Let an agent negotiate it

Early in the coming millennium the set-top box (now moved inside your television set) will be Java-enabled, and electronic commerce in the home will then truly take off. The whole of the web will be accessible from the couch. TV commercials will link you into web pages, and retail therapy will be immediate. Television will have become truly interactive, and the distinction between broadcasting and the web will break down.

If television is an experience for the slack-jawed couch-potato, the web affords an alert, dynamic, eclectic activity. The canny television shopper will be active, webwise, and will seek the best bargain. Given the vastness of the web she will need software support. She will need her own agents to roam the web in search of the very best deal. Intelligent agents will soon be indispensable in managing information overload and the info-bloat of the internet.

An agent is a system, almost always a piece of software, to which the user delegates some authority, much as she might delegate the details of her travel plans to a travel agent or of her house purchase to an estate agent. An agent acts on her behalf but does so with some autonomy. Intelligent agents act flexibly and can learn from their experience.

Experts tell us that the big growth area is in business-to-business e-commerce. In this, intelligent agents will act on behalf of corporations, disintermediating those activities that merely shift information about, activities that IT has rendered non-value-adding.

Intelligent agents are the result of the collision of network computing, artificial intelligence and object orientation. Intelligent agents are intelligent objects. Agents may be static, sitting on one machine, but they typically operate over a network. Or they may be mobile, travelling around a network and running on its distant nodes. The web is the perfect stamping ground for mobile agents.

Intelligent agents can negotiate on behalf of those they serve. They can have human-like qualities, the sort of things we like to characterise as beliefs, intentions and desires. A philosophy and an economics of agent interaction flourish in the research journals. To devotees, intelligent agents are a form of proto-artificial life, implemented in code. If the internet is to be a living thing, as some claim it can be, it will come alive with these little helpers.

Standards are important to agents if they are to communicate, negotiate, interact, roam the net and run on a variety of processors. The world of intelligent agents has quickly matured to the point at which it has its own standards organisations. There is FIPA, the Foundation for Intelligent Physical Agents, which despite the "physical" in its name is focused on software agents. There is MASIF, the Mobile Agent System Interoperability Facility, a partly competing organisation. IBM in Japan has developed a Java-based system for mobile agents, or agent applets, inelegantly called Aglets. A number of other applications are now appearing.

Nicholas Jennings and Michael Wooldridge at Queen Mary and Westfield College are two of the UK's leading academic researchers in agent technology. Their book is an edited collection of papers that introduces the idea of agents, some visions of the future, some theories about current agents and some commercial instances of the technology. It was written in 1997 and published in 1998. In internet-time, this is an old book. Some of the contributors have already moved to new jobs. Nevertheless, for the academic or the technical manager who wants an overview of the technology, this represents a good place to start.

Peter Gibbins is executive director, Virtual Centre of Excellence in Digital Broadcasting and Multimedia Technology, and visiting professor in computer science, University of Exeter.