All Global News on One Page

What IBMs Jeopardy Machine Can Teach Us Humility

Ken Jennings and Brad Rutter, Jeopardy’s two greatest champions, have brains packed with facts. In one Final Jeopardy, Rutter actually recalled that President James Garfield’s wife was named Lucretia. And he deduced from this that the Mediterranean island that shared a nickname with a 19th century First Lady was “Crete.”
You might think that IBM’s Jeopardy computer, which is taking on Rutter and Jennings next month, would “know” billions of facts. But in truth, Watson is sure of nothing. It treats each statement it comes across — pillows are soft, water is a liquid, etc. — as an assertion. In the two- to three-second process of responding to a Jeopardy clue, it builds varying levels of confidence in these apparent facts. Through the course of this painstaking work the Jeopardy computer builds up statistically based beliefs. But it’s never 100% sure of anything. It doesn’t “know.” There can always be exceptions.
An example: David Ferrucci, the chief scientist on the Jeopardy project, was telling me as I researched my book that his team was tempted to teach Watson some basic facts, including the months of the year. Months didn’t change. Why shouldn’t Watson simply know them? But then they came across a Jeopardy clue about the Muslim “holy month.” If Watson had been convinced that there were only 12 possible answers, from January to December, it would have missed “What is Ramadan?”
One reason humans are faster than Watson at answering certain questions is that our minds are full of “facts” that appear to require no further research. It makes thinking easy, so easy in fact, that we’re tempted to expand our universe of facts.

read full news from www.huffingtonpost.com