We started hearing about software agents a decade ago. Early visions suggested that every PC user would have a software agent on his desktop, carrying out simple tasks such as checking on monthly sales reports and prompting us about upcoming meetings. One example of this type of personal digital assistant, Microsoft's Bob, never caught on with the computing public.

A lot has happened since Bob went bye-bye.

Researchers are developing autonomous software agents that interact with other agents on behalf of businesses. These agents aren't carrying out simple tasks. Quite the contrary. They're carrying out processes that would take humans a great deal of time to complete. Autonomous software agents, in fact, excel at managing complexity and large amounts of information.

Autonomous agents differ from other types of software in their ability to execute in complex domains under a wide variety of situations, according to AT&T Labs senior technical staff member Peter Stone. While most application software executes in response to external prompts, autonomous agents respond to a wide variety of conditions through sophisticated sensing mechanisms.

Agents are being developed to carry out activities in E-commerce environments that include customer service, product configuration, buying, and selling. Agents are also managing computing resources in network management and security systems. Agents will also play an important role in helping humans to collect and analyze complex information, and will become part of high-end analytics applications.

Many researchers believe autonomous agents will be a part of most business systems in the next five to 10 years. IBM researchers foresee broad use of agent software in E-commerce. "We're creating a new economic species that will be created partially in our image, but significantly different from humans," says Jeff Kephart, manager of the agents and emergent phenomenon group at IBM Research. "Software agents will become players in the economy," he adds.

IBM Research has created software agents that work in online auctions, bidding on goods and services, responding to the bids of their competitors, and even developing bidding strategies with other agents to outsmart the competition.

In a demonstration earlier this year, IBM's agents actually outperformed humans. Humans were prone to make typing mistakes, thus throwing off their bids in online auctions. Granted, the agents weren't without their own problems; they started to develop aberrant behaviors, bidding too low for certain goods.

Agents are good at taking into account a lot of data in a short time period of time. Agent software could be used to set dynamic prices, to collect market data, to sense emerging market conditions, to engage in data mining across a variety of databases, and to collaborate with other agents.

This type of autonomous software could also be used to assist businesses in:

• carrying out complex negotiations between trading partners,
• searching for and helping to configure complex products,
• locating business partners across the Web,
• managing the many elements of a complex project,
• tracking a multitude of business conditions and trigging an alert that would result in human intervention.

Kephart says agents will make markets more efficient by maximizing profits, proactively carrying out corporate goals, and minimizing time lags between the initiation of a process and its completion.

Autonomous agents will also be used to look for patterns in systems. A U.K. company called SearchSpace, for instance, has developed software sentinels that create profiles of cell phone customers based on their usage patterns. When someone uses a particular cell phone account, her call is compared to the pattern of calls that subscriber usually makes. If the call seems out of the ordinary, the user is prompted to enter a personal identification number before the call is put through. If the user can't enter the PIN, chances are the phone has been stolen, and the phone company can shut down the account.

Intelligent agents are also starting to be used in intrusion-detection systems--they move through a network and look for unauthorized users. Agents could also be used to look for changes in the way workers access software. A marked change could signal the person is attempting to do damage to the corporate system.

Bell Labs scientists are looking at the use of intelligent agents to manage the telecommunications network. These autonomous agents can perform in proactive and reactive mode. Bell Labs' computer scientists are working to create a common language for agents to communicate with each other. The Lucina project, named after the goddess of childbirth, is an open platform that stipulates how agents communicate, defines what constitutes a message between two or more agents, and the rules governing the domains in which agents operate.

There's a tremendous amount of research taking place in corporate and academic research labs across the country in the area of autonomous agents. This type of software seems to take advantage of what computers are best at: processing lots of data in short periods of time. It's unlikely agents will replace human thought and reasoning, but it will go a long way to making businesses and their systems more efficient.

Sounds good to me. Why deal with a human being when you can deal with a machine? I bet some day we'll have a group for the ethical treatment of autonomous software agents.

Are you looking forward to autonomous agents? Using them already? Or are you thinking that, just maybe, this isn't such a good idea? Chime in at Karyl Scott's discussion forum.