Watson Surges Past Its Human Opponents on Jeopardy!

Four years ago, a team of researchers at IBM starting building a machine which they named Watson, after IBM founder Thomas J. Watson. Watson is the latest project at IBM Research Center to try to solve a "Grand Challenge". These are fundamental technological breakthroughs that achieve something previously considered "impossible". One of the first of these was in the late 70s, when IBM helped NASA be able to place a man on the moon. This time, they wanted to make a machine that could understand human language. To create a challenge that would exercise Watson's abilities, they set their sights on creating a Jeopardy champion! This game show provides the ultimate challenge because the game's clues involve analyzing subtle meaning, irony, riddles, and other complexities in which humans excel and computers traditionally do not.

The match between Watson and its human opponents is now airing, with the final episode tonight. Watson is competing against the top two players of all time, Ken Jennings and Brad Rutter. Ken Jennings won a record 74 straight games in 2004-2005 and Brad Rutter has won a record $3.2 million on Jeopardy! After the first round, Watson and Brad were tied with $5000 each and Ken had $2000. Last night, the second round began with Ken choosing the first question, but Watson quickly took over and built a seemingly insurmountable lead. The round ended with $35,734 for Watson, Rutter at $10,400 and Jennings with $4,800. The competitors will square off again tonight for a final Jeopardy battle.

OK, so what's the big deal? Is it really surprising that a computer with 13 terabytes of memory can answer more questions faster than a human? Actually, YES! Since the early 1970s, I have spent time reading on the attempts at machine intelligence. The main goal was to demonstrate the potential for machines that could replicate human thinking by competing with humans in games of mental skill; chess was the favorite game in the early years. There were many researchers in the 70s that thought a computer would easily be the world champion chess player by the end of the decade (After all, technology got us to the moon, didn't it?). They were very wrong - it took almost another twenty years before IBM's Deep Blue became competitive with human grandmasters in May 1997, when it beat world champion Gary Kasparov is a match. Deep Blue was a special-purpose computer with hardware designed specifically to play chess; it used heuristics from the top chess players to decide on likely good moves and then used its brute strength to search these lines of play for the best results.

Computer hardware and software technology had advance enough by 1997 for a machine to approximate the best humans could offer in a very limited environment. But this wasn't going to make a significant difference in creating machines that could interact with humans in an "intelligent" manner. In 1950, Alan Turing proposed a test to determine if machines could be considered intelligent; it is now known as the Turing test. It says that if a human "talking" to the machine cannot tell that it isn't a human; then the machine must be considered intelligent. The test is based upon understanding human language, with all of its idioms, puns, cultural references, etc. This is a much more difficult challenge since it involves much of what makes us human, at least as a thinking being.

Now, with Watson showing clear ability to compete with the best human opponents on Jeopardy!, we have achieved a significant milestone in making machines that can truly work side by side with humans. The achievement is not remarkable for Watson's ability to recall vast amounts of trivia from its databanks; what is remarkable is the ability to "understand" the question and to consider what the best response would be. During the show, you can see Watson's top three responses and what confidence it has in each. If a possible answer has a high enough confidence factor, Watson will give this response; otherwise, Watson will usually stay quiet. This is natural language processing and has been the biggest hurdle to making computers truly useful.

We can't say that Watson can think for itself, but at least we have to admit that Watson is able to understand human language. And this is a huge leap in technology that will lead to significant advancements in many fields, as humans can more easily access the power of machines to access and search through our ever-increasing wealth of data and turn it into usable information. Watch today and see history being made as machines achieve a new milestone!

REFERENCES

http://www-943.ibm.com/innovation/us/watson/

http://www.pcworld.com/article/219778/ibms_watson_dominates_jeopardy_competition_on_day_2.html

http://www.research.ibm.com/deepblue/

http://en.wikipedia.org/wiki/Turing_test