SO YOU are in Manhattan and you want to ring a friend in Tokyo, well, join the queue. AT&T, the US鈥檚 main carrier of long-distance telecommunications, now handles 160 million calls a day, up more than fifty per cent in five years. In Britain, BT has 7000 telephone exchanges, and in the past 15 years it has installed more than 3 million kilometres of high-capacity fibre-optic cable. In many European countries, as well as developing nations in Asia and South America, there are also blossoming mobile phone networks, interactive cable TV networks, and, of course, data networks.
But by building and operating such an array of complex and interlinked systems, the telecommunications industry has set itself a big problem: how do you manage these networks? Today鈥檚 telecommunications users 鈥� especially if they are large businesses 鈥� do not expect to hear that lines are busy, that a service is not available, or to find that a high-speed data transmission fails because the network cannot support the traffic load.
As the latest telecommunications and computing technologies are introduced, these networks promise to revolutionise society in the next century, making it possible to send and receive massive amounts of fast-changing data to the tiniest portable devices using the vaguest addresses. But showing that technology in the labs is one thing, delivering it is another. Many telecommunications experts are worried that it is all getting out of hand, that the networks could prove overwhelming or even unstable, and that too many important decisions could end up being made by computers, or even by software programs that wander across the world鈥檚 networks.
Advertisement
Ideally, subscribers to telecommunications systems today, and in the future, would like the network operators to manage their communications for them, directing the messages to the right place, in the right format, at the right time. In fact, they want the network to show a bit of intelligence. Most of today鈥檚 telecommunications and computer networks are based on switching devices which simply read the destination of the message. Such devices 鈥� whether they are routers forwarding messages over the Internet or exchanges switching telephone calls 鈥� carry out very little computing, they just speed the messages on their way down the dark pipes or send them across the airwaves.
If there is any processing carried out once the message or call has left its source, it usually occurs at the level of switches. These can be programmed to send back messages for analysis by network management software running on a computer at one of the operator鈥檚 main sites. These centres act like traffic control centres in large cities, which collect and distribute reports of traffic problems. Any network problems, such as a failed exchange or unexpected congestion, are reported back so that human network management specialists, will be able to reroute traffic or take other appropriate actions, such as organising repairs.
These network management centres, with the occasional well-publicised exception, have worked well in the past, but the network operators are starting to question how well they will work in the future. Deregulation in countries across Europe means that the range of companies that can offer some form of telecommunications service is growing rapidly. The result is that large parts of any national network could be under the control of different suppliers.
The fear is that these centres will become swamped by management messages 鈥� such as local traffic congestion caused by a TV phone-in 鈥� and will be overburdened as the demand for more sophisticated services such as video conferencing and online shopping creates unprecedented traffic levels. 鈥淭he old centralistic approach will not work into the next century,鈥� says Robin Smith of the Advanced Information Processing Centre at British Telecom鈥檚 Martlesham research laboratories.
BT, for example, estimates that it will have to write some 27 million lines of code to cope with the predicted growth of its telecommunications and data networks. And even after they have finished, there are fears that the system would soon be obsolete as demand continued to grow.
One BT researcher describes the software being developed as 鈥渢oo vast and too expensive鈥�. And he fears that there will come a point where the central system will get so big that it will become unstable and 鈥渆xplode鈥�. Smith says that the centralised model is fine while the network remains the same. But this kind of management begins to fall apart once the network undergoes frequent change 鈥� as is the case at the moment.
That is why telecommunications operators such as BT and AT&T are exploring the possibility of spreading the management programs across the network rather than localising them in a few centres. These are being called 鈥渋ntelligent networks鈥�. The key to their development is the emergence of software agents, small software programs which are able to react to their environment, learn from human behaviour online, and move on to take decisions independently of the humans which programmed them (鈥淪oftware鈥檚 special agents鈥�, New 杏吧原创, 9 April 1994). These agents not only make decisions on behalf of subscribers, such as which messages should be listened to and which should be thrown away, but they can also decide the best routes for sending various types of information. Smith suggests that a better way to manage networks would be to link each telephone exchange to a local PC or a workstation, running 鈥渟tatic鈥� intelligent agent software. This software would make the kind of decisions now taken by the hundreds of individuals that specialise in managing BT鈥檚 network.
Intelligence services
For BT, the 鈥渁gents鈥� would be programmed with goals or 鈥渞ules鈥�, such as 鈥渒eep traffic levels below a certain threshold鈥�, 鈥渞oute calls using the cheapest method available鈥�. But beyond those rules, there will be some autonomy. So if, for example, a digger cuts through a fibre-optic cable in the City of London, the agent has to determine alternative routes, make decisions about which services should be given priority, and negotiate for capacity with equivalent software programs running on similar computers at nearby exchanges.
There are interesting problems ahead, however. All that deregulation may mean that some of these nearby agents may be owned by another company. So, will you be able to trust the instructions of a piece of software acting on behalf of one of your competitors?
Pattie Maes, an assistant professor at the Media Lab in Cambridge, Massachusetts, has looked at the problems of using intelligent agents during her research into agent technology. This has led Maes and her group to give their agents four learning methods.
In one of the techniques, the agent searches for statistical patterns in the way a user interacts with a computer, for example if messages from a certain person are always discarded, the agent will follow this pattern. But before automating the process, the agent will first 鈥渁sk鈥� the user if they want this pattern of rejection automated. This gives the user a second chance to take back control of the process. If a user rejects a pattern, the agent will learn areas where it should not automate. The user can also give the agent hypothetical examples of what it should do (such as 鈥渇orward all messages from my boss to my home鈥�). Or the agent can send an e-mail message to another 鈥減eer agent鈥� across a network, and ask what, in their experience, they would do.
Some of these ideas are likely to be reflected in the IBM Communications System, an intelligent new network which IBM plans to launch this autumn. The goal of ICS, says Don Gilbert, coordinator of IBM鈥檚 intelligent agents group, is to build sufficient intelligence into the network so that the end user can always be found and the message delivered in the right form. 鈥淲herever you are, it will find you. And when it finds you, it will deliver the message in the form you want it.鈥�
IBM鈥檚 plans are in the early stages, but they are highly ambitious. Gilbert cites an example where a network customer is driving along a motorway when an urgent fax arrives. The network finds out where the subscriber is, and, realising that the user is not able to receive faxes, will then convert the fax into speech before calling the mobile number and reading the message over the phone. Alternatively, it may merely alert the user to the fact that a fax is waiting.
How would this be achieved? IBM says it will use an international network of computers which holds each user鈥檚 mailbox. These mailboxes will be transformed into intelligent agents by programming in rules such as: 鈥淚f I receive a fax over the weekend, please forward it to my electronic mailbox and telephone me to say it is there.鈥� To make the service easy to use, subscribers simply select options from a series of menus. Eventually, IBM wants these smart mailboxes to learn user鈥檚 preferences and take decisions on their behalf.
Gilbert, however, stresses that it is the ability to simplify the customer鈥檚 interaction with the network and forward messages on to the users, wherever they are, that is most important. 鈥淓xecutives may not yet be ready for half-baked learning technology,鈥� he says.
Intelligent routing technology also lies at the heart of another idea floated by BT researchers at Martlesham. Like IBM鈥檚 ICS, BT wants to exploit its existing network to make it easier to find people and pass on messages to them. One option is for the user to carry a small transmitting device 鈥� perhaps a personal digital assistant, a mobile phone or something simpler. This lets the network know its position by communicating with local base stations, or alternatively by directly communicating with other mobile phones or personal assistants connected to the same network. A message can then be sent back to the mailbox agent, carrying details of the local mailboxes, fax machines and printers.
Once the information regarding the local devices is available to the user鈥檚 agent, it must then negotiate for permission to use these facilities. If, for example, an executive is visiting a hotel, and wants to print a chart the office has sent, the agent may have to agree to pay for usage, and supply credit card details. Smith envisages that every piece of equipment participating in the network will have its own agent, which will 鈥渘egotiate鈥� over the kind of messages to be received, whether there will be a charge, whether the message needs to be encrypted, how long it should be stored for and so on. This way, says Smith, a network customer could receive any kind of digital message 鈥� from voice to video 鈥� at a variety of locations without carrying any specialist equipment around.
But if you want to manage information spread over innumerable mobile devices, why have your intelligence located in static mailboxes? Would it not be better to have intelligent pieces of software prowling the networks themselves? General Magic, a small Californian company set up by a group of software experts who used to work for Apple Computer, believes this could be the winning approach.
It has developed Telescript, a special computer language and software environment which allows mobile agents to explore networks and carry out their tasks. One of the key advantages of this approach is that a user can send off agents to search out information or carry out transactions, and then disconnect while they do their work. The agent is directed like an e-mail message into a remote database, from where it can initiate a process (such as a database search or order some goods from an electronic warehouse), and then leave when it has achieved its objective.
This behaviour, however, closely mirrors that of a computer virus. And it would only take a slight change in the software to turn a helpful agent into a vengeful virus, or so the argument goes. General Magic has attempted to guard against this by making all Telescript agents reveal their identity before being allowed into a computer. There are also controls on the amount of time an agent can spend in a computer鈥檚 database, the kind of tasks it can carry out and an encryption procedure for transferring data from one device to another.
Executive decisions
So far, only AT&T has developed a commercial service based on Telescript. But the approach is attracting interest from other telecommunication companies which are at an earlier stage of testing the technology. Other organisations such as BT, are also carrying out research into the potential for to manage network traffic. Last year, it used the analogy of an ant colony and the insects constant search for food (Technology, 5 March 1994) to explore the concept. According to Simon Steward, a researcher into agent technology at BT鈥檚 Martlesham laboratory, 鈥淭he ants problem is to collect food. Our agents鈥� problem is to read out network capacity.鈥� In place of the ants鈥� instructions to wander across the screen looking for food, the agents would be programmed to travel across a network collecting data about traffic levels. Once they find an overloaded node, or an underutilised one, they could be programmed to take different kinds of action. For example, they could 鈥渞eproduce鈥� spawning worker ants whose only job is to travel to a nearby exchange and ask for it to redirect some traffic. 鈥淨ueen ants鈥�, which, like their biological counterparts don鈥檛 move around, can also be used to create and destroy all the other ants.
But is this going too far? Mobile agents, says BT鈥檚 Smith, make everyone nervous, mainly because of their ability to act like viruses. 鈥淭he risk is too high at the moment,鈥� he says. Smith points out that, in fact, many telecommunications carriers are cautious about the whole idea of using agents because their ability to produce stable networks has not been proven. Smith cites the 1987 鈥淏lack Monday鈥� stock market crash as one reason why. Then, computer programs were instructed to sell when stocks fell below a certain level 鈥� thereby contributing to the speed of the crash.
How long these concerns can be allowed to flourish is, however, debatable. The Internet alone is forecast to have more connections than there are people in the world by the early part of the next century. And telecommunications companies are already offering an ever broader ranges of services. Without enlisting some help from 鈥渋ntelligent鈥� programs, the power and complexity of these networks will overwhelm the humans who created them.