I can recall having great fun talking with ELIZA more than a decade ago. Fun yes, serious AI no. It didn’t take long to get bored with the repetitive question and answer looping even if it was wrapped up in the guise of being intelligent machine driven conversation: an oxymoron then, and still one now.
Not that times and technologies have stood still, far from it, but with the announcement of the winner of the annual Loebner Prize in the form of ‘Joan’ which picked up a bronze medal and a couple of thousand dollars, have things really changed that much? I would venture that the answer is no.
We are told that Joan, a veteran Loebner entrant (its first stab being in 2003), has been evolving over the years, becoming ever increasingly human by learning from thousands upon thousands of previous ‘conversations’ but I’m afraid I just don’t buy it. Moreover, neither, I venture, do the Loebner judges seeing as the gold medal and accompanying $100,000 prize remains unclaimed some 16 years after being first dangled in front of the AI development community. Could the reason be that the criteria for a machine to be indistinguishable from a human during the course of a conversation is a technological leap too far? Well yes, frankly.
Despite British entrepreneur Rollo Carpenter’s best efforts with Joan, it still cannot pass the Turing Test required to win Loebner gold. Rollo knows what it takes to win the annual competition, he grabbed it last year with George that was powered by his Jabberwacky engine just like Joan, but that elusive big breakthrough remains as distant as ever.
Alan Turing, the infamous British mathematician responsible for helping to crack the German military Enigma codes during World War Two, posed his question regarding AI way back in 1950. He asked how would we could tell if a computer was truly intelligent, and answered with the idea that a computer capable of providing conversational responses indistinguishable from a human being would truly be artificially intelligent. Jabberwacky is not truly intelligent, it does not ‘think’ it merely selects the best fit response from an ever growing database of responses collected through online conversations at the Jabberwacky site. OK, that may be oversimplifying it, but not by much. You can make your own mind up, of course, by talking to Joan or George and seeing how long it would take you to guess you were not talking to a real human being.
It took me all of, well, less than a minute...
