AUVSI

[John Canning]

When Asimov penned his original Laws of Robotics in 1940, he set a standard for idealized robotic behavior. Today, militaries around the world are attempting to develop armed unmanned systems. How does this mesh with Asimov's Laws? Please see the attached document.

[Luke Haberkern]

I have read the three Asimov books on robotics, and though they are interesting and revealing on the aspects of AI in Asimov's time, they bear no usage here today. Even if we could apply the three fundamental rules of robotics on all our designs there's no reason we would. A robotic mine detonator would never work if it wouldn't knowingly bring harm to itself. Asimov believed that in our time or the future, artificial intelligence would actually "think". Though such concepts of fuzzy logic and ghost programming are beginning to be developed, the most complicated and intricate robots today merely "compute" - using probabilities, environment knowledge, and the insight of its creator to "act" like it's thinking. Even when fully autonomous, the robot is simply an extension of its creator's abilities. Asimov had foreseen the combatant uses of AI, but wrongly assumed that we would knowingly stunt our creativity and effectiveness with such constraints, or even that we ever could if we wanted to. Robots will be able to manage wars, civilize other planets, and pretend to be human long before they will be able to discern a boy with a gun intent to shoot innocents from one terrified to be told to do so.

[John Canning]

Yes, we realize Asimov's Laws don't really apply. That was the point of the original post. The problem is that most people's perceptions for robots are either framed by Asimov's laws and/or the Terminator movies - neither of which is an accurate depiction of where we are going with armed, autonomous systems. We are not targeting people - only the weapons that they use. And not necessarily with what one would consider to be a traditional weapon, such as a gun or missile. The point is to disarm a foe, if you can - not kill him. However, if he is dumb enough to try to hang on to his weapons when our machines go after them (his weapons), then he may be injured, or even possibly killed, in the process. We are looking at "threading the needle" between Asimov's Laws and the Terminator movies.

[Sierra]

It may be essential for humans to hardwire the Asimov rules into the robots. Also an emergency shutdown command. It is our interest, if we are not doing it - evolution will take over, and the chance of control might gone forever. If we design robots to kill humans (or even animals) we deserve our fate...

[John Canning]

The purpose of this forum is to hold debates regarding questions such as you pose, but in this vein, I would point you to the current works of Dr. Ron Arkin, Georgia Tech, to see the other end of the spectrum.

Dr. Arkin is proposing battlefield robots with an "ethics module" that would allow them to process the current Rules Of Engagement more even-handedly than current human soldiers, and thus be allowed to deliberately target and kill humans.

There is no doubt that Dr Arkin's approach would lead to a more ethical conduct of warfare, if his "ethics module" concept could be produced without errors, but that is a pretty big "if."

We think that our approach of disarming a foe is much more preferable.

[Noel Sharkey]

There are many problems with Asimov's laws of robotics. In fact Asimov introduced them so that he could play out the problems and demonstrate why the laws were not workable. There were somehow an integral part of the positronic brain that would not work without them. Except a few robots found a way round them.

Asimov's laws would of course be no good for military applications as the robots could not harm humans under any circumstances. Although they could allow harm to come to themselves if it saved a human life. So bomb disposal and demining are possibilites.

In the 1980s Asimov introduced a fourth law that really opened a can of ethical worms (or is it an ethical can of worms?). Here he allowed robots to kill humans if it was in the greater interests of humanity. So a human attempting to blow up a city could be killed.

The fine print on this one gets into the nitty gritty of ethical questions that have kept philosophers busy for centuries and is enough to fuse anything but the toughest robot's computer (or positronic brain). In ethical circles it can take the form of: You are standing beside the points on a railway when you see a train rapidly approaching. It is going to run into a group of 10 deaf children unless you pull the lever and send it down the other track. But there is one deaf child on the other track - do you let the train go on course to kill the 10 children or do you pull the lever which effectively makes you a murderer?

You can then play the game of juggling who is on the two tracks - 10 deaf children verses 20 octogenerians etc.

Another form is the triage: The powersource in a hospital is going down and there is only enough emergency power to support one person. Who do you choose.

Current robots are really not up to this kind of decision. It really requires human reasoning and an understanding of humanity outside of computational grasp.

Of course these were fictional robots that could sense as well as humans if no better. The robots of today for tomorrow battlefield do not have such sensing ability and could not distinguish friend from foe or combatant from enemy which leaves Arkin's artificial conscience bereft of accurate input. Similar problems (although perhaps more solvable) exist for John's idea of taking out weapons and not people. Accurate weapon detection is still some way off.

Another problem is that, even with accurate weapons detection, there are many scenarios where innocents could be killed e.g. As has happened in Iraq, children have been forced to carry weapons by an insurgent who will shoot them if they drop them.