Jeopardy Watson
A red-letter event in Jeopardy history was the Jeopardy Watson
match, where the IBM computer "Watson" took on the two most
successful Jeopardy contestants in the trivia game's history. Ken
Jennings was the man from Utah who captivated America by winning 75
straight games, while Brad Rutter was the all-time money winner on
the popular tv show when he took on Watson.
What Is Watson?
Watson is an artificial intelligence computer able to answer Jeopardy questions. After IBM's artificial intelligence Deep Blue defeated chess champion Gary Kasparov at chess in 1997, IBM developers were looking for a new challenge. That challenge came when one research executive was in a restaurant one night and the place stopped down to watch Ken Jennings in his amazing 74-game winning streak on Jeopardy in 2004. The executive challenged IBM researchers to design a computer that could win at Jeopardy, but potential designers were scared off by the complexity of the challenge. Eventually, Paul Ferucci took up the challenge in 2006.
In original tests in 2006, Watson was only able to answer about 15% of the questions, while top Jeopardy! players have a success rate around 95%. Early attempts also required several minutes to respond to clues, instead of the split-second pace Jeopardy is played at. This was enough to get an IBM design team full of 15 members funding for a 3 to 5 year project. By 2008, that team had a computing system that could compete with human contestants. By early 2010, Watson was able to beat human Jeopardy contestants regularly. It would be another year (February 2011) before the Jeopardy! matches between Watson, Jennings, and Rutter took place.
Part of the delay involved complicated negotiations between Jeopardy's producers and IBM Research on two key issues: how clues would be written and how the buzz-in mechanism would work. The first issue involved IBM being concerned that Jeopardy's writers would write clues that would present trick answers for an AI or what amounted to a Turing test. This was resolved by selected questions from un-aired episodes of Jeopardy. The second question involved hooking Watson to a robotic button pusher, so Watson would have to go through the same mechanical process of manipulating a device that human contestants do. Despite this inclusion, Watson's biggest advantage would prove to be its ability to buzz-in quicker than its human opponents.
How Does Watson Work?
In the layman's explanation of how Watson works, the IBM ai associates a word or words with other words. When Jeopardy clues are presented to Watson, the computer matches the keywords in the clue to a big list of other words. Using algorithms that help Watson associate the clue words with other words on the big list, the computer comes up with the best guess in terms of percentages.
During the Jeopardy shows, when the Ai Jeopardy contestant answered a question, the broadcast displayed Watson's top three choices for an answer, which gave some indication at the process Watson used to arrive at an answer. Sometimes, the percentage would be overwhelming, up around 97%. Sometimes, as in the famous (or infamous) Toronto answer, the top guess (Toronto) was only at 14%, while the correct answer (Chicago) came in second at 11%. That's why Watson "wrote" it's answer with 5 question marks, because it was programmed to write a question marks when the percentage fell below a certain level.
Watson vs Human Jeopardy Players
Watson's biggest advantage was it's ability to ring in answers quicker than its human opponents. On the other hand, Watson displayed weakness in questioning answers that contained only a small number of words. On pure trivia questions involving literature and music, such as "who was the composer of this piece of music" type questions, Watson seemed unbeatable.
Watson's sometimes caused the audience to break into laughter with a few oddball answers, because Watson sometimes wasn't able to put the words in the clue into a proper context the way a human being would. Another funny moment is when Watson asked "What is leg?" instead of "What is 'missing a leg'?" On a video about Watson's many matches against 800 other Jeopardy opponents in preparation for its television showdown, Watson broke up the audience when it answered "What is milk?" in response to the clue "the first non-dairy creamer substitute". Also, Watson's choice of money to wager on a Daily Double brought a big laugh, because of its specific choice of $6,435 instead of $6,500.
Who Won the Watson Jeopardy Exhibition?
After the first day of Jeopardy, Ken Jennings had $4,800, Brad Rutter had $10,400, and Watson had $35,734 (wagering only $947 on an incorrect answer)? After the second and final day of the competition, Brad Rutter finished with $21,600, Ken Jennings came in second with $24,000, and Watson was the big winner with $77,147. As per agreement for the exhibition, Brad Rutter went home with $200,000 for his favorite charity for taking 3rd place. Ken Jennings won $300,000 for his favorite charity, while IBM split $1,000,000 in Jeopardy winnings between two charities.
Later Congressional Contest
In a later contest between 5 members of the United States Congress (who took turns playing as a team), Watson won a much-closer competition with $40,300 against $30,00 for the combined Congressional team. The five members of the U.S. House of Representatives who took part were Democrats Rush D. Holt, Jr. of New Jersey, Jim Himes of Connecticut, and Jared Polis of Colorado, along with Republicans Nan Hayworth of New York and Bill Cassidy of Louisiana. During the battle between Watson and Rush Holt, a former Jeopardy contestant, Representative Holt had a lead over the IBM computer player.
Is Toronto in the United States?
In the Final Jeopardy section of the first day of competition against Jennings and Rutter, Watson incorrectly answered "What is Toronto?" to a "U.S. Cities" question involving clues about what its two biggest airports are named after. This created a little buzz and more than one questions about how this answer was derived, while the leader of the IBM Research team responsible for Watson wore a Toronto Bluejays jersey for cameras the next day.
Watson and the Singularity
Despite its victory against human opponents at Jeopardy!, Watson is far removed from the much-speculated "singularity". As scientific experts and commentators on the Watson competition noted, Watson doesn't really think in any meaningful way--it recognizes symbols. Despite it's impressive showing at Jeopardy, Watson is learning through associating one symbol with the next, not by applying learned knowledge to the context of the situation the way human beings do. It's noted that humans know so little about the human brain, that this imposes a major limitation on building devices that can truly simulate the human thinking process. Still, IBM Research moved the development of artificial intelligence along significantly in designing Watson, bringing computer intelligence design one step closer to truly intelligent artificial intelligence.
