Beyond Asimov; The Three Laws of Responsible Robotics

IEEE Intelligent Systems just printed our (David Woods, OSU, and my) article about "Beyond Asimov: The Three Laws of Responsible Robotics" and put "Beyond Asimov" as one of the articles on the cover... and the hate mail has started!

So what does Asimov and his Three Laws of Robotics have to do with rescue robotics? The Three Laws are being taken seriously as a framework for discussing human-robot interaction. Rescue robotics has humans behind the robot and humans in front of the robot- it's about as human-centric as you can get. I became one of the early drivers of the human-robot interaction community (I co-chaired the seminal 2001 DARPA/NSF study) precisely because I found through my fieldwork that the poor interaction was the limiting factor. No matter how bad the rescue robots were in terms of locomotion, communications, sensing- the horrible mismatch between the robots and the human cognitive abilities for the environment was the limiting factor. It isn't just interfaces, it is the set of fundamentally, pervasively wrong assumptions about how people interact with robots.

Whenever I hear some grad student talking about wanting to design robots which meet Asimov's Three Laws and thereby provide perfect human-robot interaction I get ill. One year I heard a researcher telling the press that their robot met the First Law of Robotics (a robot may not injure a human) because it was able to avoid people. Except that it was simply avoiding heat sources and people happen to be warm.

To say AI researchers tend to be technological optimists is an understatement.

The paper came about when I began to read Moral Machines (David texted me that I had to stop whatever it was I was doing and go read it now, he was so put out by the book). I next-day-ed the book, and between that and the Living Safely with Robots tome, shouted "enough with the Asimov's Laws already as some sort of gold standard for robot ethics. It was a literary device. Let it go!" My family tends to find things to do away from the house at time like that. I thought the Moral Machines actually made a strong, though unintentional, argument for why Asimov would get sued if he were a robot manufacturer.

So I whipped up a draft on alternative laws one Saturday morning. Leila Takayama and Victoria Groom from Cliff Nass' group at Stanford read it, make great suggestions, and included in their HRI workshop. I sent it to David to read and he came back with excellent ideas, tons of experiences and examples of how autonomy and automation fails, and way better prose. I insisted that we stay with three laws and that they had to be symmetric with Asimov's-- sticking with literary convention to make a point. I agree with David, if you really want laws, it'd be better to start over. Anyway, we put a version in an IEEE ICRA workshop (thanks Cindy for presenting!) and continued to refine. We ran it past Robert Hoffman who saw the possibilities of getting a more informed discussion going and after a rapid edit cycle and a discussion with Jeff Bradshaw, it's in print. (I'm sharing names in a Good Way, please don't go yell at them if you hate the paper.)

Hopefully, besides hate mail, we'll get a real intellectual discussion going instead of extreme quotes in the media. AI robotics is capable of so many things, I hate to handicap true progress by adherence to a cute literary devices designed to create problems.