Original URL: http://www.theregister.co.uk/2008/02/12/darpa_ai_air_traffic_kill_robot_flying_car_fun/

DARPA continues military AI air-traffic project

Handy for flying cars, machine genocide against fleshies

By Lewis Page

Posted in Science, 12th February 2008 12:12 GMT

DARPA, the US warboffin bureau which curls a supercilious lip at established wisdom and tweaks the nose of consensus opinion, has decided to press ahead with its automated artificial-intelligence air traffic control system.

The move is doubly welcome here at Vulture Central: not only is it a tasty tech-news nugget, but the lack of any ultrapuissant automated air-traffic systems is one of the factors delaying our long-awaited flying cars.

Flight International reported yesterday that DARPA has approved the move forward to phase two of GILA, the Generalized Integrated Learning Architecture. GILA, under development by US wartech behemoth Lockheed and academic partners, is intended initially as a planning and training aid for US military air-traffic controllers. In time, it is hoped that the system will be able to learn from its human colleagues and eventually surpass them.

"Eventually [GILA] will outperform the novice human planner by 125 per cent," Lockheed says.

The company believes that GILA will become a true expert system, able to build up knowledge in a continual process - unlike human traffic controllers, who inevitably leave the service or retire. Under phase two, following a successful progress test by DARPA, the programme gets another year of life and a further $5.2m of funds. Phase one was four years long and cost the American taxpayers $22m.

"GILA will revolutionize the way expert skill and knowledge are captured and transferred," according to Lockheed GILA bigwig Ken Whitebread. "Using planning, learning and reasoning technologies, GILA will know what to learn, why it's important to learn it, and how to focus resources to quickly achieve that learning."

DARPA hope that GILA will allow deployed US forces to manage crowded airspace above future battlefields and war zones more efficiently, allowing manned and unmanned aircraft, drones, weapons and so forth to cooperate without the crippling planning and authorisation delays common today.

This sort of automated routing and planning system - combined, perhaps, with faster-updating and more accurate air tracking - might one day permit much larger numbers of aircraft to operate safely in a given volume of civil airspace. Without something of this sort, it doesn't really matter if you can make a miraculous small, cheap, silent hovering aircraft with synthetic-vision autopilot. Or a decent jetpack, come to that. People would still never be able to fly it in any numbers because this would overload the air-traffic system.

It's early days yet for GILA, though; and most DARPA ideas never work out. Also, just because the military are willing to use a technique doesn't mean that civil authorities will ever welcome it. And, of course, the miracle hovercar isn't even close to existing either.

On another note, it seems almost superfluous to point out that military AI software controlling fleets of murderous aerial war robots - for that is what GILA is meant to become - could easily have been more appropriately named. "Skynet" leaps to mind, for instance; but perhaps there was some kind of legal problem. (The British military has felt no such qualms, however.)

The Flight report is here. ®