Computers might order our shopping, fly our planes and run our factories but to date they’ve proven to be pretty lousy conversationalists.
Take this choice interaction between a person and a chatbot, a piece of software attempting to pass itself off as human.
Human: “Could you kill someone if you stabbed them with a towel?”
Chatbot: How are you doing today? I didn’t hear you. I didn’t hear you.
Human: I am not speaking - I am typing
Chatbot: OK. Do you have any pets? I didn’t hear you. I didn’t hear you.
Not quite Noel Coward is it? This interaction seems to be pretty much typical of your average silicon-based raconteur. Computers may crunch sums at mind-bending speeds but when it comes to verbal skills they’re barely out of kindergarten. Even the iPhone’s loquacious personal assistant Siri will come unstuck once you start making complex queries.
But creating convincing chatbots matter does matter: Find a machine that can pass itself off as human in conversation and you may have found a machine that can do something thought to be uniquely human, think.
The notion of identifying sentient computers by talking to them stretches back to 1950, to the dawn of the information age, when the brilliant British mathematician Alan Turing devised a game to answer the question ‘Can a machine think?’.
Today that game is known as the Turing Test and involves judges talking via a computer to two unseen correspondents - one a human and one a piece of software, and then having to decide which is which.
Turing expected that by the year 2000, computers would be able to fool one in three judges into thinking they were human after five minutes of conversation, and that this would mean “one will be able to speak of machines thinking without expecting to be contradicted”.
That hasn’t yet happened, and, after 21 years of pitting chatbots against the Turing Test as part of the annual Loebner Prize, the closest that any piece of software has come was in 2008, when a program called Elbot fell one vote short of fooling the required one third of judges.
Yet two decades of failure has not deterred researchers from trying to win the Loebner Prize, and tomorrow chatbots will attempt to become the first computer whose responses are indistinguishable from humans - a feat that would win their research teams $100,000 and a gold medal.
Despite the cynicism with which the Loebner Prize is viewed by some academics, a competition which might - just might - identify the first glimmerings of a thinking machine is massively exciting.
So if you were one of the judges, what questions would you ask to work out whether you were talking to a human or a chatbot?