FlamingText

The Future of Warfare Pt III

Still more killer robots.

Start from the Beginning.

The Logic of War:
Hot on the heels of the SWORDS, a host of other robots with combat and support applications is being readied under FCS (future combat systems) to take their place alongside human soldiers in nearly every aspect of military operations. From mobile sentry systems like MDARS and REDCAR to unmanned combat vehicles like the robotic Stryker or the Marines’ Gladiator to integrated, multiple-vehicle systems like COUGAR, the Pentagon is aggressively pursuing its commitment to a largely automated military in the next decade.

  The two applications that are likely to be performed by robots first are supply transport and sentry duties, as these are tasks that require lower levels of true autonomy and for which technology already exists. Military planners envision supply convoys consisting of one manned (and presumably well-armored) lead vehicle accompanied by a train of “follower” trucks who would track the GPS positioning of the lead truck and drive along behind it. Current systems allow vehicles to navigate on their own and only ask for instructions from an operator when they get stuck. Navigation in straight lines down paved roads is simple enough that they wouldn't require much help. Similarly, robots have already been successfully employed as sentries in controlled areas, because patrolling a known route in a fixed location does not require the sorts of complex navigational decision-making that robots have difficulty mastering.

  The ongoing efforts to develop systems that can navigate in unfamiliar surroundings highlight some of the major obstacles that must be overcome before truly autonomous combat robots will be possible. While great progress has been made in creating navigation systems that can negotiate difficult terrain, designers have thus far been unable to overcome certain limitations in the perceptual powers of robotic vehicles. A competition held last year in the Mojave desert which offered a $1 Million dollar prize (sponsored by the Pentagon) to any fully automated vehicle that could successfully navigate the 142-mile course produced no winners. Within four hours, every vehicle entered in the competition had crashed or broken down.

  These are not necessarily limitations that will be overcome by advances in raw number-crunching speed. At the heart of the enterprise are some knotty questions about how to describe the vagaries of perception in the formalized language of machines. Consider this description, pulled from a Washington Post article from last year, concerning the performance of the MDARS sentry, “patrolling” the snow-covered back yard at the Westminster lab where it is being developed:

“It drives several feet, eyes a parking sign and halts, apparently puzzled, until a human attendant reprograms MDARS to move on.
"Compared to a human, MDARS is really not that smart," {General Dynamics Engineer} DiBerardino says by way of explanation. “

  A robot that can’t tell a parking sign from an intruder is probably not something that most military personnel are going to want to arm and send out to patrol the perimeter. Until shortcomings such as these can be satisfactorily addressed it seems likely that decisions to employ lethal force will be left in the hands of remote operators. It is worth noting, however, that almost every one of the land-based robotic systems listed above features some sort of “non-lethal/less lethal” option (to clarify: “non-lethal” and “less-lethal” usually refer to the same system – if somebody gets shot with it and doesn’t subsequently die, it’s “non-lethal”). Sentry robots will almost certainly be entrusted with the decision to employ these sorts of “delay and deny” measures long they are trusted to pull the trigger on an M-16. So the problem remains: How to formally describe, in terms a machine can understand, the difference between friend and foe?

  What’s needed is a robot that can complete one of those courses that police officers use to train for the use of deadly force in the field. If you’ve seen a Dirty Harry movie you’ve probably seen the Hollywood version of these, where life-size cardboard cutouts of bad guys pop out and the shooter has to take them down as rapidly as possible while not shooting “decoy” targets which represent innocent bystanders. The cardboard cutouts may have long since been replaced by video and then by first-person-shooter simulations, but the idea remains the same: Training those entrusted with the use of lethal force to exercise judgment when employing that force. A soldier who simply shoots everything that moves is of limited usefulness, especially in an urban setting where winning the hearts and minds of the locals is part of the mission statement. A robot that rolled around mowing down innocent women and children might well present an even greater PR problem.

  The solution is not necessarily more processing power or better hardware either. Before robots can be trusted with the decision of whether or not to pull the trigger, a hard-and-fast set of rules for distinguishing between targets and innocent bystanders must be developed. In effect, it would require programmers to be able to write code clearly defining the Logic of War. If that phrase sounds like an oxymoron to you, then you are beginning to grasp the scope of the problem facing programmers.

  Consider this: If the machines are given the instruction "don't shoot nuns," then soon there will be battlefields full of soldiers dressed as nuns marching right up to the robots and shooting them point blank in the brain box. Conversely, instructions like "shoot everybody wearing a uniform that isn't one of ours" would lead to widespread slaughter of nurses, priests and bellhops. Instructions to look for weapons would still have to define in exact terms what a weapon looks like in every possible instance. Errors of judgement would prove fatal, and baiting combat robots into killing innocent bystanders or destroying select targets would become a viable guerilla technique. Considering how difficult visual perception in the real world has proven to be for machines, optical recognition systems could probably be rendered inoperative by a good Trompe L'oeil painting.

  In the immediate future, the way around this problem will likely be to not address it at all. Sentry systems are already being discussed that have a “shoot anything that looks suspicious” instruction, which would probably mean that in controlled areas anybody not presenting proper identification (this may take the form of “smart chips” which could be worn, sewn into uniforms, or implanted) would be fair game. Anyone entering the sentry’s area would receive a command to “halt!” or be fired upon, with the robot having clearance to use force if they do not then comply.

   Sentry systems now being readied for deployment will employ a non-lethal/less lethal option. Setting up robots to shoot anything that moves in their field of vision is a relatively simple matter. It's not hard to imagine soldiers in their 14th straight hour of sentry duty looking for ways to get sentry robots to perform these duties for them, even if that meant entrusting them with deadly weapons -- the same weapons, after all, that the soldiers are instructed to use. And while the idea of machines with undiscriminating lethal responses to everything in their sphere may be troubling, in reality such machines have been present on the battlefield since the invention of the landmine. Of course landmines can’t aim or give pursuit. Yet still they cause an estimated 2000 deaths and casualties every month worldwide, mostly in civilians.

Next:The Sky's the Limit

Posted by flamingbanjo at June 8, 2005 01:52 PM