What are intelligent agents?
They are the bastard offspring of research into artificial intelligence and artificial life, with more than a passing acquaintance with computer viruses. Essentially, intelligent agents are self-contained chunks of mobile code which can act autonomously, respond to changes in their environment and communicate with users or other intelligent agents. Ideally, they also have the ability to learn, or at least be trained.
Intelligent agents are best imagined as software robots - indeed, they are sometimes called 'bots' - that can move around individual systems or complex networks performing specific tasks. For instance, a user could instruct an agent to go onto the Internet and negotiate the best price for a particular product. Agents that interact with users, rather than just with other agents, are also typically given a 'personality'. Cute talking animals seem to be very popular.
Christine Karman, founder-president of Amsterdam-based agent supplier Tryllian, says, "A change of mindset is the most important thing: thinking of your software not as something you use but as something with a life of its own."
Nick Jennings, professor of computer science at the University of Southampton and an expert in intelligent agent technology, says, "For me, agents are a natural evolution of objects. You can think of an agent almost as an object++. I believe we'll see agents in mainstream computing in the way we see objects in mainstream computing today."
What areas of business can benefit from intelligent agents?
Although agents are likely to find their way into many different areas of business, the key drivers of the technology are e-commerce and networked open systems. Intelligent agents could allow organisations to offer a tailored service to many thousands of individual customers, 24 hours a day, seven days a week, and to hike the speed of transactions up to previously unimaginable levels.
Jeff Burgett, chief technology officer of AMS's European e-business division, says, "We've filed for a patent on a set of agents that would monitor prospective customers coming in on a Web site. Based on both internal and external information on a customer, an agent would then establish a profile, or risk level, for that customer and communicate with a negotiating agent that would establish a tailored negotiation strategy or pricing plan."
Organisations that overlay their existing systems with an agent model can also overcome the problem of different systems not being able to talk to one another. If one agent is taught to query a particular system, any other agent can simply instruct it to retrieve the information it needs.
"Those sorts of applications - which are increasingly common as companies look to open up their computing infrastructure to take advantage of the Web - you can only build with agents," says Jennings.
Can intelligent agents replace human beings?
In a word, yes. In an e-business context, it would be totally impractical for an organisation to employ the armies of people needed to offer instantaneous, meaningful responses to every customer query. Soon, agents will allow businesses to expand to unprecedented levels without simultaneously employing hordes of additional staff.
We may not yet be at the stage where we can all get on with something more strategic while agents run the nuts and bolts of our businesses, but already we can get a tantalising glimpse of what the future may have to offer.
Tryllian's Karman cites an agent her company is currently developing for a major Dutch telecoms supplier: "Imagine there's a telephone switch connected to a computer network. Currently, if something goes wrong with the switch, the telco needs to send out a support engineer, which costs a lot of money. In future, the telco could send a software agent over the network to diagnose what's wrong with the switch and repair it. If it can't work out what's wrong, the agent will simply return to the phone company and obtain the upgrade it needs from a support agent or expert system."
Should you become an early adopter?
The experts think so. "The sooner you get in, the more of an advantage you'll have," says Karman. Jennings adds, "I think a lot of companies are pleased they have been early adopters and are starting to see the rewards. It's true in telecoms and e-commerce and I think it's going to be true in the more traditional customer-focused areas such as the financial sector, because you get the ability to personalise customer relationships and to build much more sophisticated open systems."
Burgett says, "Now is the time to become an early adopter of agents to automate back-end processes and respond more effectively to customers. The reason we got into agents four years ago was to solve complex problems. We found an agent approach was much more effective than the traditional procedural strategy. The technology allows you to break a problem up into manageable pieces, with agents collaborating to solve more complex problems."
But the window for getting in early may not be open for long. Tom Ilube, CEO of agent developer Lost Wax, predicts, "In 18 months to two years, it will seem odd to be launching an e-commerce presence - be it business or consumer - without some sort of dynamic or intelligent agent capability, because customers will expect to be able to negotiate in some way with your business online."
What businesses are researching, using and making money from intelligent agents?
In the IT and telecoms sectors, research into and deployment of intelligent agents are already fairly well advanced. "You could name 50 top companies and I could tell you they're all doing agent research - Sun, Microsoft, Hewlett-Packard, BT, IBM - all those sorts of organisations," says Jennings.
At BT, research into intelligent agents takes around 5% of the company's £600m annual R&D budget. The company is now at the stage of commercial implementation of some of the fruits of this research (for example, using agents to personalise information for customers). Outside IT and telecoms, take-up has been more limited. However, Bent Thomsen of ICL's research and advanced technology department and a veteran of intelligent agent research, says, "We've been reasonably successful at applying agent technology in the financial services sector. We're using it for share trading in the banking sector and at the user interface level for things like mortgage advice - users can communicate in natural language with a computer through an animated character and the intelligent agent goes off and queries databases to find the best mortgage."
Burgett says, "If you loosen up your definition of what you consider to be an intelligent agent, then they're in broad use today by most companies: intelligent search engines, intelligent knowledge management systems, products such as Autonomy and Excalibur."
What about standards?
If intelligent agents are indeed to become the dominant model for software development, agents will need to be able to communicate with one another. Not surprisingly, there are a number of competing standards.
Thomsen says, "The Federation of Intelligent Physical Agents [Fipa] has gone a long way towards standardising things such as KQML [Knowledge Query and Manipulation Language] although it doesn't look like industry's picking up on those standards. Then there's the Object Management Group [OMG] , which has a subcommittee on agent standardisation. It's enriched the Corba standard for agent interaction, which has had some pick-up, particularly in the telcoms area, but again it's not really widespread.
"Personally, I think neither standard will win out because Microsoft has just launched its .Net infrastructure. Bill Gates has spoken a lot about agents talking to one another using XML as the language and .Net as the environment for executing the agents. Our prediction is that this is going to be where the big take-up in terms of multi-agent systems and communication standards will be."
Jennings, on the other hand, thinks all talk of standards is premature: "For those who think standardisation is a good idea, there are still lots of ongoing debates about what these standards should look like and I think that's partly because they're trying to standardise a moving target."
What potential problems exist with intelligent agents?
According to many experts, one of the biggest problems with intelligent agents will be getting people to trust them. People will be suspicious of agents because they make decisions on their behalf rather than following set instructions. This could drastically affect take-up.
Lost Wax's Ilube believes the solution lies in getting people to take ownership of agents that act on their behalf. "What would be really interesting is if a body like the Consumers Association or some other trusted third party offered people agents that acted genuinely in their best interests," he says. "You can imagine everyone really trusting those agents."
Of course, having lots of third-party agents trawling your corporate systems makes clear the other big problem with the agent model: security. Jennings thinks it's a problem we're going to have to cope with, like it or not: "Open systems where you have code from multiple suppliers and buyers are increasingly going to be a reality," he says. "If you're not going to play in that space, you're going to struggle to expand. Yes, it's more chaotic. Yes, it increases security risks. But it's inevitable. You have to deal with it."
I'm still interested. How can I take things further?
First, make sure you have a clear idea of what agents can do for your business. "Ask whether the technology opens up new business models," advises Illube. "Will it result in cost reductions? Are you expecting it to lengthen the life of customer relationships?" Jennings recommends anyone interested in tracking developments to look at www.agentlink.org, a European network of excellence in agent-based computing. "Membership is free and it gets you access to the sorts of agent resources that are out there," he says.
When it comes to choosing suppliers in a cutting-edge area such as this, it's not a simple case of looking at a top 10 list. Use the Web to talk to other early adopters and ask for recommendations. Find suppliers in tune with your way of thinking and that have some experience of your industry. You may prefer to deal with a large systems integrator such as ICL, a big-name supplier like Microsoft, or an innovative smaller player such as Lost Wax or Tryllian, but your choice will depend largely on which company's approach best fits in with your own organisational requirements.
In terms of in-house skills, retraining staff in agent technology and techniques should not be all that difficult. "Although it's a new way of doing things, it's not completely radical," Jennings points out. Within five years, he believes, students will be coming out of university with a good grounding in agent technology.
BT's secret army
A key mover in the field of intelligent agents, BT has been researching the technology for five years and is one of the founding members of agent standards body FIPA. Nader Azarmi, AI technology manager and technical manager of BT's intelligent systems research group, says the company is working on agents in two main areas - personal agents which find and customise information for users, and multi-agent systems for complex problem-solving.
"We spent quite a bit of time initially working on a toolkit called Zeus to enable us to develop agent applications," says Azarmi. "This won us a BCS innovation award 1997."
The company has used Zeus to develop a number of trial multi-agent systems. "The three main areas where we've been trying to apply collaborative agent technology are in network management, e-commerce and business process re-engineering," he says. "We've developed a number of prototype systems and we're hoping those could made available first internally and later commercially."
BT has also been developing a personal agent framework. "The idea here is to develop a central profile for the user - in which you identify your interests, expertise, what you're looking for, etc," explains Azarmi. "Then a number of agents can use that profile to go and find information for you on the Internet, an intranet or extranet."
This framework has already borne fruit. "One agent, Bugle, will generate a daily newspaper based on your interests," says Azarmi. "Another, called Ivine, facilitiates networking. Essentially, it tries to put you in touch with a community of people with similar interests."
Then there's Radar - named after Radar O'Reilly from the television series M*A*S*H - a just-in-time information delivery agent. "Essentially, Radar can watch over your shoulder as you're typing in MS-Word, for example, and find documents relevant to the paragraph you're typing in real-time."
For more information on BT's agent research and deployment, see www.labs.bt.com/projects/agents.htm
This was first published in December 2000