We have a ritual in TechMan’s house.
Every weeknight, TechMan and TechMa’am sit down after dinner to watch “Jeopardy.” And every evening TechMan blurts out the answers, right or wrong.
Well, on Feb. 14 to 16, the ritual will be a little different. TechMan will not be competing against the usual contestants, such as a librarian from Poughkeepsie or an engineer from Tecumcari, but against a supercomputer from Armonk, N.Y.
On those days an IBM computer will play two Jeopardy champs for a top prize of $1 million.
Why would one of the most respected corporations in the world spend the money and time of highly paid experts to compete on a game show?
Let’s find out. Supercomputers on “Jeopardy” for 200 please, Alex.
Watson. What is the name of the supercomputer software that IBM has spent four years readying to play “Jeopardy?” It is named after Thomas J. Watson, a longtime leader of the company.
Ken Jennings and Brad Rutter. Who are the Jeopardy grand champions that Watson will play against?
Stormin’ Mormon Jennings holds the record for the longest streak of Jeopardy wins — 74. A software engineer from Salt Lake City, Utah, at the time he was on the show, he ended up winning more than $3 million.
Brad Rutter holds the record for winnings on Jeopardy — more than $3.2 million and a pair of Camaros. A resident of Lancaster until 2007, he now lives in Los Angeles and is pursuing a career as an actor.
QA. What is Question Answering, the scientific field that Watson excels in?
IBM thinks Jeopardy is the perfect challenge for a machine trying to understand natural human language.
“Jeopardy” requires understanding language, which is completely open-ended, often ambiguous and requires context to understand.
When the announcer reads a question, it is fed electronically to Watson, which first has to understand the question, no mean task for a computer.
” ‘Jeopardy’ clues are expressed using complex, often tricky, natural human language. Just understanding what the clue is asking for is a challenge,” Watson Research Manager Eric Brown told Daniel Terdiman in an interview recounted on a Cnet news blog.
Confidence. What must Watson calculate before giving an answer? In “Jeopardy,” contestants are penalized if they give a wrong answer, so it is important to decide how much confidence you have in your answer. Watson calculates a percentage of confidence in each answer. In determining confidence levels, it can assess the dollar amounts left on the board, how far behind or ahead it is and how well it is doing in a given category.
Unlike some human contestants (usually losers), it is programmed not to answer unless it has sufficient confidence that the response is correct.
But the confidence required for an answer can vary. If Watson is far behind, it will accept a lower confidence level in an attempt to “go for it.” But if it is far ahead, it can refuse to buzz in despite a high confidence level.
Dictionaries. What is one of the sources of information Watson uses to answer questions. Also, encyclopedias, news stories, books and Web content. All this information must be stored internally because the rules forbid Watson from being connected to the Internet while playing.
Watson has 1,000 programs to apply to the info equivalent of millions of books. It understands anagrams, puns, wordplay and has memorized every Shakespeare soliloquy, major river and world capital.
Sparring. What does Mr. Brown call Watson’s preparatory practice matches? Watson has played 79 games against former “Jeopardy” contestants and 55 games against Tournament of Champions players. The computer’s performance is evaluated each time.
Deep Blue. What is the name of an IBM chess-playing computer that beat world champion Garry Kasparov in a famous 1997 showdown?
“But chess is fairly mathematical and well defined — each game state and the corresponding possible moves can be easily represented by a computer,” Mr. Brown said. “Jeopardy” requires an understanding of human language, which is much more difficult.
TechMan’s not a betting man, but if he were, his money would be on Watson.