AI gets down to business

An electronic butler that answers your questions using all the resources of the Internet. An artificial infant you raise from birth, teaching it and helping it develop its own unique personality. A security system that learns your attacks and improves itself.

No, not science fiction: these are all applications of artificial intelligence in use today. We may not have a HAL-like computer system today, but AI is being used to enhance a wide variety of products and create new applications, and it is fuelling some of the latest hot Internet startups.

The definition of "artificial intelligence" depends on whom you ask, but businesses have all sorts of uses for smarter computers. "One common definition is that if a computer does something where, if a person did that job, you'd say the person was intelligent, then you can say the computer is intelligent," says Dave Cliff, a former associate professor at the AI labs of the Massachusetts Institute of Technology, now technical lead at Hewlett-Packard Labs, in the digital media systems department.

"People are expensive and error-prone and hard to get hold of. In general AI can be used to improve products and make machines do things that otherwise you'd need a human to do," Cliff says.

That could be something as simple as the self-repairing photocopiers and laser-printers of today, that eliminate the need for constant visits by a repairman. In reality, such simple applications aren't usually considered AI. "AI has always been plagued by this problem: as soon as machines have the ability to do something, people think maybe you didn't need to be that intelligent to do it after all," says Cliff.

He uses the example of chess-playing, which was once thought to require intelligence. "Then the Deep Blue team beat Gary Kasparov, and now it's not considered an AI problem," Cliff points out. Likewise, speech recognition is within reach of any reasonably powerful PC, but is not considered a mark of intelligence.

One of the hottest frontiers for AI is e-commerce, where e-tailers are hoping to make the online world an ever more human place. Web sites such as Amazon.com already use a process called collaborative filtering to compare our buying patterns with those of other customers and make recommendations.

A new generation of products hope to use AI to one-up Amazon's methods and win customer loyalty for the sites that use them. Products from US startups such as Saffron Technology and Manna aim to learn the individual user's buying patterns and make personalised recommendations accordingly.

Such technology could be the best way of giving retail sites a more personal feel. "You need to have neural technologies in place, constantly updating themselves, so that they can deal with conditionalities," says analyst Derek Brown with Robertson Stephens. "One day I might be interested in Manchester United, but that doesn't mean I always am. Sites need to change depending on what my behaviour is at the time."

But Brown sees a bigger role for AI in making sense of the Internet's chaotic mass of information. Today, companies such as the UK's Autonomy specialise in software that can sift through all sorts of documents used by businesses and put them into a clear order.

The software uses a form of AI to recognise patterns in the documents and guess at the gist of their overall meanings. It can then sort the documents into a hierarchy and link to other relevant documents.

In the future, wide use of XML -- which includes tags identifying a document's content -- could make such software unnecessary. But AI could still be used to replace simple sorting jobs humans do today, such as reading a news story and determining which section of a Web site it belongs in, or listening to a piece of music and deciding whether it's country, pop or jazz.

Web robots don't necessarily carry out tasks for one Web site. Many researchers envision a world of semi-autonomous "agents", roaming the Web and carrying out various tasks for their owners. Present software such as the "mobile agents" of Netherlands-based Tryllian could be the forerunner of intelligent bots making purchases and carrying out other business transactions without human intervention. Such agents could be given a rough idea of what we want, do some comparison shopping and order the best deal, just like a real personal assistant.

Cliff even envisions virtual organisations composed of autonomous agents, which could form spontaneously to carry out a specific task and then disband again.

Of course, there's the question of whether we really want to cede control over our affairs to an artificially-intelligent piece of software, which might even have its own legal powers. "Some autonomy is good. Absolute autonomy is scary," Cliff says. "Our legal system isn't prepared yet for high autonomy systems. For example, if I give it my credit card, would I be violating my credit card agreement?"

On a more practical level, it isn't clear users would really find agents useful. After all, it was clear-headed, original thinking about AI that resulted in the much-reviled personal assistants in Microsoft Office. Another potentially irritating application, under development by Autonomy for Tesco, would notice when you visit a competitor's food-shopping site and then suggest Tesco's lower prices.

"I see the Internet as a pull medium," says Brown. "People know roughly what they want, and they seek it out. The idea that I want things to be delivered to me automatically might be interesting, but it isn't a killer app."

A compromise solution is to build a more human interface for tasks that have already proven popular, such as Web searching. Sites such as Ask Jeeves are using AI in the hope of making the Internet a more intuitive place, where you get things done the same way you do in the real world.

Ask Jeeves believes human-style interfaces will eventually help people do everything from shopping to setting up their new answering machine, and will be indistinguishable from talking to a real person. Sports giant Nike is already using an Ask Jeeves interface to help sell shoes on its Web site, asking the user what they want and then finding the best matches. "It talks you through the refinement process," says Donald Clark, product development manager at Ask Jeeves UK. "It's like putting your best salesperson on the Web."

Ask Jeeves' humanistic objectives are shared by MIT's Project Oxygen, a new research project aimed at making computing more pervasive and friendly. The project, launched last year, is supported by Acer Group, Delta Electronics, HP, NTT, Nokia Research Centre and Philips Research.

One very big business in which AI plays a key role is the video-game industry, comparable to the film business in size. Unlike in the movies, it's often up to a computer or game console to create a sense of reality for the gamer, and standards of realism are going up all the time.

"It's important for a game to make people feel that the characters who are inside the game are kind of real," says Jez San, chief executive of Argonaut Games and a major figure in the UK's world-leading game design business.

Some games use neural networking technology to create characters who can learn as they go along. Characters in a fighting game, for example, might be taught combat skills in much the same way as a human.

A new genre of games has even emerged around interacting with simple AI personalities -- something like the Tamagotchi phenomenon. The best known example so far is probably Black & White, still in development by Peter Molyneux and Lionhead Studios, which will allow players to nurture a creature through its life and help build up its behaviour patterns.

Demand for realistic AI entities may be about to emerge from an unexpected source: online multiplayer games, which link thousands of people together in role-playing or strategy extravaganzas. "The problem is that people have to eat and go to the bathroom etc," says San. "Wouldn't it be nice if you could send in an artificially intelligent agent, and your character could keep on playing while you did that? It's all theoretically possible."

But San admits truly intelligent, autonomous AIs aren't on the cards for gaming -- and won't be, until there is a clear demand. "When games exist that need that intelligence, it will come," he says.

In the mean time, the same intelligence that lets a video-game fighter predict your next move is also helping antihacker security systems better protect sensitive computer networks. The idea is that a smarter system will be better able to recognise hacker threats -- such as an intruder gently probing defences over a long period of time -- and take action.

One such project, spearheaded by Sandia National Labs in Albuquerque, will use AI agents to pool the forces of as many Internet-connected computers as possible into one vast, collective security force. The Linux-based system will require no specialised security computers and will be able to share information and form a consensus about the nature of any irregularity it spots.

The system can take action by closing data ports, rejecting viruses and cutting a hijacked computer off from the network.

No industry so far has found a need to come up with a "true" artificial intelligence -- the sort of conscious artificial being imagined by Isaac Asimov or Arthur C Clarke. It's a question of usefulness, say experts. "A lot of AI companies are evangelical about their vision, but what they evangelise about doesn't necessarily turn into a practical or useful product at the end of the day," points out Brown.

And there's no guarantee the ends of scientists and businesspeople will get any closer together than they have ever been. "Creating something conscious, there isn't necessarily a business case for that," says HP's Cliff. "There might be a Nobel Prize in it, but it's more of a scientific goal, rather than engineering. The question we have to ask is, can we create products which do useful things?"

Read ZDNet's Artificial Intelligence Special to learn about a future of sentience, human implants, and moral questions -- and the threat robots could pose to humanity.

Have your say instantly, and see what others have said. Click on the TalkBack button and go to the ZDNet News forum.

Let the editors know what you think in the Mailroom. And read what others have said.