Robots Taught to Deceive 239
Posted
by
CmdrTaco
from the of-this-will-be-fine dept.
from the of-this-will-be-fine dept.
An anonymous reader found a story that starts "'We have developed algorithms that allow a robot to determine whether it should deceive a human or other intelligent machine and we have designed techniques that help the robot select the best deceptive strategy to reduce its chance of being discovered,' said Ronald Arkin, a Regents professor in the Georgia Tech School of Interactive Computing."
Proof that humans are dumber than dogs (Score:5, Insightful)
That a human being would teach a robot to deceive only proves that we humans are dumber than dogs, as dogs don't shit in their own backyard unless they have to. We humans will shit in our own backyard by choice.
Re:Proof that humans are dumber than dogs (Score:5, Insightful)
Incidentally, dogs are actually smart enough to intentionally deceive their owners.
Re:A few things.... (Score:3, Insightful)
define:belief - any cognitive content held as true.
Not the, "Oh look, Johnny5 died and he came back as a T-551 model because he was good! Praise Serial number 00000000001!!" kind.
I think the whole concept of deception is a necessary step in robotics for communication. What's the difference between deception and non-literal communication? Not much.
For the first crappy example that comes to mind, if I'm talking to someone and they use a double negative, I have to deceive them into thinking I heard a single negative. If this deception fails, the communication might get awkward or fail, and tho whole relationship could change ("They think I'm an idiot.", thinks the other person).
For the terribly imprecise (for most) nature of human speech, the whole concept of knowing someone is incorrect and figuring out what they *actually* meant rather than what they spoke, and tolerating someones belief/opinion that you think is wrong all involve deception to keep the communication smooth. At least IMO.
Of course, someday I might find myself dead and robbed in an alley after following what I thought was some robot woman who needed my help (since us great apes are so comfortable in the trees) getting her robot kitty from a robot tree. :-\
Re:Duh (Score:1, Insightful)
Allowing robots to develop their own behaviour unsupervised would be enough. Wouldn't take long for them to find out that deception can be benifical under some circumstances.
Genetic algorithms 'like' to cheat. If your specification of the problem isn't strict enough and there are way left open to cheat, the algorithm will occasionally stumble over it and use that solution, if it's a superior way to achieve the goal.
Re:Exit Asimov (Score:5, Insightful)
Asimov himself wrote about robots that were capable of lying.
"Liar!", from "I, Robot", is about a robot who develops the ability to read minds and lies to people because he interprets hurting their feelings as a violation of the First Law.
"Little Lost Robot" (same book) is about a robot who, after being told vehemently to "get lost!", manages to hide among other robots of the same model and deceives its owners trying to obey that command.
There is no Law of Robotics that states that a robot shall be truthful to a human being, or by inaction allow a human being to be deceived.