By Mike McPhee
The game show “Jeopardy!” featured an unprecedented contestant this week: Watson, an IBM-created supercomputer with highly sophisticated software, competed against “Jeopardy!” superstars Ken Jennings and Brad Rutter in a trivia test of man versus machine. For a machine that was designed to win at trivia competitions, Watson’s performance generated more questions than answers.
The idea of a computer that plays “Jeopardy!” has drawn media attention back to the development of artificial intelligence, something that has been quietly worked on by computer scientists and engineers for years. How soon will it be before machines become smarter than humans? Is Watson the first step in that direction?
Watson’s “Jeopardy!” performance does not tell us much about that bigger picture. Its most strikingly “human” trait is that it is not perfect. Despite IBM’s four years of research and development, there were still some shortcomings to Watson’s “thought” processes that became obvious over the course of the three-day “Jeopardy!” special.
For example, on the second day of the showing, Watson got 24 of the 25 questions in Double Jeopardy right. When prompted with the Final Jeopardy question, “U.S. Cities: Its largest airport is named for a WWII hero. Its second largest is named after a WWII battle,” Watson answered “Toronto.” The studio audience was shocked that this supercomputer came up with an answer that was so obviously incorrect. Watson also had an odd tendency to end all of its Daily Double wagers with atypical amounts instead of a round multiple of $100.
Watson did win the competition on “Jeopardy!” In fact, it earned more money than both Jennings and Rutter combined. But compare its success rate to other electronic devices you own. It wouldn’t be acceptable if your cell phone only answered 24 out of 25 of your phone calls, or if sometimes your computer misinterpreted what you clicked on screen. Watson’s algorithm needs to be further developed before it can be branched out to other applications.
Those potential applications are quite interesting. Watson’s technology could be used to replace jobs that are currently held by humans – for example, answering questions when you call a help number for a broken computer, or even conducting telemarketing more intelligently.
In addition, Watson is very good at taking vast quantities of data as an input, receiving a question and formulating a useful output; in this way, it could function as a more advanced form of a search engine for important data-intensive fields like medical research.
Before this can happen, work still needs to be done. The computers that Watson is built out of fill an entire room, draw an enormous amount of power and certainly are not cheap. It is definitely not cost-effective to use a Watson to replace a human at a job yet.
The real take-away message for people who watched the “Jeopardy!” special is not that we should “welcome our new computer overlords,” as Ken wrote during his Final Jeopardy answer on Wednesday. We should keep an eye on the developing technology of “deep thought” computer programs as enhancements to human intelligence, not replacements for it.