IBM’s Watson: Hope or Hype?
Earlier this week, many millions of people tuned their TVs to Jeopardy and found a non-human player competing with previous human champions Ken Jennings (74 wins) and Brad Rutter (20 wins). The non-human (hereafter referred to as the machine, or its proper name, Watson), not only competed but wiped the floor with its puny human competition. But was it really a competition? And what does it really mean?
First, that name: Watson. No, it does not represent Sherlock Holmes’s Dr. Watson, but rather IBM founder Thomas J. Watson, Sr. But who outside of IBM remembers—or cares? Or that Junior was president, chairman, then chairman of the board from 1952 to 1979? Sorry IBM, but Sherlock’s Watson is still better known (consider the 2009 Sherlock Holmes movie).
Second, the machine. My first question: How was it getting host Alex Trebek’s questions? Or answers, because, as you no doubt know, Jeopardy’s questions are in the form of an answer—and the player’s responses must be in the form of a question. And the answer to my question is that the machine was “fed” the question in text form at the same time Alex spoke it to the humans. Text? What the hell (and more in a moment)? But the larger point here is that they made it look like Watson heard the question. They could have shown the question to viewers as he read it, but then people would have been suspicious about how it could appear as text if Alex just selected it.
Third, about that text. One of IBM’s big brags about Watson is that it does natural language processing. Or does it? I know natural language processing is a big deal, but it has to be more than text. Hell, I did a version with text back in 1968! Processing speech, as you may have noticed, is way more difficult. (As in, could a machine understand the colloquial “way more”?) So if Watson can’t “hear” Alex’s questions, why should we take it seriously?
Speaking of Alex’s questions, who taught Watson how to answer a Jeopardy question (actually an answer) in the form of a question? Did it hear the rules as spoken by Alex and thereby understand how to answer—as did the puny humans? And more. How did Watson acquire its knowledge base? Did they instruct it to search and associate knowledge (how and for how long?) or did they simply give it a knowledge database? For example, did it learn anything new from the correct answers of its competitors? I doubt it. Score at least one for the puny humans.
And my last question: How would Ken Jennings have done if he had the time and resources—and the staff—that were put into IBM’s Watson? Sure he (and you or I) might look puny next to Watson’s capabilities, but then Jeopardy viewers did not see what was behind the machine’s curtain. If you want to know what’s behind Ken, read his book, Brainiac: Adventures in the Curious, Competitive, Compulsive World of Trivia Buffs.
Now for the answers. Are we in jeopardy from IBM’s Watson? No, despite its apparent cleverness. Are we in danger from The Machine? An interesting question (asked in the 1957 movie, Desk Set, starring Tracy and Hepburn). My answer. Well, I could say you’ll have to read the book (look back at the top of this post: Triumph of the Machine)—but I’m still writing it. Anyway, my short answer is that the machine has already won. If you want to know why now, read this post and the future book’s web pages