FUTURESPEAK

American Demographics, April 1, 2003

AD: How will new software agents be different from those available today? For instance, Amazon.com maintains a list of shoppers' preferences based on past purchases. How will new software agents improve on this technology?

JB: The software agents will have more experience, greater cognitive power or ability to infer and learn from past experience, and greater social competence. Current technology is information-impoverished. If you want the agents that assist you to be smarter, they need to have more experience with you than just within the limited scope of a single consumer Web site or appliance, and thus the "same" agent needs to accompany you wherever you go. The agent would also have to be more capable of observing and understanding what you do than simple agents, which merely observe as you click buttons.

AD: What makes agents more intuitive?

JB: If we ignore supernatural inspiration, intuition is based on two things: experience and intelligence. The more experience I have with you, the more likely I am to encounter repetition of activities and situations that help me learn about you. The smarter I am, the more I can abstract from those experiences to find connections and patterns among them - and then infer from past experience in these abstracted situations how best to be helpful in a new, similar situation.

AD: For now, space and military needs seem to drive software agents' development. How might agents help the military or address our business needs?

JB: Just as glasses enhance vision, agents could help soldiers augment their cognitive capacity and thereby enhance awareness. It might help me see things I otherwise couldn't see, because it's infrared or too far away. When I look left, agents might look to the right for me. Because agents can move around and actively seek out information without waiting for a human to specifically program them, you could imagine b-to-b applications. Relatively simple agents currently track inventories and initiate purchase orders on their own, to keep the supply chain running. Similar principles could work in a factory setting.

AD: Down the road, when it comes to simplifying consumers' lives, how ubiquitous will software agents become?

JB: It's hard to say. But they will be deeply embedded in our lives, helping us get things done or gathering information. You can imagine agents that assist in broader ways to do what you normally do on the Web. In the future, you may want something that searches all over the Web and knows your preferences for all kinds of things. Instead of waiting for you to go to those sites, it might constantly monitor items you want and purchase them on your behalf. Obviously, that will require a lot of trust on your part. It's going to be across all spectrums, not just specialized channels for movies, books or clothing purchases. It would be something that can carry your identity and, like a real estate agent, generally act on your behalf.

AD: How soon could software agents populate our world?

JB: In 5 to 10 years. But it depends partially on the health of our economy. A lot of agent companies had problems when the stock market tanked. If the economy recovers and people start thinking about what's next rather than just preserving what's now, I think that we're still 10 years away from agents that have the ability to move between multiple sites and do sophisticated reasoning in a general-purpose way.

JEFFREY BRADSHAW

Imagine a softball-size flying robot that could help astronauts with mundane tasks, such as patrolling the hallways of spaceships and monitoring everything from experiments to atmospheric readings. Even if a fire broke out in a cargo bay, this "personal satellite assistant" would be smart enough to take action without waiting for instructions from a human.

The robot would have greater cognitive and social powers than today's robots, says Jeffrey Bradshaw, research scientist at the University of West Florida's Institute for Human and Machine Cognition, who is developing just such an assistant, with NASA's backing. It would have the autonomy to act on a human's behalf, and it would learn from experience and collaborate with others. It would also pick up on social cues. For example, it might notice when somebody's busy and wait a few minutes before interrupting.

What controls these robots are called "software agents," pieces of computer code that inhabit hardware systems, like computers, or software worlds, like cyberspace. While early agents have been used in autonomous vehicles, satellites and probes, they'll likely appear next in robotic systems that interact with people and in advanced military and Internet applications. Delegating tasks to agents, says Bradshaw, will help us deal with information overload. American Demographics' Sandra Yin spoke with him recently to learn more about what software agents might do for us.