Wednesday, February 16, 2011

Language, Artificial Intelligence and WATSON - the new super-computer Jeopardy Champion

This week’s Jeopardy IBM challenge pitted two human super-champions (Ken Jennings and Brad Rutter) against a custom-designed IBM super-computer named Watson. The game was intended to showcase machine capability not just for fast hard-core calculation but also for softer but equally complex skills like language processing. Thanks to the publicity in several media channels, I managed to catch the second and third game of the three-day series. Curiously enough, I found myself in an Us-and-Them mode, rooting for the human players (who of course lost thoroughly to the machine). 

It has been 42 years since Stanley Kubrick introduced the fictional super-computer HAL in 2001 Space Odyssey that was perfected to a human-like complexity to display emotion and language processing ability well beyond its artificial intelligence. The ongoing quest to develop a real-life HAL remains an evolving goal, although IBM seems to make that goal one of its promotional targets, while providing some entertainment to us.

First was the introduction of Deep Blue in 1996 to play chess against grand master Gary Kasparov. There was, however little human about Deep Blue, its chess moves still came from brute force computation of established rules.  Even that achievement remains embroiled in controversy. Kasparov, who lost complained of unfair human intervention by IBM to reprogram the machine during game.

Introduction of Watson on Jeopardy’s stage has passed without any such controversy. The human competitors seemed to gracefully accept its lightning speed and its agility with the buzzer. In fact one of them added the following sub-text to the final jeopardy response “I for one welcome our new computer overlord”

Over and beyond our cognitive abilities, language with its puns, metaphors, analogies, double meanings represents a core human attribute. One that gives expression to our personality, attitude and perhaps to more abstract concepts like our consciousness. In fact, voice and language abilities are often used to give human characteristics to animals or other natural objects in art, literature, movies. Computer scientists have even defined a “Turing Test” (named after Alan Turing) to judge a machine’s intelligence by testing its natural language or conversational abilities. An electronic system is said to have passed the Turing Test if its conversation is indistinguishable from that of humans.

With its stacks of ninety servers and instantaneous ability to process text of the answer, Watson's victory was not much in question. More significant part of the game probably was the few rare clues where Watson faltered. One example was a clue on what Shengen treaty opened up (I’m paraphrasing) – Watson’s response “Passport” was related but not quite the correct contextual one – “National Border”.  Probably the real-life machine still has some work cut out to pass the Turing test !

Regardless of those rare faux paus, Watson definitely represents a milestone for computational progress at processing natural language and parsing convoluted statements often with multiple meanings. We sure are making progress to the day when we may have “Turing Test” compliant appliances, cars and droids that would interact more “humanly” with us. 

No comments:

Post a Comment