When it comes to "killer robots", my philosophy -- as most of you know by now -- comes down to, "We should think about it, but there's no need to panic." Fortunately, there's a group that's doing just that: the International Joint Conference for Artificial Intelligence. Their meeting last month generated some really interesting discussion, including calls for a wider debate on the use of AI in military and security applications, but also discussed the benefits that more advanced artificial intelligence could bring us. (You can read many of the papers online here)
Keep in mind that the definition of "artificial intelligence" reaches far outside of robotics, so these discussions also considered technology like computer viruses -- which some researchers claimed had reached "cockroach-like" levels of self-preservation and reproduction -- and software that tries to break CAPTCHAs. Spambots and the Conficker worm generally don't figure into my day-to-day robotics engineering but they might be closer relatives to Skynet than my Roomba is.
The gist of it is hesitancy to hand decision making over to an autonomous robot. This comes on the heels of a plan released by the US Air Force to move toward decison-making aerial drones by the year 2047 (pdf) -- an idea that would make almost anyone nervous, and an interesting change from the Army's Future Combat Systems' insistence that they want to keep a human in the loop. But the IJCAI's concerns about autonomous decision making have legs well outside of military robots; there is much discussion and nervousness lately of automated computer trading systems on the stock market. (As I'm writing this I'm also listening to NPR, who have just informed me that the SEC has already moved to ban this practice)
At the same time the proceedings of this conference came out, an interesting article (behind a pay login -- sorry, non-IEEE members) was published in the IEEE Intelligent Systems magazine that referenced the work of a couple of roboticists who are interested in rewriting Asimov's Three Laws of Robotics for modern application. The good news is, they got it right that Asimov's laws are plot devices and neither can nor should be taken as gospel for a robotic code of ethics. The bad news is... well, take a read for yourself, and then I'll tell you what I think.
Here are Asimov's original laws:
- A robot may not injure a human being or, through inaction, allow a human being to come to harm.
- A robot must obey any orders given to it by human beings, except where such orders would conflict with the First Law.
- A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
And here are the proposed new three laws:
- A human may not deploy a robot without the human-robot work system meeting the highest legal and professional standards of safety and ethics.
- A robot must respond to humans as appropriate for their roles.
- A robot must be endowed with sufficient situated autonomy to protect its own existence as long as such protection provides smooth transfer of control which does not conflict with the First and Second Laws.
While it's a cute PR move, copying the famous Three Laws infrastructure, these really appear to be some guidelines of which there happen to be three. They also are guidelines for the human creators of robots, not robots themselves -- and while the creators acknowledge this fact (you can read it in the abstract linked above), it defeats the purpose of the original Three Laws, which were hard-coded into the positronic brain of every robot and were supposedly failsafe. And let's face it... these laws, especially the first two, are pretty darn vague. You think Asimov's laws had loopholes to be exploited... boy, what happens the first time a robot gets around to reading some history books and finding what used to be some of humanity's highest legal and professional ethical standards?
It is all tongue-in-cheek, of course, but I think this Three Laws buiness just shows how uncertain we are of what we want our robots to do. We can't even make an absolute statement of "don't kill anyone." We can't tell if robots should be doing our bidding as mechanical slaves, or if there are roles in which it is "more appropriate" for a robot and a human to interact on a more equal level. We've got a long way to go until we understand where artificial intelligences actually fit in our world -- and the good news is, I think we've got a decent amount of time on our hands to do this.
So in summary: keep thinking, folks, but keep not panicking.