IBM's New Business Plan: Beating You at Jeopardy

(MoneyWatch)  Back in 1997 IBM's (IBM) super computer, Deep Blue, beat human chess master Gary Kasparov. The event was a PR triumph for the company, but didn't lead to any new products worth selling. Now IBM has designed a computer named Watson to compete with humans at Jeopardy. And this time the machine's natural language skills hold real promise for IBM's future business.

According to a new piece by Clive Thompson, IBM has invested three years and millions of dollars in the project, and plans to pit the computer against human Jeopardy champions as early as this fall. Watson not only understands human language well enough to answer straightforward questions, it can even handle the elusive puns familiar to fans of the long running game show. "The name of this hat is elementary, my dear contestant?" Answer: What is a deerstalker, the preferred hat of Sherlock Holmes.

The obstacle for this kind of artificial intelligence has always been the labor intensive process of educating the computer. Previously, humans had to enter all the knowledge into a computer by hand, and create the links within this database that told the machine which items were connected and why. It was possible to make a computer that could answer every question in Jeopardy, but it had to be spoon fed all the facts first.

According to Thompson, two things changed in the last decade that enabled I.B.M. to get beyond this hurdle. Computers became powerful enough that they could learn, statistically speaking, how huge sets of words and terms were related to each other. At the same time, massive amounts of information were digitized, meaning there was plenty for the computers to study without a human having to enter it manually.

Watson's set of skills matches up well against the current paradigm, which is search. When users ask Google (GOOG) a question, they are usually giving it a data point, hoping it will return results that lead them towards the information they're seeking. Watson, by comparison, can give a direct, factual answer. Image a law firm taking a big case with a tight deadline. A team could scan mountains of evidence into the machine, allowing it to quickly and easily answer questions that would have taken hours of human research.

John Kelly, head of IBM's research labs, envisions a version of this machine that processes the torrent of new medical research published on the web, allowing doctors the most accurate, up to date answers to their questions. "Computers need to go from just being back-office calculating machines to improving the intelligence of people making decisions." As cloud services lessen the demand for IBM's hardware, creating high value services like Watson will be critical to the firm's future growth.

Image from David Ray