The most paranoid will no doubt see the beginnings of Skynet, this IA which rebelled against humankind and triggers war between robots and humans in Terminator.
Gordon Briggs and Matthias Scheutz researchers the laboratory of interaction human/robot at Tufts University in Massachusetts come to present the results of their work.
Holding the 3 laws of Asimov notoriously inadequate, they wanted to give the robot a more “humane” approach and learned the robot to… disobey.
Learn how to robots to refuse certain orders
‘Sorry, I can’t Do That’, is named the research topic presented by these two researchers in Washington, at the recent symposium “ have for Human-Robot Interaction”. They developed algorithms that allow a robot, a Nao in this case, to refuse the orders of a human if the robot has a good reason for it. An operator asks Nao to advance while it lies on a table, refuses because the fall is inevitable.
The goal for these researchers is to implement in the robot close behavior of human behaviour where the conduct of an individual depends on what is called the condition of success (felicity conditions). beam elements that make the person will judge that it must execute the order or not. Human like robot, this judgment goes beyond the simple 3 laws of Asimov (not to hurt a human / obey humans without infringing the law n ° 1 / protect his own existence without contravening the first two laws). In the behavior implemented by the researchers, the robot will consider 5 factors prior to obey an order.
Some are pretty trivial: knowledge (is what I know do the task that me is requested), capacity (am I able to do it), priority (can I do it now), the next two are most amazing for a robot. It will firstly assess the social aspect of its act. Should he obey the person who gives it an order? In addition, it will explore is there permission to carry out his act. It is this stacking of rules “of life” that prevent Nao of falling of the table although requested to do so by his master.
Translation : Bing Translator
Source : “Researchers Teaching Robots How to Best Reject Orders from Humans”, IEEE Spectrum, November 19, 2015