fatale, she gives him sideways looks and issues wisecracks before sashaying off his screen to go and search for the novel's outlaw heroines in the seedy back-alleys of the Internet.

Of course that's science fiction, but only just. Looking at the intelligent agent search engine Autonomy you can see where the idea came from. Autonomy's agents use the type of animated 'toon (as in cartoon) interface with the user that is becoming increasingly popular. In Autonomy's case it's a dog which digs away while the agent searches the Net on your behalf, makes an appealing face when begging to be 'let back in' (i.e signalling that it's found something) and - just like your dog at home - it gives you a thumbs up sign when it's particularly pleased with what it's retrieved for you.

Like a real dog, the agent has to be trained. By giving it instructions as to what you want, and 'rewarding' it when it does well, you gradually refine its searches so, as it gets to know you better, it comes back with more targeted information and fewer time-wasters.

At the moment, the technology is still at an early stage of development, so perhaps canine intelligence is an appropriate level at which to peg the metaphor - dealing with it will need a lot of patience, but may bring its own rewards (no doubt the mistakes it makes in the early days will be as unerring a source of humour as that old failsafe the spell-checker). But as anyone who has watched the development of technology in the last ten years knows, things will change, and they will change very, very quickly.

In the future, perhaps, the cartoon interface will look like one of those media librarians (horned-rimmed glasses, tightly scraped hair bun etc - fill in your own stereotype and/or fetishistic accoutrement here). More importantly it will almost certainly do a lot of what librarians currently do.

Laurent Lachal is a consultant at Ovum Ltd, which has produced a number of reports on intelligent agents. Librarians, he says, have two choices: they can either ignore this technology and hope it goes away, or they can learn to work with it. If they want to stay in a job, he argues, they would be well advised to do the latter.

'There is always going to be a need for people who are smarter than the agents,' he says. At the moment that's like being a millionaire in lira, not that difficult, but those inclined to be smug should remember the recent fate of chess master Gary Kasparov. 'They aren't currently that smart, but they will get smarter, and librarians need to keep one step ahead.'

The advent of the World Wide Web and its conventional search engines have already altered the role of the librarian, particularly in a business setting, Mr Lachal argues.

'We have three librarians here, but since the advent of online databases I am using them far less. Previously if someone wanted to know, for example, the definition of "enterprise resource planning", they would put a call in to the librarian who would know the right source to go to. Now you can go to something like Alta Vista and get a fairly good idea of what the term means in about a minute and a half.'

Intelligent search agents will, he believes, accelerate this change in the librarian's role, enabling more complex tasks, like gathering together a portfolio of information on a particular topic, to be done electronically. But this needn't be seen simply as a threat.

'It is also an opportunity,' he argues. 'Searching for things is actually rather boring, and if machines can be developed to do this form of drudge work it allows librarians to get on with more interesting tasks.

'There is still going to be an information overload. The future librarian will be someone who can span a number of subject areas, without being a subject specialist in each of them, and provide meaningful summaries of the information available.'

So what are intelligent agents? The most popular answer to that question is to say that as with intelligence itself, there are no clear agreements on an actual definition, but there are calibrations of some of their functions. In other words we can't say exactly what they are, because they take a number of forms, but we can say some of the things that they do.

They may stay static on your own machine, like a program which sorts e-mails as they come in, but the ones which have received the most attention are those which, like Autonomy, are mobile. They react to their environment and, most importantly, they learn from experience.

What takes them beyond simple keyword searching is the faculty of 'fuzzy logic'. This enables them to recognise concepts which are related to the training text, rather than just zeroing in on specific words.

Where it begins to sound like a science fiction plot is where the agents start to interact socially, as the more sophisticated ones already do, with other agents and with human beings. This function is particularly used at present for agents which organise complex timetables for meetings and events. At the simple end, such applications might do what a network of PAs would do - schedule several people to be in the same place at the same time, and inform them all. At the more complex end is the visitor hosting system, already operating in an American university, which schedules and organises events, taking into account such variables as the visitor's interests and status in his or her organisation.

Autonomy runs what it jokingly refers to as a dating service where the dogs carry around profiles of their owners. If during their travels they run into other dogs whose owners have similar interests, they exchange details.

This social networking facility of agent software is an extension of services such as Firefly which in themselves replicate some of the functions that literature promotion projects have fulfilled in public libraries.

Firefly itself is a selection system for music and movies. On your first visit, you are asked to rate a series of titles. From your responses the software forms an impression of the type of thing you are likely to enjoy and comes up with a list. Theoretically you then go away and watch all the films, buy all the CDs and rate them in turn. This gives the system information with which to refine your profile, and so on. It then matches you up with other Firefly members of similar interests, gives you their e-mail addresses and tells you whether any of them are online at the time. If they are, you can go to chat rooms and 'talk' to them.

This is directly analogous to some of the schemes run in ventures such as Sheffield's Opening the Book and, on a more philosophical level, it performs the same sort of role of uniting people by common interests that Comedia's Libraries in a World of Cultural Change report outlines for public libraries. Of course, the fact that it is geographically diverse is both a weakness and a strength. The fact that someone who shares your interests lives in San Francisco and not down the road in Sheffield may be a disadvantage if you're looking for a friend, on the other hand it might be an advantage if he or she turns out to be a pest.

Of course this is not in itself a threat to the librarian, but it is a sign that the Internet is genuinely becoming, in a philosophical sense, closer and closer to being a library without walls.

A more serious challenge to the librarian's role comes from the various customised news services which are emerging. At the lower end these are still apparently primitive: NewsPage Direct, for example, puts together a Web page for you according to various interests which you select from a checklist. It then delivers this to you via your e-mail address. Unfortunately if you don't have an html-enabled newsreader, and many corporate users using products like Microsoft Exchange do not, you'll get the information you asked for, but it will be interspersed with raw html code to an extent that will make it unreadable.

There are more sophisticated packages available, like PointCast for example, which aim to give you an 'active' desktop environment. It offers you a number of news 'channels' to chose from and interfaces with your own Web browser or an inbuilt version. Corporate versions update automatically, while versions configured for home usage update only when you go online. It also gives you news-based screen savers, so your machine broadcasts a series of headlines when it's not in active use - presumably to give you something to read while engaged in a boring but necessary telephone call.

The channels supposedly feed you the news you want to know, although the problems for a British-based user are immediately apparent as you succumb to the temptation to hit the temptingly-positioned sports channel only to find yourself deluged with information about the Ducks, the Bulls and various other animals you have no interest in whatsoever. Even references to the Reds turn out to mean not Manchester Utd or Liverpool, as you might have expected, but the Cincinnati Reds, whoever they might be. In other words, you only get out of these things what someone else is putting in and in PointCast's case the content is overwhelmingly American. The other thing is that the filters are currently fairly blunt instruments. One American librarian confesses in a Stateside discussion list to getting rid of PointCast 'ostensibly' because of system conflicts 'but really because it was causing me unwarranted and unnecessary information anxiety' - and that's from an American.

But all the applications mentioned so far are pioneering software and almost certainly the period pieces if not of tomorrow then certainly of next year. The quantum leap in the use of agent technology will happen shortly with the release of new Web browsers from Microsoft and from Netscape.

Using intelligent software, the new breed of browser will visit your favourite sights and deliver the information to you for you to peruse offline without ticking up your phone bill. This is what is referred to as the 'push revolution'.

Cynics point out that tailoring the information may also at some stage allow information providers to sell the sort of tailored advertising that the commercial world has been yearning for on the Web. In other words, one of these days 'push' will come to 'shove'.

But Microsoft's plans particularly have gained some credence within the business information world (Microsoft's perspective on this can be found at the link below). Effectively they have the potential to turn the free-for-all of the Web into a form of focused broadcasting, or 'narrowcasting' to use the advertising speak. In other words, you get the message across to the people who want to hear it (and who you want to send it to). The commercial potential of this is clear to see.

The library and information world has its own intelligent agent project, funded by eLib, and going under the title of NewsAgent. This trial project, which goes live this month, will provide subscribers with information initially fed from a selection of LIS journals, including the Record and Library Technology.

Once again the difficulty with this type of system is going to come in the levels of information that are fed into the system, says Robin Yeates of the Library Information Technology Centre at South Bank University, one of the partners in NewsAgent. 'I suspect that the users will feel that there is not sufficient volume of information, while the information providers will be concerned about the amount of work that it takes to make information available to the system.'

NewsAgent operates on what is called the Dublin Core metadata, an agreed protocol which is added to online material. There are currently two difficulties with this. One is that adding the metadata to html pages is still fairly labour-intensive - each item requires about three quarters of an A4 page of information to be added to it (principally source information, author name, appropriate channel and keywords). The second is that any off-line sources, like the vast majority of the Record, must be converted to html before the metadata can be added in the first place.

This all takes time and, as we know, time is money, which brings us to the next area of uncertainty about this new age of information and intelligent agents, which is how the economics of information is going to function. One look at PointCast is enough to convince you of its strength as an advertising medium and advertisers have been trying to figure out a way of using the Internet effectively for some time. The constant updating means that a range of adverts can be pushed to people's desktops with the information being used as bait.

But ultimately information providers have to pay the salaries of the people who are working for them. This means that, for example, Wired magazine (a channel on PointCast) is not going to provide extensive content for nothing, not at least for ever. Either they are going to see the Internet as a free medium and one on which to broadcast the concept of Wired magazine, in other words the information available is going to be in 'taster' form, designed to get you to buy the magazine in print form, or it is going to cost money to access. Whether as a mixed economy (user + advertisers) like print media or advertiser-driven (like commercial television) somebody is going to have to pay. Obviously the shape of the information will be affected by the shape of the economy.

When it comes to the future, the one thing we can be certain about is that we don't know quite what it's going to be like. But we can also be sure that it won't be a place for those who are stuck in the past.

Links:

NewsAgent list of links and resources:

AgentWeb (general information)

Software agents: an overview:

Autonomy

Firefly

Feedback or comments on the contents of the printed
Record or on the Record Web pages are welcomed

© The Library Association
Last updated: 16 June 97