In fact, not a lot has changed since Joseph Weizenbaum's 1966 Eliza program used Rogerian psychotherapy techniques to convince people that the computer was capable of being human. (You can still try it out on the Web at sites like www.binary.net/dturley/eliza/ eliza.cgi.) One of the first people to seriously dream of artificial intelligence was Alan Turing (considered by many to be the father of modern computer science). In a paper he wrote in 1950, Turing posed the question -Can a Computer Think?". He observed that the answer came down to how we define -think". And so Turing came up with a logical behavioural test: if an outsider cannot tell the difference between a computer and a person (in other words, if each is picked as the human 50 percent of the time, there is no basis for saying that one can think and the other can't). In that fifty-year-old paper, Turing predicted that by the year 2000, a computer would be identified as human 30 percent of the time, if tests were limited to five-minute conversations.
To find out just how close computers have come to fulfiling Turing's prediction, each year Flinders University sponsors the Loebner Prize competition. In this event, judges log into a -chat window" and carry on a five-minute conversation with either a human or a program developed by one of that year's entrants. In the 1998 competition, the winning program fooled 15 percent of the time, earning it the US$2000 Annual Prize and Bronze Medal. (According to the Web site, -once entrants reach the 30 percent level . . . there will need to be a playoff for the $25000 Turing Prize and Silver Medal.") To see how entrants fared in this year's competition, you can visit www.cs.flinders.edu.au/ research/AI/LoebnerPrize and read the conversations between judges and computer/human entrants.
It's an intriguing challengeââ,¬"I watched many of the conversations live over the Web when the contest was held in January. What was disappointing was the fact that a lot of the computer entries looked as if technology hadn't improved much since the days of Eliza. The winner this year only managed a 10 percent -Turing rate".
Why has there been so little progress? I'd have to guess that the skill had yet to find its commercial niche (contest prize money only lasts so long). But maybe that is beginning to change. Artificially intelligent interfaces (let's not even bring up Microsoft's Bob at this point) are beginning to appear on a great many Web sitesââ,¬"and they're better at conversing than anything I've seen in a long time.
Many are designed for companies desiring a virtual customer service representative (a rep who'll work 24 hours a day without going on strike). They are impressive in their ability to learn (and because they're basically advertising themselves to potential customers, the better they are at that the better their chances of -survival" in the marketplace. Here are a few you can test for yourself: there's Andrette at www.bigscience.com, Roy at www.artificial-life.com, Red at www.neuromedia .com, and Alice at birch.eecs.lehigh.edu/alice.
If you've come across any that deserve a bit of recognition, let me know at bhaverty@zdapn .com.au.
Brian Haverty is the editor of PC Magazine Australia
