On the Web everywhere is as close as everywhere else, so location is
irrelevant. The next shop is no longer a tiring bus ride away: it is just a
couple of clicks down the I-way. Far more important than location is
reputation—the way a business treats its customers, how it settles their
complaints, and how it deals with its trading partners.

This used to be the case in the real world too. When folk bought everything
at the corner shop, everyone knew when the owner was treating people badly, and
the shop would lose trade. It is early days for electronic commerce on the
Internet and everyone feels they have just moved in and are finding out where
the shops are. Because it does not take much to make a website look like a
reputable business, some people are, sadly, getting ripped off. Gradually,
however, people are learning, and they are helped by sites that are able to
reveal who can be trusted and who is offering the best deals.

One such company is SeaFax of Portland, Maine, which provides services for
fishermen and seafood suppliers on the East Coast of the US
(This Week, 29 August, p 15).
SeaFax acts as an honest broker and tells those with a catch or
warehouse full of soft-shell crabs who wants to buy such things and, just as
important, who has a history of being a good customer. Restaurants or
supermarkets who don’t pay their bills on time could soon find that no one is
willing to sell to them.

Seafood becomes inedible more quickly than most other foodstuffs, so the
speed of the Internet is vital to the success of the SeaFax service. But there
is no doubt that honest brokers will spring up for other industries, too. If you
never meet your trading partners in person you have to be sure that it is good
business to strike a deal with them.

The importance of a good trading reputation is growing rapidly. Already it is
possible to use intelligent agents to trawl the Web for the cheapest music CDs,
software and pretty much anything else. It is just as easy to get a snapshot of
a reputation—all you do is post a message to a relevant Usenet newsgroup,
asking who can be trusted. The Internet is a massive consumer watchdog, and
asking the question will probably elicit more than enough information to make an
informed choice.

This can only be a good thing. It gives power back to consumers and ensures
that they get a good deal. It may also mean that we come to value relationships
rather than objects—a profoundly different situation from now.

But just as businesses must watch their reputations carefully, so too must
customers. Many people believe that the growth of electronic commerce will mean
that eventually we will be using intelligent agents to do most of our shopping.
The agents will need an element of autonomy because of the speed with which
bargains come and go. However, by the time your agent has tracked you down and
asked if it should buy that antique vase, chances are that the vase will have
been snapped up by someone else.

Autonomous agents will need to be told how to strike a deal. If you create an
unprincipled agent that rips off a lot of other agents you could find that no
one will sell to you because of what your agent has been up to.

The importance of reputation does not end with shops and consumers. It may
even extend to politicians. Project Vote Smart in the US tracks politicians and
lets voters know who is sponsoring them and how they are voting. If they promise
one thing to voters and then vote the other way in the chamber, they will not be
able to hide the fact. Who knows? One day we may have too many honest
politicians to choose from.

Most mistakes in spreadsheet programs arise when the programmer links the
wrong columns or someone enters the wrong figures. Research by Ray Panko of the
University of Hawaii has found that over 30 per cent of spreadsheets contain
errors, many of which arise because the programs are now so easy to use
(This Week, 16 August 1997, p 13).
Spreadsheets generally allow any combination of
columns to be linked without checking what is being represented by those
columns.

Now Jocelyn Paine of Oxford’s experimental psychology department has created
Model Master—a software interface for spreadsheets that he claims makes
these mistakes less likely. Unlike the scripting languages used by popular
spreadsheets, Model Master resembles a programming language. It forces people
who use it to be much more rigorous about the variables they employ, and how
they are combined.

Instead of numbered cells, which are easy to confuse, Model Master uses named
variables. This reduces the chance that the programmer will make a link to the
wrong cell or column. “Because you have to give variables names, it guards
against major slips,” says Paine. He adds that Model Master can check for
“dimensional” errors—ensuring that money is not being multiplied by money,
for example. It is also much simpler to reuse old formulas on new spreadsheets
because they are written in discrete chunks, called objects, that can easily be
exported.

Panko says he is aware of Paine’s work. “It’s just one of several new tools
in the works for developing spreadsheets and auditing them,” he says. “Hopefully
they will reduce errors to reasonable levels.”

Paine is now working on a “point-and-click” interface to Model Master so that
people unfamiliar with programming can build in formulae.

Caldera is seeking damages for lost sales due to what it alleges is a
built-in incompatibility. Microsoft denies the claim. Although Microsoft has
handed over some of the code that Caldera wanted, it says it has lost some of
the source code for Windows and the older MS-DOS software. Jim Cullinan, a
spokesman for Microsoft in Seattle, says of the missing pieces: “They’re very
difficult to find because they’re six or seven years old . . . Source code is
our most valuable asset, but in the technical field [this particular source
code] is far outdated.”

Caldera chief executive Bryan Sparks says the lack of some of the source code
should not damage his company’s case, but he believes that the source
code—complete with programmers’ comments—might have helped in court.

Many surfers object to being tracked in this way, especially if the
information being garnered is used commercially. So Michael Reiter and Avi Rubin
of AT&T Labs in Florham Park, New Jersey, have found a way to give Internet
surfers more privacy by making them a mere face in a crowd.

People using their software, called Crowds, become anonymous members of a
pool of surfers. When a member of the pool clicks on a link or types in a Web
address, the request for data does not go directly to the desired site. Instead,
it is sent along a randomly determined route through the crowd. The chances are
that it will turn up at the website with another member’s Internet address,
spoiling attempts at profiling individual users. The downside of mixing up
addresses is that a surfer could get into trouble for apparently visiting
websites, such as sex sites, that were actually visited by another crowd
member.

To become part of a crowd, users download a program called a “jondo” after
John Doe, the name sometimes given to unidentified corpses in the US. The jondo
effectively acts as a “proxy” server, taking the request for information and
randomly picking a member of the crowd whose address will be used to pass on the
request.

After a few hops around the members’ computers, the request reaches the
desired website, which sends data back down the path that has just been
established. The request looks to the destination website as though it has come
from some random member of the crowd, says Reiter.

Passing requests from machine to machine slows down surfing slightly, so
Reiter says crowd membership is currently limited to those with at least a 64
kilobit per second ISDN link to the Internet.

Reiter says there has been a steady level of interest in the Crowds program.
AT&T’s crowd has 900 members, not all of whom work for the company. The
larger the crowd, the greater the anonymity of each member and the lower the
delay on any surfing trip. Even when not surfing, crowd members often keep their
links to the Internet live, giving the program as big a choice of addresses as
possible.

Crowds does not give perfect privacy, Reiter admits, but he claims it has
advantages over other “anonymiser” services that funnel requests via central
servers. It allows people to legitimately claim that it was not them that
initiated a request to a prohibited website, for example, even if their machine
originated that request.

As an added safeguard, Crowds routinely strips out cookies, the small packets
of data generated by many websites and stored on a visitor’s computer. Reiter
says this can put a few websites off limits because some allow access only to
surfers with valid cookies. He adds that Crowds protects privacy best when Java
and Javascript are turned off, because both these Web page add-ons can
compromise anonymity.

Neal Workman, a former fisheries debt collector, thinks he has the answer:
sites that act as trusted third parties can become market moderators, acting as
honest brokers. He has turned his knowledge of the financial state of
restaurants, supermarkets and food-processing companies into a profitable
website called Go-Fish, which allows fishermen working the seas off the coast of
Maine to find solvent buyers for their catches.

The service is hosted by Workman’s company, SeaFax of Portland, which
provides financial and credit information on the food industry. The SeaFax
website is also the gateway to several other services. It provides a 24-hour
news feed about suppliers and restaurants, lists the credit ratings of these
companies and hosts a Java-based market that matches sellers to buyers.

As electronic commerce becomes more widespread, Internet experts believe that
services like Go-Fish will become increasingly important.

Chris Solomon and his colleagues in the physics department at the University
of Kent have developed a way of creating a unique identifying code for every
face by combining images from a library of “universal” faces.

To create this library, the researchers produced digitised images of 290
faces, most of them of students. They then mathematically analysed the images to
capture their most important features, such as nose shape or eye width. From
this information they created another, smaller library of definitive “universal”
face types, which they ranked in the order of the degree to which they represent
the diversity of features within the population. Faces that represent the
largest sector of the population were put at the top of the list, while the
least typical ended up at the bottom.

To encode the image of somebody’s face the researchers begin with the
“average” face—the most representative one. Then other universal faces are
overlaid on this first face, emphasising or playing down features such as
eyebrows or noses until the image reaches the closest match with the
individual’s face.

“It’s possible to show that anybody’s face can be produced by an appropriate
combination of these faces,” says Solomon. “With a large enough population, it
should be possible to re-create any human face efficiently. “The method differs
from the Identikit images police use to build up distinguishing features because
it builds up images using whole faces, rather than separate facial sections.

The faces in the library are numbered, and the unique combination needed to
recreate the face of a person defines their “facial PIN number”. Solomon says it
takes between 50 and 100 faces to match an individual’s face adequately. This
combination of numbers easily fits into 50 bytes and can be recorded on the kind
of magnetic strips used on credit cards. One of the tracks on these strips is
largely unused and has 65 bytes of space in it. By contrast, the widely used
JPEG image compression standard leads to much more data. “There’s no way on
earth that JPEG could get near to 50 bytes,” says Solomon.

When a credit card carrying the unique number is swiped through a reader a
computer at the checkout could use the PIN to re-create an image of the face for
visual identification. Alternatively it could be used as part of an automatic
recognition system to control access to a building, or to cash in a bank
account.

Smart cards, which have tens of kilobytes of memory, can store larger images
but are expensive. The ability to store an image on the far cheaper magnetic
strip card is important to companies like banks that might have to make millions
of cards for their customers.

Richard Freeman and his colleagues at Scientific Generics in Cambridge make
their can with four layers of thin cardboard. The layers are staggered so that
the seams of each sheet meet at 90-degree intervals around the can. The seams
are usually the point of failure in a tube, Freeman says, but with this design
each seam gains strength from the sheets above or below it.

A thin aluminium membrane is sprayed onto the inside of the cans to make them
airtight, and the ends plugged with shallow cones that point outwards. Opening
the can involves removing one of the cones, turning the sealed tube into a
beaker. Freeman says that prototypes have withstood pressures of up to 480
kilopascals (about 5 atmospheres), higher than the 275 kilopascals fizzy drinks
are usually pressurised to.

Scientific Generics is looking for a partner to develop the idea. Freeman
fears that large drinks companies will be wary because they have so much
invested in aluminium cans. He says: “The challenge is finding someone with the
will to make it happen.”

The smooth running of the skies over Britain affects the whole of western
Europe and even North America. Thousands of flights between continental Europe
and the US cross British airspace, and that space has never been more crowded.
Officially, the volume of traffic is growing by about 5 per cent a year;
controllers, however, think the annual rate is edging closer to 8 per cent.
British controllers frequently have to ask other airports to keep planes on the
ground while space clears over Britain.

Traffic jams in the sky are also becoming more common. In June, the number of
planes flying through British airspace broke previous records on three
occasions. And in one month this spring there were six
“overloads”—incidents when designated areas of airspace contained more
planes than controllers normally expect to handle. This compares with a dozen
overloads in the whole of 1997.

For passengers made nervous by these statistics, the latest developments are
not reassuring. There are “fundamental” problems with the delayed computer
system, according to Peter Ladkin, an expert in computer networks at the
University of Bielefeld in Germany. Ladkin has given evidence to a select
committee of MPs investigating the problem.

The whole issue of air-traffic control has been dogged by controversy since
last year when, in a conspicuous U-turn, the newly elected government announced
plans to privatise National Air Traffic Services, the body responsible for
managing the flow of planes in British airspace. The sale of NATS is supposed to
raise capital to help improve the control system. So far, however, there is no
obvious buyer and no one is clear how the service could make a profit. Unions
representing the air-traffic controllers are waiting anxiously for a government
consultation document—due to be published shortly—to learn their
future.

For the past 30 years, most air traffic over Britain has been handled from
three control centres—at West Drayton near London’s Heathrow Airport, at
Prestwick in Scotland and at Manchester. The ageing equipment is running out of
capacity.

The new computer system was conceived in 1988 as part of an ambitious plan to
cope with the predicted growth in air traffic. Originally designed by a division
of the computer giant IBM that was subsequently taken over by Lockheed Martin,
the system was one of the most advanced in the world and was intended to come
into use in 1996. The plan called for the closure of the Manchester centre and
the building of a new, larger centre in Swanwick, near Southampton, to replace
West Drayton. At the same time, Prestwick was to be upgraded so that it could
take over from Swanwick in an emergency.

The US Federal Aviation Administration was installing a similar system, also
built by the former IBM division. But the FAA encountered numerous technical
problems with the system and abandoned it in 1995. Undeterred, Britain pressed
on. By January 1996 it was clear that the system would not be ready for launch
that year, but still NATS persevered. With the fifth delay announced last week,
the system is now running four years late. Swanwick stands idle.

In April, MPs sitting on the Select Committee on Environment, Transport and
Regional Affairs issued a highly critical report into the development of the
system. “We find it remarkable that the previous management of NATS decided to
press on with pioneering this highly complex project after the US Federal
Aviation Administration abandoned a similar system.” The select committee
declared itself “astonished at the apparent complacency of the Department [of
Transport] about progress with the project in recent years”.

The committee recommended that the government should appoint an independent
auditor to scrutinise the management of the project and find out whether it
could be saved. If not, the report said, the government should consider
scrapping it and starting again. Some of those who gave evidence to the
committee said that the management of the project showed “some of the classic
early-warning signs of a public sector computing disaster”.

An internal check of the new system in 1995 found 21 000 defects. Details
have not been released, but Malcolm Field, the chairman of Britain’s Civil
Aviation Authority, the parent body of NATS, says that 95 per cent of the faults
have been eliminated. Ladkin believes that the remaining problems will be harder
to resolve.

What little is publicly known confirms this. When the system was tested on 30
workstations it worked well. But when it was tested on 160, it failed. “This
means that there are problems at a fundamental level,” says Ladkin. When
operational, the system would need to work on about 200 workstations at a time.
Lockheed Martin says that its system has undergone extensive testing and now
works well on 194 workstations. It says that only minor faults remain to be
cleared up.

Ladkin points out that if a large software project suffers problems while it
is being developed, it is likely to be hit by even greater problems when it is
being “bedded in” and tested. He points to lessons from the US, which is slowly
installing a replacement for the system it abandoned in 1995. The FAA is now in
dispute with the American air-traffic controllers’ union, which claims that the
new system is flawed in many respects. They say the flaws only became apparent
when controllers were trained in its use and discovered that information they
were accustomed to having was no longer available. Until the alleged problems
are ironed out, the FAA is using its system only at night and when air traffic
is light. Lockheed Martin, the system’s developer, says only small changes are
needed to correct the problems.

In Britain, the outlook remains uncertain. The government has expressed
concern about the delays, but has also said it is confident that West Drayton
will be able to cope until Swanwick becomes operational. It will be asking the
Defence Evaluation Research Agency to audit the project. In a response to the
MPs’ report, the government said that it sees “no fundamental reason why the
Swanwick centre should not be brought successfully into operation”.

Ladkin is not so sure. “The rule of thumb is that 20 per cent of costs are in
forward development and 80 per cent are in maintenance,” he says. So far the
development of the Swanwick computer system has cost ÂŁ163 million. If
Ladkin is right, then fixing the problems may cost ÂŁ650 million, pushing
the total costs above ÂŁ0.8 billion.

This may explain the government’s eagerness to privatise NATS, in the hope of
netting an expected ÂŁ300 million to reinvest in the service. But the
unions suspect that financial concerns may lead to a compromise on safety. One
union, IPMS, has launched a campaign to fight the deal.

If the government is successful, NATS will become the only privatised
air-traffic control system in the world. It will also be operating a computer
system of a type that others have abandoned. This could be a real flight into
the unknown.

Some of these simply slow browsers down but others are written specifically
to damage a computer, often exploiting security loopholes to steal or corrupt
data. Growth of the Internet is giving these malicious programs great scope for
destruction.

Now the International Computer Security Association (ICSA) is setting up the
Malicious Mobile Code Consortium to look into the problem, set standards for the
products created to tackle bad applets, and educate the public about the
dangers.

Larry Bridwell, anti-virus program manager at the ICSA, says although there
are only tens of malicious applets compared to the thousands of conventional
viruses, the industry does not want to be caught out as it was with macro
viruses—programs that can infect documents. These first appeared in summer
1995 but now they are the most prevalent type of virus.

Bridwell says the combination of security loopholes and anti-virus software
that cannot catch malicious programs means that the problem could grow quickly,
given the vast number of potential users.

The computer simulation has been developed by Maja Mataric from the
University of Southern California in Los Angeles, Victor Zordan of Georgia
Institute of Technology in Atlanta and Zachary Mason from Brandeis University in
Waltham, Massachusetts. Their chief aim, says Mataric, is to create mechanisms
that robotic hardware and software can use to learn any task.

Earlier work had established that humans look at their tutor’s hands when
learning a physical routine, such as the movements associated with a dance. But
when we actually come to do it, the correct arm, head and neck movements follow.
Our own set of behavioural primitives—the basic movements on which more
complex ones are based—means that when we do something with our hands, the
rest of the body follows automatically.

Mataric says imitation is extensively used by animals and could make robots
much more flexible. Although it is possible simply to copy control software from
one robot to others, because each is unique they will not do the job as well as
they would if they had “picked it up” by observation.

Before Adonis can start to watch and learn, he first needs a basic set of
behavioural primitives to build upon. So the researchers decided to teach him to
make the arm and hand movements of the macarena. They decided to simulate a
dance because it is easier to spot when Adonis has got it right. More subtle
movements would be harder to evaluate. “If you like the macarena, you’ll know if
it’s right or not,” says Mataric.

She adds that the macarena is ideal because it involves distinct movements
that can be broken up into easily learnt parts. In the dance, the hands are
moved and placed on different parts of the body. Mataric says the changes in
direction and speed in the dance will provide useful points for splitting up the
behaviours later.

The basic behaviour set is based around 12 hand and arm-related movements of
the 14-step version of the macarena. Spanish singing duo Antonio Romero and
Rafael Ruiz—aka Los del Rio—had an international hit with the song
Macarena in 1995. It was released in several versions with different
tempos and numbers of movements. The dance involves extending the arms, touching
hands to different parts of the body and crossing the arms. The two movements
that Adonis can’t do involve jumping a quarter turn and swinging the hips.
Mataric and her colleagues simulated only the upper body so Adonis has no feet
to jump with and no hips to swing, though he can bend at the waist.

Adonis’s head, neck and arms are subject to the same constraints as those of
humans. His joints have the same degrees of freedom that human elbows and wrists
do, and he can’t cheat by swinging his hands through his head. He also has to
contend with the software equivalent of gravity and inertia.

Now that Adonis can do the macarena, Mataric and her colleagues are planning
to add a sensory component to the simulation—for example, video
cameras—to allow Adonis to watch someone do another dance such as the
mashed potato or the watusi and then use parts of the macarena to mimic what he
has seen.

This will be a key advance in artificial intelligence. “If you are learning
from scratch every time, that is not as powerful and not the kind of thing that
is seen in biology,” says Mataric.