Monday, September 13, 2010

"How to Train Your Robot (to Lie)"

In an earlier entry I expressed my discontent with our current terminology about robots - specifically, that we use the word "robot" to cover too many distinct types of artifacts. This article from ScienceNow illustrates another problem with how we talk about robots, somewhat reminiscent of Jaron Lanier's concerns over The First Church of Robotics.

The claim in the article's title is that a robot has been taught how to lie, but it has only learned how to lay a false trail. This can be seen as deception, or misdirection, or perhaps even camouflage - all of which can be seen in non-human animals and even plants - but hardly lying. No doubt the work represents a breakthrough, but is the hyped-up language necessary or productive? Am I the only one who thinks longingly of the days when science news was a bit more sober and restrained?

Ken Pimple, PAIT Project Director

No comments: