The Quiz Contestant Had No Internet

1. A different challenge

IBM researchers built Watson to answer Jeopardy! clues. The task required handling language, evaluating possible answers and deciding when to buzz. The clues relied on language, wordplay and factual knowledge. Finding the right response was different from merely retrieving a matching sentence.

Fast arithmetic alone would not win this programme.

2. An answer needs a degree of confidence

A quiz system can have several plausible responses to a clue. Choosing between them matters when a wrong answer carries a penalty and the time to decide is limited. The technology therefore had to do more than retrieve information: it needed a way to decide when a candidate answer was worth committing to publicly.

Knowing something was only half the buzzer problem.

3. The preparation

Watson processed large collections of written material before the competition. It played without an internet connection, using its prepared system and information. A useful system needed to rank competing answers. Confidence mattered because a wrong response could also lose money in the game.

The contestant could not quietly search the web.

4. The opponents

In February 2011, Watson faced Ken Jennings and Brad Rutter. Both humans brought outstanding records and experience with the show’s clues. The contest placed the system under public time pressure. Its preparation had to translate into decisions during a real broadcast event.

The computer had excellent company at the podium.

5. An imperfect player

Watson sometimes made obvious-looking mistakes, including a weak response in a U.S. cities category. Strong performance still included conspicuous failures. An incorrect answer revealed limits alongside the successful responses. A public demonstration could show both capability and error in the same competition.

A machine could dominate and still look confused.

6. A public failure is visible too

A competition records errors as well as correct responses. That made the demonstration useful evidence of both the system's capabilities and its limits. The audience did not need to imagine a perfectly reliable machine in order to recognize an important result in the particular game being played.

The score could contain success and a very conspicuous mistake.

7. The outcome

Watson won the televised contest. Its achievement advanced question-answering research while leaving the distinction between scoring answers and understanding the world open. Winning the game established a specific technical achievement. It did not mean every later application of the technology would perform equally well.

The scoreboard gave an answer to one specific question.

Sources

• IBM — Watson, Jeopardy! champion