Robo Software Reports For Duty

Robots have long captured man's imagination as tireless, eternally loyal servants. But even today, with many factories having gone robotic, few people have ever seen a true robot, much less owned one that they could boss around.

All that may change now, as a new kind of robot starts prowling the digital highways and byways of cyberspace. It's what computer scientists call an "agent"--a kind of software program that's powerful and autonomous enough to do what all good robots should: help the harried humans by carrying out tedious, time-consuming, and complex tasks. Software agents just now emerging from the research labs can scan data banks by the dozen, schedule meetings, tidy up electronic in-boxes, and handle a growing list of clerical jobs. In the next couple of years, experts predict, agents will be available from all the major software companies and quite a few startups as well.

SUPERHIGHWAY SURFEIT. At Hewlett-Packard Co., an agent supplied by Edify Corp. is already helping automate a quarterly wage-review process that covers approximately 13,000 salespeople. Edify's software, running on a PC, performs essentially the same tasks that a team of 20 administrators has been doing by hand. First, the agent program dials into the HP personnel system and gets a list of who works for each of 1,200 sales managers nationwide. The agent then electronically mails each manager a list for his verification. It collects any changes by e-mail and enters them back into the personnel system--exactly as if a human were working at a terminal.

Now, the agent repeats the entire process, but this time it includes proposed salary changes supplied by management for each employee. Managers can approve the changes or alter them simply by phoning the PC and keying in new data in response to the Edify agent's synthesized voice prompts.

An entire menagerie of software agents is starting to debut--some sophisticated, some simple, some quite visible, others programmed to work strictly behind the scenes. Ideally, they will mimic just what an intelligent human would do, except with greater stamina and accuracy. This might be the only way that mere mortals will be able to cope with the widely hailed Information Superhighway and its cornucopia of services and machine-generated messages demanding immediate attention.

The windows-icon-mouse interface that Alan Kay and other computer scientists developed at Xerox Corp. in the 1970s isn't keeping up with the complexity of networks. Unassisted, people can keep track of only so many graphical icons, for instance. "You'll be connected to millions or billions of different resources," predicts Kay, now a research fellow at Apple Computer Inc. "One of the biggest problems is to find information. We don't believe you can browse for it." To find what you need, he says, the computer "has to be flexible enough to take on many of your goals. You either tell it or teach it what you're interested in, and dispatch agents. The agents will find things and screen them."

PLAUSIBLE LIES. To make that happen, agent developers are building on the past decade's research in artificial intelligence (AI). Agents are incorporating techniques for understanding human language and learning their masters' wishes by observing how they search for information. Says Ted Selker, a researcher at IBM and professor at Stanford University: "We want to create a tight feedback loop between the user and his computer assistant." That way, an agent might scour a large network of data banks to answer a difficult question, such as "How's the Asian retail market doing compared with Europe's?"

Until recently, even simple agents weren't practical, although AI researchers had sketched out the concept in the late 1950s. "The way you do research is to make up lies that are plausible--then invent the technology to make them true," Kay quips. Now the technology is in place: sprawling networks of powerful computers that are capable of running complex software. In the next few years, everything from personal digital assistants (PDAs) to desktop PCs and supercomputers will likely run some form of agent software. Just how big a market it will be is hard to estimate: The term "agent," like AI 10 years ago, is being applied to new products quite sloppily--in this case, to just about any program that works on your behalf.

In the tradition of those wise-cracking robots in sci-fi movies, software agents are also subject to much anthropomorphizing--and in some cases for good reason. At Carnegie Mellon University, Stanford University, and software think-tank Interval Research Corp., researchers are trying for "believable agents"--programs that embody human character and present themselves as animated cartoon faces that react to stimuli with convincing human expressions. At Carnegie Mellon, a family of such agents called Woggles--Shrimp, Bear, and Wolf--populates a whimsical virtual world. For entertainment, people can interact with them via another, mouse-controlled Woggle. One day, similar agents may make computers more human by giving their operating systems believable screen personas--frowning if you do something foolish, say. "We want to create agents that gain our confidence," says Joseph Bates, senior research computer scientist at Carnegie Mellon.

Not all agents strive for humanity, though. The Internet, connecting thousands of research computers around the world, relies on agents called gophers, which automatically retrieve information from all over the network. And Rupert Murdoch's News Electronic Data Inc. is getting ready to unleash Oliver, which will fetch personalized news and travel information. It's to appear on screens later this year as a Labrador retriever: "Oliver is loyal," says John Evans, president and CEO. "He retrieves personalized information for me. He's tireless, and he doesn't make a mess. But he's a stupid dog: If you throw two sticks, he'll get confused as hell. We're just trying to lower people's expectations" about agents.

American Telephone & Telegraph Co., meanwhile, is heavily promoting the intelligent software assistants scheduled to be available on its forthcoming PersonaLink computer network. They'll be programmed with software from General Magic, a Silicon Valley startup that's funded by AT&T, Sony, Motorola, and several other large electronics companies. The agents will help consumers communicate and purchase information, goods, and services from companies such as Mead Data Central Inc. A mail agent might have you paged when important e-mail arrives. A personal shopping agent could browse electronic malls for you and even make purchases using your credit-card number. Who knows? Perhaps in the future people doing business in cyberspace will just say: "Have your agent talk to my agent."


How personalized intelligent assistants, or software agents, can help people cope in the Information Age:

NETWORK AGENTS They'll scan data bases and electronic mail, schedule meetings, and help with travel arrangements

CLERICAL AGENTS In offices, they help answer phones, tap into computers for customer data, and send faxes

USER-INTERFACE AGENTS They'll make computers easier to use by learning your work habits and finding information you need

BELIEVABLE AGENTS With human facial expressions, they'll make computers less intimidating and easier to control

Before it's here, it's on the Bloomberg Terminal. LEARN MORE