The idea that software could think for itself is almost incredible, but intelligent agents are here and causing a bit of a stir among programmers and business users. Their forerunners were the expert systems of the 70s, which did not take off because the applications were never quite right and memory and processing power constraints meant they could never be run properly on the desktop.
Today all that has changed. The power is certainly available, and now that the volumes of data most businesses have to deal with are spiralling into the terabyte league, users need software which will intelligently browse, seek and analyse functions.
Colin Murray, a product manager with the SAS Institute, explains: 'The amount of data which most companies have to deal with is so large that there is no way an individual can manage it. Even with basic data mining tools, the data becomes useless because it can't be analysed or put to any good purpose.'
Dominique Verdejo, Internet project manager with Ilog, has been following the rise of intelligent agents for six years. He says that they are increasingly being integrated into various industrial applications, most notably in the telecoms sector and for developing enhanced services on the Internet.
'There are many definitions of intelligent agents, but broadly speaking they are autonomously acting software programs which deliver real services in a virtual world.'
Murray says intelligent agents have three main uses: filtering, to remove junk mail from a daily list of email messages; making information more understandable by putting it into some sort of context; and looking for things the user didn't know existed.
It is the last of these which offers most excitement. The most frequently quoted example is that of a supermarket chain which, through judicious use of an intelligent agent, discovered that the sales of disposable nappies and beer peaked at about 10pm each night, as husbands were sent out to buy more nappies and bought themselves a few cans at the same time. As a result, the supermarket sited the nappies and beer closer together, and increased its beer sales even further.
Verdejo sees intelligent agents being used primarily in the telecoms market, including the Internet, and in electronic marketing and commerce.
'In the telecoms market intelligent agents with supervision, filtering and correlation capabilities are being used to automate processes in network management,' he says.
'US Spring, a major telecoms operator in the US market, has designed a Wan supervision applications called Clearline which uses an intelligent agent to correlate alarms and detect the origins of communications failures.'
But the use of intelligent agents in the sphere of electronic commerce has raised some fears about security. They are able to move from one machine to another, bringing with them encapsulated data and rules which allow them to negotiate on behalf of their owner in electronic transactions.
Verdejo says: 'There are interesting implications for electronic commerce here, and the security issue raises some important questions.
How secure are intelligent agents? What role do they play in other areas of business? What is their future role?'
Gary Smith, managing director of Red Brick sees intelligent agents as a natural development of query and reporting software, which in a short time will become a standard feature. 'Previously a straightforward query would do for most people to mine down into and information store to get the figures they need,' he says.
'But it's not just the fact that the volumes are now so big that the speeds are starting to slow, but often people do not know the questions the need to ask. Without the right question you don't get the right answer.'
Smith points out that until intelligent agents, the biggest benefits from data mining often happened by accident. 'Rather than having armies of analysts surfing the data warehouse, it is so much more sensible, cheaper and more efficient to get software to do it.'
Smith says that it has taken the programmers some time to reach the point where intelligent agent software has reached the stage where it can be bolted on to almost any application. But now resellers and integrators should be investigating what is available to enhance the solutions they have already installed.
'Previously users and corporates had islands of information which was searched and reported on, but they were essentially disjointed and now able to speak to each other or make links or connections. But with intelligent agents all kinds of links and associations can be made,' he says.
Smith explains that by giving the intelligent agent software some constraints and parameters it will come back with all kinds of associations that would not otherwise be obvious. 'The way I understand it, it will display certain links in the form of dots on a chart, and then the user looks for clusters of dots, which will indicate a trend.'
There are certain similarities between the way intelligent agents work and fuzzy logic, but, says Smith, intelligent agent software is more like object technology. 'It will certainly become a standard extension or addition to any situation where historical data is stored, although it is in its early days at the moment.'
Mike Lynch, MD of Autonomy, a Cambridge-based software firm, has given the whole category of intelligent agent software a new name: agentware.
That is the name he has also given the product Autonomy has developed, which roams the Internet like an intelligent browser, searching out information for the user.
Lynch says: 'A big problem for Internet users is getting at all the information.
The limitation of ordinary searching or browsing software is that it needs a user to be there. Agentware works on its own.' agentware can be used for a variety of tasks including blocking access to certain types of Internet information such as pornography, and amassing data from disparate Web sites.
Not only will an intelligent agent search for specific information, but it will keep an eye out for other things it thinks the user might want to see, based on its knowledge of the user's interests. Although Lynch does not suggest that agentware will be the end of browsing and mining software, he says conventional browsing often takes users down many dead ends in their search for information.
Murray says: 'Intelligent agents are in many ways an evolution of expert systems because they can be used to capture the experience of very expensive and highly skilled people and make it widely available.' The skill of analysts, consultants and programmers, for example, to 'read' data and interpret it for marketing purposes, can be transferred to an intelligent agent which could be bought almost off-the-shelf to be bolted on to a data warehouse app.
'Intelligent agents are the only way to handle the volumes of data that we are seeing now, and because the processing power is available and they can be made simple enough for a fairly experienced programmer to bolt on to existing code and a data source, they can be embedded into many different applications,' says Murray.
He adds that intelligent agents are things which make life easier for programmers and users as they refine existing tasks. One day they will also be able to initiate tasks themselves, he says. 'They are currently seen as something novel which can be bolted on to an application or data warehouse as an afterthought, but it is only a matter of a short time before they are automatically embedded into the applications as they are developed.'
Murray says that several organisations and products already use intelligent agent software, so all resellers and integrators should be learning about them, what they can do and how they can be implemented. 'The White House in Washington has installed an intelligent agent-based text analysis system that reads the content of email and routes it to the appropriate department for reply. This includes routing the threatening and abusive ones to the FBI.'
WHAT IS AN INTELLIGENT AGENT?
Intelligent agents, software agents or intelligent information agents are the terms used to describe an automated way of approaching and managing the large amounts of data that need to be processed, data mined and analysed these days. They draw from the worlds of object orientation, artificial intelligence and knowledge-based systems.
Intelligent agents have been called 'objects that think'. They can interact with the user, with systems resources and with other agents to achieve their goal. Agents go one step further than objects and have data and methods which will act on a mass of data. They are also attributed with having 'beliefs, commitment and goals' - a degree of reasoning applied to the data store or source which guides the gathering of extra information in order to achieve goals which have been set.
Like objects, agents are able to communicate via messages. Agents may be complex or simple and either work alone or in harmony, creating a 'co-operative society'. Other attributes include the ability to migrate themselves across nodes of a network to perform their tasks and then report back on their findings.
INTELLIGENT AGENTS OF THE FUTURE
Verdejo says the second generation of intelligent agents are called 'wondering agents', but their capabilities can only be guessed at.
Smith says the greatest demand for intelligent agent software is coming from second-generation data warehousing users that have decided that they need to do more to get most value from the information they are gathering.
'They can see that by continuing to ask the same questions and generating the same reports they are starting to miss things. They need to introduce a wild card, if you like, which will suggest potential links and deductions which may not have been considered,' he says.
The next generation of intelligent agents will have far more intuition, according to Murray. 'At the moment intelligent agent software can produce a series of events or data groups which appear to form a trend or show some kind of link, but it will be unable to tell you why. The next generation of intelligent agents will be able to show you why there is a link.'
Common standards also need to emerge to interoperability between agent systems. At the moment there is no standard for intelligent agents from different environments to work together.
Agents will eventually be used to control business processes, says Murray, and develop new processes as necessary. 'As soon as you start talking about intelligence in software user's expectations and imaginations will start to soar. Then it's up to the programmers to deliver, and I think that they will.'
Ovum recently published a survey of the embryonic intelligent agent market, with a forecast of the likely development of this market.
The market was split into a number of segments of which the potential implementation of intelligent agents and the interface, development tools and information retrieval are the most interesting. The report concluded that all the market segments would total $1,700 million by 2000. Of that, $450 million will be in interfaces, $650 million would be in development tools and $680 million would be for information retrieval applications.
The balance would come from industrial and other applications.
View pictures of all of last night's fights
Acquisitive comms provider swoops on Frontier Voice & Data and StoneHouse Logic
Cybersecurity firm rakes in £3.6m for unwanted unit
Results, reaction and pictures from last night's CRN Fight Night