Slashdot Mirror
The Question of Robot Safety
An anonymous reader writes to mention an Economist article wondering how safe should robots be? From the article: "In 1981 Kenji Urada, a 37-year-old Japanese factory worker, climbed over a safety fence at a Kawasaki plant to carry out some maintenance work on a robot. In his haste, he failed to switch the robot off properly. Unable to sense him, the robot's powerful hydraulic arm kept on working and accidentally pushed the engineer into a grinding machine. His death made Urada the first recorded victim to die at the hands of a robot. This gruesome industrial accident would not have happened in a world in which robot behavior was governed by the Three Laws of Robotics drawn up by Isaac Asimov, a science-fiction writer." The article goes on to explore the ethics behind robot soldiers, the liability issues of cleaning droids, and the moral problems posed by sexbots.
Asimov's rules were always applied to intelligent robots. No-one (to my knowledge) has ever suggested that a hammer should have a sensor to recognise if it is hitting a nail or a thumb and refuse to obey the "command" of its operator if it is targetting the latter. The purpose of Asimov's three rules was to prevent himself from falling into the trap of writing yet another Frankenstein story. That said, I believe there are some proponents of handgun biometrics that believe guns should override the commands of their operators if the operator is not authorized to use it. In the future you may not be able to (legally) purchase a handgun that will fire on a human being.
How we know is more important than what we know.
Whoa, transport me back to when E.L.O.'s "Time" album came out (Yikes! 1981) and the song "Yours Truly 2095":
But I digress (before I was ever on topic)... there won't be any moral dilemma for this crowd. The first sexbots will be programmed for "No Geeks" which will only increase their allure for that very crowd. They'll be hacked to remove that restriction, and while they're at it they'll be programmed to hang out at retirement homes, PTA meetings and church services. That'll pretty much doom them to be recalled, pulled from the market, and there'll be only a few remaining examples in the Smithsonian and certain institutions of higher learning for, ummm, "research".
Remember, you read it here first.
You don't use science to show that you're right, you use science to become right.
I'm a post-grad student working on a robot helicopter. It has extremely fast rotor blades and is a very real threat to humans if mishandled, so I can speak from personal experience in working on robot safety critical systems. To me, robot safety is more of the same problem faced by machine safety in general and more of the same problems faced by robots in particular.
Firstly all potentially dangerous machines require correct operation to avoid injury. No one can stop an idiot from ignoring a safety railing of a machine, automatic or robotic. To expect safety after defeating barriers and interlocks is stupid for microwave ovens and toasters, let alone high energy robotic systems. To expect robots to be safe outside of their defined operating parameters is like expecting a car to be made of sponge so no matter how much you ignore the speed limit, you can't kill anyone.
Secondly, robots seem to suffer a higher demand for intrinsic safety because of the expectation of robot cognition. The reality is, this is the place in robotics where the technology least developed. How do people possibly expect a robot to implement the three laws if the robot cannot flawlessly recognise a human as human? Furthermore, the three laws make no sense for a system that generally works far removed from humans. Putting the sensors and intelligence into a factory robot that should never encounter a human in its powered up state is just stupid. A simple barrier or laser curtain is more than adequate as an interlock, but as we've seen, that doesn't keep humans out all the time. The best the industrial roboticist can practically do is build robot systems that are reliable and stay within their work envelopes.
For mobile systems like my helicopter, it becomes more difficult since you can't control its workspace - cognition bites you in the arse once again. However, the reality of robot-human safety is that dangerous robots working around humans simply should not be autonomous without direct supervision. We are decades away from machines that are autonomously safe around humans. Software is brittle and easy to confuse no matter how well coded it is - you just can't capture all of the edge cases in the real world when you have millions of possible states. Don't imagine robot helicopters flying around people without a monkey in control - it just won't happen.
It seems to me that people need to change their idea of robots away from R2-D2 and towards reality. Treat industrial robots like an piece of industrial equipment - with respect. The same idiots who jump the fence of a robot workcell are probably the same idiots who misuse power tools and ignore safety directives. You just can't stop idiots from earning darwin awards. Seriously, it's not hard to stay outside the yellow tape.
Take your three laws and return them to science fiction, from which they came - they belong to the same realm of fantasy as FTL travel - which is to say, maybe one day but not for a long time.
Scientists point out problems, engineers fix them
altslashdot.org: The future of slashdot.
Actually, there are very good reasons to make a robot aware of its own existence. Certain types of reasoning and learning are helped significantly by the ability to reason about the existence of oneself.
Consider the following experiment, which toddlers have difficulty performing prior to 4 years, but are able to after. A tube is presented to them, with the logo of a candy company on it, "smarties," not the American ones, but the British ones. The child is asked, what is in this tube? At this point, the child invariably says, "smarties!" The conductor of the experiment then opens the tube, revealing pencils. The experimenter asks again, "what is in this tube?" The child says, "pencils." Now, "if I ask another child what is in this tube, what do you think they will say?" Before 4, the kid will say, "pencils." After, they will say, "smarties."
This reasoning task requires the kid to model themselves prior to the revelation that there are pencils in the tube. It requires a model of what happened after. It, further, requires a model of the other child, of what they will be like without this knowledge. This is actually part of a model of self-awareness, but it's not the entire model. You might ask, "why would a robot need to know this?" Well, actually, it's quite important if the robot is to interract with people, because people will expect the robot to behave in an appropriate manner. Dangerous scenarios could arise because the robot does not understand that things that are in its field of view, for instance, are not in the field of view of a person. An example might be a robot handling dangerous materials, during a construction task. Perhaps the person can't see that it's handling hot metal. A person would warn the other person, avoiding danger.
As for the three laws, they were written in a body of fiction. I think that too much attention is paid to them.
- He needed to test something with the power on
- He would have had to reboot the robot
- Getting in through the safety gate was hard
- Someone told him it was safe to enter (that the product feed was inactive)
I've been smacked by robots a couple of times, and while the ones we have are simple lift and put robots that don't hit too hard, it still hurts and if it hit you the wrong way could probably do some serious damage. For the reasons why it's happened to me, see above.[FUCK BETA]