Wouldn't you like end-to-end control of the entire logistics pipeline? Simultaneous planning and execution? A logistics control system that operated at all levels and during all phases of operations? A system that allowed multiple simultaneous communications and decision-making over a vastly complex, spread out theater of operations?
Well, of course you would, although it probably sounds as far-fetched as having psychic access to your customers' order planning, or complete control over the weather.
It's not. The U.S. military and the defense departments in other countries, including the UK and Australia, are shooting for these exact goals, and they're coming close to achieving them. It's all made possible by something called intelligent software agents (see box).
Ironically, intelligent agents were originally developed in the commercial sector, and advocates initially had high hopes for an early rollout there. Last year, we published a story about intelligent software, quoting several sources who promised significant pilot projects in the next year. That proved overly optimistic, and a year on, adoption in the commercial sector is still far from reality. In the interim, however, intelligent software agent technology has steadily found its way into mainstream military applications.
R & Defense
It may be that the military has pulled ahead simply because it needs this level of sophistication more than commercial shippers do. "People ask, why aren't you like Wal-Mart," says Mark Greaves, program manager at the U.S. Department of Defense's Defense Advanced Research Projects Agency (DARPA) in Arlington, Va. "I say, well we are like Wal-Mart except our stores are moving all the time and Christmas comes on a random date every year."
And so for now, the military's where the action is. "The military has been in the front of this agent stuff," confirms Dr. Noel Greis, director of the Center for Logistics and Digital Strategy at the University of North Carolina's Kenan-Flagler Business School. "In the commercial world, the interest in agents [really took off] during the e-commerce era, and most of the applications focused on the interface between the customer and the process." (Amazon.com, for example, uses intelligent agents to automatically recommend new titles based on the customer's previous purchases.) "The military is a bit different," says Greis. "They dug in immediately on how agents can help with internal processes, especially logistics—the deployment of people, ammunition, fuel, water. In the military, agents are part of this whole digitization of the battlefield and systemization of operations. They want information captured in real time so they can see how the environment is changing dynamically and make sense of the information coming in."
On that front, perhaps the most promising intelligent software system being developed in the United States is UltraLog. Project UltraLog was established by DARPA in 2001 to take over from the Advanced Logistics Project that ran there from 1996 to 2001.
The UltraLog project's stated goal is "to create a comprehensive capability [that] will enable a massive scale, trusted, distributed agent infrastructure for operational logistics to be survivable under the most extreme circumstances." In other words, researchers are seeking the best way to use smart technology for maximum operational efficiency in combat situations—with the ability to continue operating at 80-percent capacity with up to 45-percent information infrastructure loss, for example.
UltraLog's not the only project under way, however. Others include Control of Agent-Based Systems (CoABS, also under DARPA), and Log Net, a joint-venture between the University of North Carolina, software vendor Saffron Technology of Morrisville, N.C., and Boeing Inc., headquartered in Chicago, which is developing technology for the U.S. Marine Corps. The Office of Naval Research announced in August 2003 a $5.74 million contract with the University of Southern California and Vanderbilt University to use agent software developed by computer scientists at the two schools to handle Navy and Marine Corps pilots' schedules. There is also a more ambitious and long-term project called Future Combat Systems, fostered by the U.S. Army.
There are two main advantages to using intelligent software agents to help out with dynamic logistics management— that is, logistics management that responds quickly to changing situations.
First, intelligent agents form a collaborative network and can act independently. They don't need to operate in lockstep with a "master" computer that's overseeing everything; they can make a request, say, from a transportation management system to a local shipper's office without going through the head office's computer banks. This is called a distributed system, and it mimics much more tightly the physical structure of a complex logistics network.
Second, intelligent software agents are able to make autonomous decisions that humans would likely make if only they had the time. This is where the technology intersects with artificial intelligence, although on a relatively small scale. Intelligent agents are programmed with parameters that cause them to make what humans would consider "good" decisions, and they interface with humans on overall plans that spring from those decisions.
"What we're trying to do is tackle something that is a critical need, which is to unload humans from a lot of the mundane tasks they need to do, and to get humans and computers to collaborate," says Dr. Andrew Lucas, managing director at Agent Oriented Software Pty Ltd., a Victoria, Australia-based technology firm. "If you could do that, you could get a lot done," adds Lucas, who's currently working for the U.S. Air Force Research Laboratory at the Wright-Patterson Air Force base in Ohio, applying intelligent agents to military simulation.
And you could get a lot done fast. Enlisting computers to collaborate and make decisions lets users create a top-level logistics plan for a major maneuver in under an hour instead of the weeks it would take using humans alone. The USC/Vanderbilt program researchers reckon intelligent software cuts down the time needed to produce a daily schedule for AV8-B Harrier jump jets from six hours to four minutes.
Furthermore, intelligent software gives users greater ability to predict and limit the loss of operability caused by damage to the logistics system. The DARPA UltraLog program stresses the capability to build clusters of intelligent software agents that adapt so they can function in damaged and stressed environments. The military needs these capabilities more than anyone else because, for them, a logistics hiccup might mean losing an entire port or fleet of aircraft.
Greis says that the Future Combat System will also concentrate on smart ways of figuring out when and how to restock supplies in the field. "What they'd like to do is to make a lot of the logistics processes—re-supply, moving materiel from boat to shore—as autonomous and automated as possible. Right now, someone gets on the horn and says: I need such and such, and they're just responding. What they'd like to do is to develop agent systems that are more anticipatory and also respond autonomously," says Greis.
Greis describes one test project conducted recently, where vehicles in the battlefield were fitted with sensors that relayed information about engine temperatures, water and oil levels and so on. The software agents made their own decisions about how much to re-supply, who would do it, when it would be done and how long it was going to take. "We've developed a system with mobile agents, learning agents, collaborative agents that work together to manage the logistics," says Greis. "We're moving away from the push model to more of a pull model of supply. It's the same concept as the commercial supply chain."
Other military forces are catching on, too. Dr. Lucas is doing simulation work for the Defence Science and Technology Laboratory (DSTL), part of the UK's Ministry of Defence. Lucas is also working with the Defence Science and Technology Organisation (DSTO) in Australia to apply his company's agent product to unmanned air vehicle mission management. The firm is also working with the military in Canada and France.
Beginning of a beautiful friendship
Lucas is confident this smart technology will end up in the commercial sector, but he doesn't expect to see widescale adoption for five years or so. "It's not being used in the commercial sector, but it's close," he says. "As a technologist, I recognize that risk management is a big element in introducing new technologies. It's about showing people you have a plan for proving the technology and reducing the risk to the customer, over a number of years, and we've chosen to do that with the military first."
Part of the problem is that the technology's not quite ready for prime time. "We don't want to minimize the extent to which we think this would be useful in the commercial world," says Tony Rozga, research fellow at LMI Government Consulting, a not-for-profit government consulting firm based in McLean, Va., that supports the UltraLog project. "It's just not there yet." LMI and DARPA have worked together to develop Cougaar, a computer architecture for distributed agents. Notably, it's been made "open source"—that is, non proprietary—to encourage commercial application.
Saffron Technology's co-founder, chairman and chief scientist, Manny Aparicio, says one obstacle to adoption in the commercial sector is the issue of trust, especially with learning agents—ones that gather information about habitual human preferences and responses over time. People are nervous about delegating responsibility to a piece of software, Aparicio says. But, once they learn to trust it, things change. "The user can have it set so it just makes recommendations, but then they might get bored seeing it's right all the time and let it build a whole re-supply strategy," Aparicio says. There's always a safety net, he adds. "Whenever an agent sees a situation it doesn't know, it can still send an exception back."
Meanwhile, Greis's Center for Logistics and Digital Strategy's new "Intelligent Enterprise Initiative" aims to help industry leverage advances in digital technology, especially intelligent agents, to redesign their business processes. Greis says the commercial logistics sector is still wrestling with collecting sufficiently fine-grain data to feed to intelligent agents—something the military is far ahead on. "As you get all this real-time information, the agents are going to become tools for interpreting that," Greis says. "So people are focused on the RFID part of the puzzle [for now], but ultimately the RFID and agents will be brought together."
Intelligent software agents are a form of artificial intelligence —software packets that act as autonomous decision- making entities, capable of coming up with solutions to problems and acting on them automatically. Think of the job performed by a travel agent hired by a business manager who needs to schedule a meeting. The manager contacts the travel agent to deal with flight schedules and make rental car and hotel arrangements. The travel agent then works independently with the manager's assistant—who's planning other details of the meeting, such as time and the list of attendees— to coordinate efforts so that they fit in with personal schedules, location information, the other attendees and even the weather.
To the manager, the end result, ideally, is a simple set of outcomes—flights, hotel booking, a reserved conference room full of people he wants to meet with. But the travel agent and assistant have collaborated in choosing these outcomes from an almost infinite number of possibilities, many of them interrelated. Change one factor and you change others—flying direct rather than to the local airport means renting a car. A single hiccup can send the whole process back to square one—a crucial client becomes unavailable on those dates. And so on. People respond and make decisions as the situation changes.
Software agents do the same job. They are goal-oriented, autonomous, collaborative, adaptive, proactive and mobile. Further, they allow "systems of systems," meaning you can link any number of agents together to get on with highly complex and widely distributed series of tasks. In practical logistics operations, such as a theater of war, that means you can have data and business processes distributed throughout the operation, instead of centralized data warehouses with longreach data feeds. This makes the flow of information adaptable to fast-changing environments, according to DARPA. It's also more robust and more reliable.