Å®ÉúСÊÓƵ

Another stretch of the information superhighway should be slotted into
place within the next five to ten years, according to British Telecom. Telephone
and videophone calls in future could be arranged by autonomous software
‘agents’ which will use artificial intelligence to negotiate with each other
on behalf of their respective owners – the caller and telecoms companies
– to find the optimum mix of bandwidth, price and time for making a call.

The user, however, will be blissfully unaware of the entire process,
and will simply be offered a range of options, such as a low-quality but
cheap connection or a more expensive, high-bandwidth link, before making
the call.

Robin Smith, BT’s strategic development manager at its research and
development headquarters in Martlesham, Suffolk, says that intelligent agents
will soon take over control of BT’s networks, which are becoming so complex
that they are quickly outstripping people’s ability to manage them directly.
‘Centralised control is becoming a major bottleneck in BT’s network management,’
he says. ‘And NTT in Japan, the world’s largest telecoms company, has to
employ 20 000 people just to produce software to run its network. And if
that software fails then the network is out of control.’ BT has begun using
a simplified system with 150 agents in demonstration projects, and expects
to use them on the network in three or four years’ time.

However, these internal agents are only half the story, says Smith.
As telecoms networks grow in number, size and complexity, users will find
it increasingly difficult to remember what the best choices are – such as
which networks offer cheap calls and at what time, and which offer the best
price for a high-bandwidth, long-distance mobile call. In the future, such
prices might even vary minute by minute, depending on demand. Users will
thus need their own ‘customer agent’, which will negotiate with the networks’
agents to find the best prices.

Agents are small programs which may reside on one computer, when they
are known as ‘fixed’ agents, or copy themselves to other systems and carry
out their instructions there (‘mobile’ agents). Nicholas Negroponte, director
of the Media Lab at the Massachusetts Institute of Technology, describes
them as ‘electronic butlers’ which will learn their owners’ likes and dislikes,
and navigate through the increasingly complex web of computer systems holding
useful information.

The most difficult capabilities to confer on agents are negotiation
skills, says Chris Winter, a manager on BT’s development team. ‘People know
how to trade off one thing for another, but programs tend to have fixed
goals,’ he says. ‘We have to teach them how to trade.’

A key debate among software developers is whether agents should be mobile
or fixed. Like the infamous viruses, a mobile agent arrives at a particular
computer from outside, copies itself onto the computer and uses its resources
to carry out instructions there. ‘What greatly disturbs me, and other large
users, is the idea of a virus being spread in the guise of an agent,’ says
Smith. ‘Only time can tell if those fears are real.’

But fixed agents may be less effective for certain tasks, says Winter.
‘Some customer agents work better if they can move around; others if they
stay still. Fixed agents can only communicate with others if they send some
sort of electronic mail. The problem then is that if they are tracking a
user who is moving around a lot, the amount of signal traffic that is generated
just to track a user starts to become overwhelming, and takes away capacity
on the network which the user really needs. In that situation mobile agents
are better.