Brian Christian’s article, “Mind vs. Machine,” in the March 2011 issue of The Atlantic covers the Loebner Prize competition which administers the Turing Test to artificial intelligence (AI) programs. This involves an instantaneous form of instant messaging (IM) in which judges have to determine within five minutes whether they are conversing with a computer or a person.
What computers teach us about being human when we attempt to program them to converse is really how formulaic our conversations can be. Therefore, are we really putting any thought or effort into them? The computers sometimes fool the judges, often enough to make us stop and think. When the first such conversational computer arrived, Eliza (more), many people mistook the experience for a human one.
For instance, casual conversation tends to be relatively predictable and easily programmable because it’s driven by the most recent comment. It rarely refers back to prior comments made three, four, ten or more in the past. In other words, we don’t require any knowledge of the conversation’s history to continue it. We don’t need to remember and sort through relevant points to formulate our own ideas. We only need to remember the most current comment.
Returning to what computers can teach us about being human, I was prompted to recall that someone had emphasized to me the importance of “getting down to a human level” when dealing with people. This made me wonder:
If our conversation is programmable and actionable by a computer, can we really be human with one another?
Perhaps this explains why we find some people’s conversations shallow. They’re conversing at a level no better than a computer.