My interest in IBM’s Jeopardy-winning Watson was initially syntactical. It has taken decades to teach computers to understand and speak a human language almost as well as your average three-year-old. Add to that a timed challenge, and a question-shaped answer, and the linguistic complexity is mind-boggling. I had to know more.
It turns out, it’s bigger than that–and simpler. Winning has as much to do with knowing when you know, and when you don’t.
In a Q & A, someone asked about Watson’s advantage in reading the question and beginning millions of computations before a human contestant could hear the question. Watson’s team responded, “The clues are in English–Brad and Ken’s native language; not Watson’s. Watson must calculate its response in 2-3 seconds and determine if it’s confident enough to buzz in, because as you know, you lose money if you buzz in and respond incorrectly. This is a huge challenge, especially because humans tend to know what they know and know what they don’t know. Watson has to do thousands of calculations before it knows what it knows and what it doesn’t.” [Emphasis mine.] (See Reddit’s Q & A with the team here.)
In this game, as in life, what’s called “confidence”–the accurate assessment of relative knowledge or ignorance–matters at least as much as the answers themselves. Homer Hickam, author of Rocket Boys, said, “It is better to confess ignorance than provide it.” Turns out that knowing when we’re ignorant might be the better part of intelligence.
Quoted Article, Related Article, Related Video
- IBM Watson Research Team Answers Your Questions (reddit.com)
- John Maeda: Watson Is No Match for Humanity (huffingtonpost.com)
- Watson Research Team Reflects on this Accomplishment (ibm.com)
Text © Gwyn Nichols 2011
Writers' Resort LLC 