5 ms·

We Can Now Build Autonomous Killing Machines. And That's a Very, Very Bad Idea

17 points by Libertatea 12y ago

sukilot 12y ago
Can't stop it. Any autonomous toy plus some poison, explosive, or sharp warhead == weapon.
- pj_mukh 12y ago
  I think this is referring to target identification systems that will autonomously pull the trigger without a human in the loop. Not a thoughtless landmine or blanket explosive (we know those are bad)
pj_mukh 12y ago
Or in other words https://xkcd.com/652/ https://xkcd.com/652/
- X-combinator 12y ago
  So true... http://imgs.xkcd.com/comics/more_accurate.png http://imgs.xkcd.com/comics/more_accurate.png
mullen 12y ago
Maybe I don't see the problem, but I don't see the problem with robots hunting down and killing humans. As long as the robots are controlled by the side I am on, I am good with this. Yes, there are some technical issues to overcome that guarantee we keep them under our control, but those issues will be overcome before deployment. I wouldn't mind having a future where someone attacks my country or needs to be wiped out (ISIS, Al Qaeda in "Some cesspool of a country") and we just unleash the robots on them. Set some parameters and tell the robots where they can and can not operate and under which circumstances and be done with it. Some place like Syria would be a no brainer. Drop a 100,000 robots in there and that's the end of ISIS, Al Qaeda and Assad. Let the AI figure out who the bad actors are and then help rebuild the country when done getting rid of said bad actors. In the big picture, I don't see any downsides with this.
- VikingCoder 12y ago
  > those issues will be overcome before deployment. Really? We'll invent a 100% secure computer system? With no way to be hijacked? How about we build that system first, and use it to protect our social security numbers, credit cards, email, voting, and patient health information! It's because there is no way to make a perfectly secure system. In a worse-case scenario, our enemy hijacks those 100,000 robots you wanted to send, all at once, and sends them against good guys. > Set some parameters and tell the robots where they can and can not operate and under which circumstances and be done with it. Some parameters... Like, holding a firearm? Mad at the United States' policies? Defending their home against invasion? Bowing to Mecca? Skin pigmentation? Remember, war isn't even fought between nations any more. Most of our recent interesting enemies have been merely citizens inside of third world countries. If your GOAL was "kill everyone inside of these borders," then maybe you could control that. That's never the goal, though. Not since Hiroshima and Nagasaki. > Let the AI figure out who the bad actors are Humans can't even do that. > are and then help rebuild the country That's a neat idea... How about we win hearts and minds with agriculture / health / construction / education robots FIRST, rather than sending an army of death robots to murder every human in a death zone?
  - yellowapple 12y ago
    > Really? We'll invent a 100% secure computer system? With no way to be hijacked? That's obviously not actually a requirement of killing machines if humans - being as phenomenally easy to deceive as they are - are acceptable options for assuming that role. The system wouldn't have to be 100% secure in order to be good enough, by the standards we've already set as a species. It would only need to be more secure than a human.
    - landryraccoon 12y ago
      It's not at all clear that a human being is easier to deceive under battlefield conditions than a machine would be.
VikingCoder 12y ago
I recommend the techno-thriller fiction novel "Kill Decision" by Daniel Suarez.
yellowapple 12y ago
I honestly don't think this is a bad thing. Or, at least, it's no worse than humans doing the killing. The article complains that such machines will never be able to make correct judgements about their targets with 100% reliability. However, neither can humans. Furthermore, I'd reckon that an artificial intelligence - with the hypothetical ability to tap into exabytes of raw data to base its decisions upon - is far better suited to those sorts of decisions. The article incorrectly assumes that there is a distinction between the brain and any other ordinary computer. Once one realizes the sheer lack of such a distinction between organic and inorganic computing devices beyond rather-minor implementation details and computing capacity, the whole premise of the article is moot. We already are using autonomous killing machines, and have been for hundreds of thousands of years. Those machines are called "humans".

17 points by Libertatea 12y ago

sukilot 12y ago

Can't stop it. Any autonomous toy plus some poison, explosive, or sharp warhead == weapon.

pj_mukh 12y ago
I think this is referring to target identification systems that will autonomously pull the trigger without a human in the loop. Not a thoughtless landmine or blanket explosive (we know those are bad)

pj_mukh 12y ago

Or in other words https://xkcd.com/652/ https://xkcd.com/652/

X-combinator 12y ago
So true... http://imgs.xkcd.com/comics/more_accurate.png http://imgs.xkcd.com/comics/more_accurate.png

mullen 12y ago

Maybe I don't see the problem, but I don't see the problem with robots hunting down and killing humans. As long as the robots are controlled by the side I am on, I am good with this. Yes, there are some technical issues to overcome that guarantee we keep them under our control, but those issues will be overcome before deployment. I wouldn't mind having a future where someone attacks my country or needs to be wiped out (ISIS, Al Qaeda in "Some cesspool of a country") and we just unleash the robots on them. Set some parameters and tell the robots where they can and can not operate and under which circumstances and be done with it. Some place like Syria would be a no brainer. Drop a 100,000 robots in there and that's the end of ISIS, Al Qaeda and Assad. Let the AI figure out who the bad actors are and then help rebuild the country when done getting rid of said bad actors. In the big picture, I don't see any downsides with this.

VikingCoder 12y ago
> those issues will be overcome before deployment. Really? We'll invent a 100% secure computer system? With no way to be hijacked? How about we build that system first, and use it to protect our social security numbers, credit cards, email, voting, and patient health information! It's because there is no way to make a perfectly secure system. In a worse-case scenario, our enemy hijacks those 100,000 robots you wanted to send, all at once, and sends them against good guys. > Set some parameters and tell the robots where they can and can not operate and under which circumstances and be done with it. Some parameters... Like, holding a firearm? Mad at the United States' policies? Defending their home against invasion? Bowing to Mecca? Skin pigmentation? Remember, war isn't even fought between nations any more. Most of our recent interesting enemies have been merely citizens inside of third world countries. If your GOAL was "kill everyone inside of these borders," then maybe you could control that. That's never the goal, though. Not since Hiroshima and Nagasaki. > Let the AI figure out who the bad actors are Humans can't even do that. > are and then help rebuild the country That's a neat idea... How about we win hearts and minds with agriculture / health / construction / education robots FIRST, rather than sending an army of death robots to murder every human in a death zone?
- yellowapple 12y ago
  > Really? We'll invent a 100% secure computer system? With no way to be hijacked? That's obviously not actually a requirement of killing machines if humans - being as phenomenally easy to deceive as they are - are acceptable options for assuming that role. The system wouldn't have to be 100% secure in order to be good enough, by the standards we've already set as a species. It would only need to be more secure than a human.
  - landryraccoon 12y ago
    It's not at all clear that a human being is easier to deceive under battlefield conditions than a machine would be.

VikingCoder 12y ago

I recommend the techno-thriller fiction novel "Kill Decision" by Daniel Suarez.

yellowapple 12y ago

I honestly don't think this is a bad thing. Or, at least, it's no worse than humans doing the killing. The article complains that such machines will never be able to make correct judgements about their targets with 100% reliability. However, neither can humans. Furthermore, I'd reckon that an artificial intelligence - with the hypothetical ability to tap into exabytes of raw data to base its decisions upon - is far better suited to those sorts of decisions. The article incorrectly assumes that there is a distinction between the brain and any other ordinary computer. Once one realizes the sheer lack of such a distinction between organic and inorganic computing devices beyond rather-minor implementation details and computing capacity, the whole premise of the article is moot. We already are using autonomous killing machines, and have been for hundreds of thousands of years. Those machines are called "humans".