Debating a Ban On Autonomous Weapons (thebulletin.org)

Lasrick writes: A pretty informative debate on banning autonomous weapons has just closed at the Bulletin of the Atomic Scientists. The debate looks at an open letter, published In July, 2015, in which researchers in artificial intelligence and robotics (and endorsed by high-profile individuals such as Stephen Hawking) called for 'a ban on offensive autonomous weapons beyond meaningful human control.' The letter echoes arguments made since 2013 by the Campaign to Stop Killer Robots, which views autonomous weapons as 'a fundamental challenge to the protection of civilians and to international human rights and humanitarian law.'

But support for a ban is not unanimous. Some researchers argue that autonomous weapons would commit fewer battlefield atrocities than human beings—and that their development might even be considered morally imperative. The authors in this debate focus on these questions: Would deployed autonomous weapons promote or detract from civilian safety; and is an outright ban the proper response to development of autonomous weapons?

Comments

Autonomous robots, guns and duct tape

[Kjella]

What's the difference between a search-and-rescue bot and a kill bot? The function is going to pretty much identical right up to the point the target is located, just duct tape a gun to point in same direction as the camera and wire the "person located" signal to pull the trigger. It's one thing to ban ABC weapons because they're very specific technologies, but this is way too generic to work. And it's not like the military is going to avoid developing it for intelligence gathering and decision support systems, even if you keep a human in the loop it's literally going to be one flip of the switch to full automatic where the computer's recommendations are implemented by itself.

The primary reason to keep soldiers in the loop today is because you're trying to fight a "good war" and avoid antagonizing the civilians so you want manual confirmation of each target, if you take the gloves off and say if you're found outside after curfew we'll shoot to kill and live with the collateral you could automate much more. And don't get up on the high horse, when the US nuked Hiroshima and Nagasaki they knew there's be about 100-200k civilian casualties. In a real war nobody's going to give a fuck if the robots are just 99% or 95% right, if it can save our troops and civilians and end the war for sure we're going to let them fight for us.

Self Defense

[im_thatoneguy]

Arguably while war is all about winning, it's not at "all cost". We (ignoring GOP presidential candidates) will not bomb an urban civilian center these days in order to kill a few hostile bad actors. Arguably the availability of precision munitions changed the moral balance from "bomb a city into submission" into the modern sensibility of "only bomb specific targets of military importance."

Even though war is a terrible and bloody affair, we as societies have constantly been moving towards more humane and less deadly conflict. It's one thing to shoot someone shooting at you. It's quite another to kill someone in cold blood. War is arguably largely about self defense today: "I have to shoot you in order to not die." The shift I see happening with autonomous weapons is that there is no imperative to shoot people shooting at you. It might be expensive or costly to lose an autonomous infantryman, but if you can capture without killing I suspect we'll expect our autonomous soldiers to exert "self control". Obviously if it's a shooter killing civilians you would be morally justified in stopping them using violent force if necessary but otherwise autonomous troops effectively become more akin to police officers than soldiers in their relationship to the population.

Uh huh...

[Pseudonym]

Some researchers argue that autonomous weapons would commit fewer battlefield atrocities than human beings [...]

No, it's just that if an autonomous weapon does it, it would be more difficult to call it an "atrocity". If a dozen villagers are killed because of a minefield that some idiot decided should go near where they live, the only reason you can't call that a "massacre" is that there was no human making the targeting decision.

In the 1920s, there were some who argued that aerial bombing would be more humane because they could be far more precise than field artillery, hitting only the target that you want to hit. Look how well that worked out.

Good luck enforcing it:

[Hartree]

Many restricted military technologies are fairly easy to detect. Nuclear weapons require a massive industrial input that has well known signatures, for example.

A robot is different. It can be something that's dual use. One day it's a regular robot. Look at the software. All civilian and a nice strong optics mount point on it.

Change the software and swap some of the camera gear for a small machine gun and use the rest for aiming, it's a killbot. This is just one example. Another obvious one is putting an autonomous drone software package into the flight computers of an airplane that can also be manned. This game can go on and on with just about any weapons system you can think of.

It doesn't take industrial facilities that are different from usual ones to make them. If you can make versatile robots for civilian use, and separately make weapons you just have to put them together at the last minute. They don't have any particular signatures the way chemical weapons and their precursors do. Most nations are already making ordinance, so who's to say whether a human is going to be in the loop to fire it or if it's triggered by an AI?

If people want to cheat on this, it'll be pretty easy to do so.

So far, the landmine bans haven't seemed to have slowed down the planting of them a bit in various wars. We have to have demining teams, not just for cleaning up old wars, but the very ones that are going on now.

I don't expect this to have a much greater effect.

Autonomous robots do MORE attrocities

[gurps_npc]

Look, do you know how hard it is o convince a solider to commit an atrocity? They don't just start doing it. It takes years and years of training and exposure to the horrors of war before soldiers are willing to do those things.

But robots could be turned on and ordered to do atrocities that no human would agree to do.

If a rogue general wanted to convince US soldiers to kill everyone in the White House, or to kill everyone in Congress, he would find it very difficult. It would take almost nothing for a rogue General to convince a squadron of drones to do the exact same thing.