Defense Technology International - March 2008 - (Page 12)
SCIENCE WATCH MICHAEL DUMIAK WISH FULFILLMENT All a sailor should do to ﬁnd career happiness in the U.S. Navy, Arpad Kelemen muses, is punch a few instructions into a mobile phone. An intelligent data bot will do the rest. The right sailor gets matched to the right job. Morale improves. Money is saved. What makes Kelemen’s vision possible are advances in computer engithe 35-year-old Hungarian bioinformatics expert says. Kelemen worked with project leader Stan Franklin, author of “Artiﬁcial Minds,” on a team that developed 400,000 lines of Java code designed to replicate the work of a Navy detailer—the guy who tells you where your next job is. Collectively, the code makes Miami Dolphins cheerleaders join a ﬁre drill on board the aircraft carrier Roosevelt. The U.S. Navy is funding research to better match sailors with jobs they really want—perhaps like this one. neering, employing real-world data input, statistics, theories of psychology and social studies, that mimic how and why the brain makes decisions. Software bots, agents and models are being developed at a rapid pace in armed services around the world for a number of reasons. In the U.S., it’s to better train medics; in Canada, to exploit technology more effectively on warships; in Holland, to simulate riot situations for peacekeepers. Much work remains. But Kelemen, a computer science and biostatistics associate professor in the University of Buffalo’s Neuroimaging Analysis Center, made his mark with a team working on an everyday concern—how to ﬁt sailors into jobs they like. Because the elements involved here are more deﬁned, the researchers were able to make strides. “We can actually train software that would behave comparably to a human for this purpose,” 12 up modules known as IDA (Intelligent Distribution Agent). Fifteen large-scale IDA modules, Kelemen explains, reﬂect learning, emotion and deliberation. “We created a model that uses memory much like humans do, in terms of being able to create new associations between things, and to forget.” On a module Kelemen worked on, he had to set values for “hard” restraints, such as Navy policies on alternate shore-to-sea rotations, and “soft” constraints, like a sailor’s preference for location. The model tuned its algorithms through repetition, recalling satisfactory results. The code employs neural-network architecture, Kelemen says, to ﬁnd hidden relations between inputs and outputs. Neural networks are a programming architecture that arrange data and processing elements to mimic neurons. Decision-making modules, or nodes, are interconnected like neu- rons, and the connections are usually weighted by importance. Arranging the architecture this way reveals hidden links among data and moves beyond binary computing. This may seem just a painstaking construct of order-wielding authority. But there’s more to it. “We have also learned which policies or particular constraints are considered to be more important,” Kelemen says. “We wanted to include such things as emotion, learning, human language understanding and generation, even consciousness. We have been able to replicate consciousness to a degree.” The consciousness of a detailer does not guarantee a happy sailor. So the team created another agent, one that would reside with the sailor. “It acts 24/7, looking for jobs, searching databases and web sites, able to write and read e-mail. It looks for opportunities for you,” Kelemen says. “Just evaluate the results and send feedback.” The Navy also has an agent. The interaction would create a valuable data bank. “It eliminates the detailer’s workload, increases speed and efﬁciency, increases sailor happiness, and saves money.” Not quite yet, though. Franklin says the IDA work is in limbo, as it needs to go to a developer to move ahead. But the Navy clearly values the research. Late last year, the Navy put out a call for similar work, recently awarded to longtime intelligent systems partner Paciﬁc Science & Engineering Group Inc., of San Diego. PS&E coordinated Navy research into work modeling decision-making under stress, a project with roots in the 1988 shoot-down of an Iranian airliner by the USS Vincennes, which killed 290 civilians. Although models can be married with graphics to create a virtual environment, by and large they are different animals than computer game sims. It’s not about technology, it’s about brain and biology: the idea is to build open-ended engines that arrive at decisions intuitively. Ideally, researchers and coders introduce elements of chaos and the model reacts to those naturally, realistically and in the moment. The massive, predictive models are still in the future. But in the meantime, the notion of military job satisfaction might become less of a contradiction. I AviationWeek.com/dti U.S. NAVY DEFENSE TECHNOLOGY INTERNATIONAL MARCH 2008
If you would like to try to load the digital publication without using Flash Player detection, please click here.