Man vs. machine for $500


Jeopardy is a long-running US quiz show that gives contestants an 'answer' to which they must respond with the appropriate 'question'.  Each round will have a range of questions of increasing difficulty (and of increasing prize money value).  It's enjoyable to play along if you find yourself sitting in a US hotel room at around tea-time with 30 minutes or so to spare, although the US-centric nature of some of the topics might put some of us at a disadvantage.  Well, that's my story at least.Earlier this month (February 2011) Jeopardy's greatest champions returned to the studio to take on a new contestant - IBM's Watson.  Over three years in the development, Watson was created for this specific purpose, although the lessons learned from the project go way beyond 'how to win a quiz show'.   Watson needed more than to simply have access to millions of 'facts'.  He needed to be able to master puns, humour, allusions, slang and nuance, and weigh the probability of each potential correct 'question' being correct before buzzing in. This article, published on McKinsey Quarterly, tells the story of the complexities of the development stage.  The article takes just one 'answer and question' as an example.  In a round called 'Diplomatic Relations' Watson must dissect and respond to this:"Of the four countries the United States does not have diplomatic relations with, the one is the farthest north."In order to answer this correctly, and more quickly than any other contestant, Watson needs to perform many functions, ranging from identfying the 'type' of question (is it a historical fact or a limerick for example?); working out the grammar of the clue (nouns, verbs etc); working out what 'country' refers to (country music?!); before going on to identify potential countries and geographic locations.This YouTube video shows Watson taking on the champions, displays lists of his alternative potential answers (some of which are very entertaining) and shows some footage of the early stages in IBM's development of Watson.