Intelligent Agents

Intelligent agents and multi-agent systems are hot research topics in the software community, and agent oriented software engineering (AOSE) shows promise as an key enabler for machine intelligence and robotic autonomy. This article briefly discusses what distinguishes AOSE from alternative approaches and how agent-oriented paradigms compare to object-oriented analysis/object-oriented design (OOA/OOD) architectural abstraction.

Object-oriented and agent-oriented methods are closely related. To clarify their relationship, consider the levels of abstraction that make up programming paradigms in general and realize that object-orientation and agent-orientation are instantiations of that general case with advantages realized by a priori patterning. By embedding artificial intelligence into the agent entities, the agent-oriented paradigm makes a major contribution to software engineering that will change the way we think about software systems.

Emerging from the world of Artificial Intelligence, early efforts to develop intelligent agents focused upon an integrated life-like software entity that considered all complex options (much as a human might) in an attempt to handle ad hoc situations. Nowadays, implementers focus upon distributed agent subsystems that tackle specialized discrete tasks for more tractable responses.

From “ Mobile Intelligent Agents” John Geirland, in “The Feature”, Dec 11 2002:

Imagine you're driving from Washington D.C. to New York City to see a friend. The personal agent living in your car's telematics system sends an email to your friend's personal agent suggesting dinner. More agents are “spawned” that zero in on your shared dining preferences (sushi) and access Mapquest or some other location-based service to find a conveniently located restaurant. Still other agents oversee the whole process.

Of course, some fear agents could act like pushy concierges. What if an overzealous agent compares a user's calendar with their location (via GPS) and makes the snap judgment that the user won't make it to the opera on time? “I sold your tickets,” the agent notifies the hapless users. “But look at the fine restaurant I booked for you with the money I saved on your tickets!” “A lot of these deployments will require an opt-in feature,” notes Jonathan Prial, vice president of business development and sales for IBM's pervasive computing division.

University laboratories, government agencies, and industry engineers/scientists are evolving new generations of wireless intelligent agents. Carnegie Mellon University’s Intelligent Software Agents Group (http://www-2.cs.cmu.edu/~softagents/) developed an agent development toolkit called RETSINA. CMU professor Katia Sycara and her team are building agents for human-machine collaboration, to prevent auto accidents and your social life securely on track. She advised the CMU and University of Pittsburgh Raptor team for the RoboCup Rescue US Open 2005, which placed highly in all competitions. The research validated concepts of coordinating multi-agent heterogeneous robot teams with software agents and disaster response personnel.

University of Southern California Intelligent Agent and Multi-agent Systems research center (http://agents.usc.edu/) investigates intelligent agents that manage distributed information systems, intelligent agents that emulate human behavior, and robotic agents that collaboratively use mobile robots and distributed intelligent sensors to perform complex tasks. Among the practical creations of this agent technology are virtual reality simulation trainers for young military officers (demonstrated during TEAMS Week in Huntsville this year), placing them in simulated crises with emotional interactions, training and equipping them to handle immediate tactical decisions typically encountered in urban battle scenarios.