Computers play chess? So what. Shogi? Whatever. Jeopardy!? Now we’re talking.
And while my opening line might seem sarcastic, it really isn’t. That’s because Jeopardy!, while easily dismissed as TV froth, is a genuine challenge to computers.
That’s because of the way the show works: for those not familiar, the game involves host Alex Trebek supplying an answer, and then the contestant attempting to figure out the correct question. Success requires a range of skills that don’t necessarily fit with what computers are best at. First of all, the contestant has to decipher the answer, often combating puns and other linguistic challenges, to decide which parts are relevant to the task.
Then, as opposed to finding the single correct answer to a question, the contestant needs to figure out every possible question that could have the answer, then decide which is most likely to be the one the quiz setters were thinking of. That requires a combination of reasoning, judgment, and considering several ideas near-simultaneously, a set of skills to which humans are better suited.
That’s the theory at least, but IBM believes its machine Watson (as in Thomas, the IBM founder) could compete with the best. As a result, the computer is taking on Ken Jennings and Brad Rutter, two of the all time greats at playing the show.
The rules of the challenge say that Watson can only work from data stored on the machine: it’s not allowed to network with other computers or access the Internet. That said, Watson does take up the equivalent space of 10 refrigerators: for practical reasons it’s represented at the podium by a flat-screen monitor. It’s been set up to analyze possible answers and rank their likelihood, only buzzing in if it has an answer with a set level of certainty.
In preparation for the contest, Watson has played more than 50 games against former champions, but this week’s test is the first time a result has been made public. The final score saw Jennings on $3,400, Rutter on $1,200 and Watson taking glory with $4,400.
Although there’s a million dollar prize at stake in the challenge, IBM will donate any winning to charity. It says the aim of developing Watson isn’t TV glory, but rather to have a practical way of coming up with a technology that reverses the usual question-answer format. For example, it could lead to computer systems that do a better job of diagnosing a medical condition from a list of symptoms, or performing a similar role in tech support. It’s also possible it could lead to search engines that are more able to understand natural language questions.