IBM unleashed Watson, a computer designed with the singular purpose of playing, and winning, the TV quiz show Jeopardy. In his first public attempt, Watson went up against the two top Jeopardy players in Jeopardy history. He literally murdered his human competition. Well, no, not literally. Although he did sound at least a little bit like HAL from 2001: A Space Odyssey.
The challenges involved around teaching Watson to properly interact with natural human language are actually quite deep and could be some of the most important steps in the evolution of the machine. The clues on Jeopardy are written for human competitors. They rely on factual knowledge, but also on linguistic idioms, double meanings, cultural references, and even cheap puns. Computers have notoriously bad senses of humor.
Watson, because he is a machine, does not understand things as we do. Instead, Watson processes an incredible amount of data and then calculates statistical likeliness of fit with the information given in the clue. This requires a huge amount of data analysis that goes beyond mere keyword recognition and begins to emulate human understanding.
A couple of weeks ago, TED.com posted a panel discussion of IBM insiders talking about the challenges involved in building Watson and possible future applications of Watson technology. First up on the panel was Kerrie Holley, coauthor of 100 SOA Questions Answered. You may remember our interview with Holley last month when he explain SOA infrastructure, among other things.