Chatting To Machines

More and more often, when we call a service provider in frustration because something isn’t working as it should and used to, we go through a phone tree. Most of these are prerecorded human speech. However, in many circumstances we find that we are talking to a computer.

How do we know if the reply is from a human or not? Well, we’re good at listening to the responses and can’t be fooled, can we? The aim of the robot designers is to get their robot to pass the Turing Test. Not heard of it? It’s a simple test that a judge looks at text responses from a human and a machine. If the judge is fooled, the robot graduates.

Bad news; to date none have graduated. For some years there has been a contest for The Loebner Prize for someone to build a robot terminal that cannot be distinguished from humans by a panel of judges. Data from the past ten years has been analyzed by Lortie and Guitton from Laval U in an attempt to understand why judges judge as they do (1). You see, some humans were judged to be robots after a conversation with the judges.

Most of the effort of the IT brigade has been to build on what makes a robot appear to be human. The authors came up with the thought that fresh light might be shed on the problem by looking at why some humans seemed to be robots.

So what makes a judge decide on whether they are talking to a human or a machine? First of all they like to hear lots of chat with personal questions and lots of polysyllabic words. Machines are, after all, known to be brief up to the level of brusque if not rude. Not too much laughter or cheerfulness though, that’s not good. We humans wouldn’t be so casual.

Setting things up to get the judge to ask lots of questions was a sure-fire winner. Judges, like the rest of us, like to think that the entity that we are communicating with is actually interested in us, and what we have to say. Hence the conclusion that the judges' predujices play a major role in their "impartial" judgments.

It would seem valuable to use this research for training some customer service personnel rather than programming more companiable robots


Leave a Reply