In February 2011, IBM’s Watson computer competed against Ken Jennings and Brad Rutter, the two most successful human champions in the history of the quiz show Jeopardy!, and won. According to IBM’s own history, Watson finished with $77,147, far ahead of Jennings and Rutter, and the winnings were donated to charity. The match was watched by millions on television.
Watson was built by an IBM research team led by David Ferrucci, who had pitched the idea in 2006 as a way to push natural-language processing forward. The underlying system, called DeepQA, is described in the team’s paper “Building Watson: An Overview of the DeepQA Project,” published in AI Magazine. Watson parsed the wordplay-laden clues, searched a large body of stored text, generated many candidate answers, and ranked them by confidence, all in a few seconds and without any live internet connection.
Where Deep Blue’s 1997 chess win was a triumph of search over a closed game, Watson’s victory was about open-ended language understanding. It became one of the most visible public demonstrations that machines could handle messy human questions, and it gave the word “Watson” a lasting place in the popular story of AI.