When I first started to programme in BASIC on the Sinclair ZX-81, the programme that captured my interest and imagination the most was Eliza. I spent hours entering strings of phrases and took delight when the programme appeared to engage with me in ‘conversation’.
I was interested to read about Elbot, an artificial intelligence programme, that fooled 25% of its interlocutors into believing it was a genuine human being during the final round of the Loebner Prize in artificial intelligence this year. A score of 30% would have been sufficient to pass Turing’s criterion for a true artificial intelligence.
Sadly though, Elbot is so underwhelming and unconvincing that you begin to wonder more about the intelligence of the 25% of interlocutors who thought it was human. As the Guardian notes,
The winning software, designed by an American and called Elbot, fooled 25% of respondents. “We really, really have come very close,” said Warwick. His sample size, in a field made up largely of computer experts and journalists: 12.
Here’s a sample conversation I had with Elbot just a few minutes ago.
Elbot: Well, this is indeed a surprise. A visitor from Sri Lanka!
Me: Wow. How did you guess?
Elbot: In order to guess, I just look up ‘accepting things at face value’ in my instruction manual and do whatever it tells me to do.
As long as the conversation is this grammatically correct, stilted and isn’t littered with the odd expletive, anyone who is a tad more intelligent than Elbot won’t have a hard time distinguishing between a real person’s and a computer generated conversation.
I have no doubt that computer speech synthesis and recognition, with real time semantic language processing run on massively parallel processing power (perhaps the combined power of the Internet via the cloud?) will make it impossible to distinguish between human and machine conversation. Like KITT in Knight Rider, or HAL 9000 in Space Odyssey.
Until such time though, I think it’s a safe bet to keep peace negotiations to humans.