Armed Robots Make History

For the first time in the history of warfare, robots with guns have been deployed. [Tip of the hat to KurzweilAI.net]

One of the topics I covered in my presentation on "Nanotechnology and the Future of Warfare" at the recent World Future Society conference was, you guessed it, robotics. I mentioned that it was only a matter of time until unmanned automated roving machines -- robots -- would be armed with weapons and deployed in battle. I knew they were in development but I did not realize how quickly my prediction would come true.

Here is the story from Wired's "Danger Room" blog:

Robots have been roaming the streets of Iraq since shortly after the war began. Now, for the first time -- the first time in any warzone -- the machines are carrying guns.

After years of development, three "special weapons observation remote reconnaissance direct action system" (SWORDS) robots have deployed to Iraq, armed with M249 machine guns. The 'bots "haven't fired their weapons yet," Michael Zecca, the SWORDS program manager, tells DANGER ROOM. "But that'll be happening soon."

Note that these robots are not fully autonomous: they are "remotely controlled by a soldier through a terminal."

It seems only a matter of time, though, and perhaps not much time at that, until the 'problems' of comparatively slow human decision-making will lead to the next step -- turning over the fire command to an AI program.

How soon might that be? And what sort of ethical debates will then be triggered?

Stick around, because we're likely to learn the answers before long.

Mike Treder

Tags: nanotechnology nanotech nano science technology ethics weblog blog

Comments

[John B]

I dunno, Mike - that seems to me to be rather problematic. Just getting machine vision to the point of being able to recognize valid targets in a visually complex environment is going to be non-trivial, IMO.

Sure, there's been great improvements in machine vision - but there's a long ways to go yet, at least as far as what I've read.

And that doesn't begin to take into account secondary effect consideration - what's 'down-range' of the target, for instance. *shrug*

Am I being optimistic here for once? I dunno. . . *wry grin*

-John

[Dan S]

>>Just getting machine vision to the point of being able to recognize valid targets in a visually complex environment is going to be non-trivial<<

I belive it is already there. Ability to recognize military targets with 80%-90% accuracy exists from early 1980s. Note that machine is not limited to human visual spectra - it can use infrared data and other sensors to eliminate ambiguity.

Also recent developments match humans in rapid image recognition tasks

There are of course other problems with military robot AI beyond simple identification of targets, but still...

[Tom Craver]

At least initially, if not used with great care, these weapons may be weapons of terror.

Imagine one rolls into your neighborhood for the first time. You're not sure if it's remote controlled or programmed to shoot at anything that moves.

If you want to add to your predictions, the next event to predict will be the hacking of an armed robot - possibly just jamming it, or possibly a more sophisticated cracking and turning it against its owners.

Yes, communications with the robot are almost certainly encrypted. But suppose the robot rolls into an alley, out of direct sight of it's controllers. Suddenly the camera feed goes black (blanket thrown over it from above). Then radio feedback is lost. A few minute later, trailing some cut wires and sporting a duct-taped flourescent green OLPC computer, it comes charging back down the street, firing as it comes, now under the control of the enemy.

Or more simply - the camera feed goes black - and while no command is sent, the US troops hear its gun firing. They hasten forward and find the robot - now minus the blinding blanket - with gun still smoking, beside the home of some innocent civilians - who've all been killed.

[Tom Craver]

Though on reflection, the next significant and predictably *controversial* events will be the first kill via robot, followed by the first killing of an innocent civilian, followed by the first "friendly fire" robotic incident.

[Brian Wang]

There have been fatal industrial accidents before
http://www.cdc.gov/mmwr/preview/mmwrhtml/00000505.htm

Then there are all of the autonomous killing (accidental and not) from mines, claymores, booby traps etc...

There is also a certain amount of autonomy from the old gravity dropped bombs.

There is also the separation in time between the decision to utilize (fire) a weapon and when it actually kills.

I would say that AI/robots would be a step up in selectivity from minefields and booby traps. Plus humans control does not mean no-friendly fire or mistakes..see Tilman and other examples

[Tom Craver]

Brian - Of course! But this is...

KILLER ROBOTS!!!!!!!!!!!!

[NanoEnthusiast]

Throw a blanket over it; problem solved.

[John B]

Tom/Dan - Thanks for the corrections. Looks like I've some reading ahead of me. *wry grin*

-John B