Slashdot Mirror
The UN Will Consider Banning Killer Robots (hrw.org)
Friday the United Nations agreed to discuss a ban on "killer robots" in 2017. The 123 signatories to a long-standing conventional weapons pact "agreed to formalize their efforts next year to deal with the challenges raised by weapons systems that would select and attack targets without meaningful human control," according to Human Rights Watch.
"The governments meeting in Geneva took an important step toward stemming the development of killer robots, but there is no time to lose," said Steve Goose, arms director of Human Rights Watch, a co-founder of the Campaign to Stop Killer Robots. "Once these weapons exist, there will be no stopping them. The time to act on a pre-emptive ban is now."
schwit1 reminded us that IEEE Spectrum ran a guest post Thursday by AI professor Toby Walsh, who addressed the U.N. again this week. "If we don't get a ban in place, there will be an arms race. And the end point of this race will look much like the dystopian future painted by Hollywood movies like The Terminator."
schwit1 reminded us that IEEE Spectrum ran a guest post Thursday by AI professor Toby Walsh, who addressed the U.N. again this week. "If we don't get a ban in place, there will be an arms race. And the end point of this race will look much like the dystopian future painted by Hollywood movies like The Terminator."
Not just that - this is a macrocosm of 2nd amendment arguments in the US. Just like people who advocate banning assault rifles would only affect law abiding citizens, similarly, such a ban would affect law abiding nations, but do nothing about rogue nations who sooner or later would have that capability
Besides, I disagree w/ this proposal for a simple reason. We should avoid intervening in other countries, such as Syria. But if we have to go in, I'd rather send in killer robots after ISIS rather than American (or any other) humans who'll get killed or maimed for life. We should have killer robots substitute soldiers: it would also solve the issue of a depleted military as well as the idea for a draft. If we transform our soldiers into simulation controllers where each soldier controls up to say, 10-16 robots, we then have an army that can theoretically take on every one of our adversaries should we need to fight them - China, North Korea, Iran, Syria, Pakistan, Saudi Arabia, the list goes on...
The other thing about killer robots: it's automation in the military. In the civilian world, we've seen a whole lot of harmless but grunt jobs get automated - from store clerks to robocalls. In the military world, this is a function ideal for automation: it not only saves the lives and limbs of our soldiers, but also extends their efficiency. They get to learn more deeply about robotics and computing principles, and come out of it body intact, no PTSD and merge seamlessly into the civilian work force. In fact, the military is where we should have as much automation as possible, not less.
Like many of the proclamations from the UN, such a ban will have little influence over the development and use of "killer robots".
Some arms control treaties have been relatively successful; but I wouldn't be optimistic here:
Somethings are just too convenient to ban; and those are usually a lost cause; but among stuff you can get support for; there seems to be a difference between hardware where you can say 'no legitimate use, period' and hardware where certain uses are forbidden; but there are enough legitimate use cases that the relevant hardware remains in stock, widely available to people relatively low on the food chain, and easily amenable to quiet 'off label' use.
Military small arms ammunition, say, tends to be pretty reliably jacketed, even disreputable outfits don't tend to produce their own dum-dums and hollowpoints; though irregular forces and police-derived paramilitaries are obviously more likely to be using weapons not concerned with Geneva convention compliance in the first place.
Stuff with both 'legitimate' and 'illegitimate' uses has been harder to keep a lid on. Phosphorus is a lovely illuminating agent; and produces great smoke; but it's considered poor taste to use it as an incendiary. Hard to make that stick when a large number of people, relatively low on the food chain, have access to it because of its legitimate applications, though.
In the case of 'killer robots'; the obvious problem is that all the hardware; and much of the software for a 'killer robot' will be identical to that of a 'human directed' robot with some automation of routine navigation stuff; machine-vision-assisted targeting and IFF, and so on. So long as you pinkie-swear that a human will have to press the 'make it so' button; you can develop all the elements, navigation, targeting, etc. that an autonomous killbot would need; but make sure that the firmware running on anything pesky journalists get to see has a human pressing the red button to approve what the autonomous killbot comes up with.
"Good faith" adherents to a 'no killbots' rule will likely find themselves easing their way toward autonomy with some (admittedly plausible) special cases: "We can't keep a human in the loop for our CIWS/RAM system; human reflexes aren't fast enough for contemporary missile intercept; but don't worry, the CIWS turrets are bolted onto the ship and aren't going to go wandering off." Not false; but an autonomous killbot. Then we'll need an 'emergency defensive protocol' for human-oversight robots that lose their link to HQ, whether to technical failure or hostile jamming; which will be OK; because it's strictly for the robots to engage in defensive actions in their existing location until communication is restored!
People who don't give a damn, of course, will just stub out the 'ask human for confirmation' part and carry on with their day.
Aside from this, some lucky logic-chopper is going to have the unenviable task of explaining why existing, more or less universally accepted, 'fire-and-forget' missiles and other similar hardware that gets its activation command from a human; but thereafter guides itself to target without external intervention, isn't a killbot; but the more drone-shaped hardware that gets its initial activation command from a human; but thereafter guides itself to target without external intervention, is a killbot.
People claiming that diplomatic pressure and arms control conventions are totally useless are seriously exaggerating their case(land mines, chemical, and biological weapons certainly are way down; people have been jumpy about blinding laser weapons, etc.); but there is a lot less room for optimism when you can't draw a bright line around whatever you are trying to ban and declare it and everything similar Off Limits.
With Killbots, I'm not optimistic: too much of the tech is shared with 100% legit 'human approved, machine assisted' systems; and the excuses to get a foot in the door(even if you care enough about perception of legality to not simply quietly switch off human approval) are too plausible.