Many readers will be familiar with the Jeopardy! television show, which is the most popular quiz show in North America. One of the present bloggers confesses to watching it almost every weekday evening when not on travel (and hardly any other television program). The other blogger is also a keen armchair contestant.
In this show, clues are presented on a video screen to the three contestants, and after Canadian host Alex Trebek completes reading the clue, contestants must first ring in with a handheld button. The first contestant to ring in then has five seconds to provide the answer, which must be phrased in the form of a question. If he/she is correct, then he/she wins the amount of money attached to the question, which varies from $200 to $1000 in the first round and $400 to $2000 in the second round; if wrong, this same amount is deducted from his/her score.
Recently IBM presented the Jeopardy! show with the opportunity to compete on the show with a computer system they have been developing that answers questions presented to it in English. The system is based on IBM’s “BlueGene” supercomputer, which at present costs (in a configuration sufficient to this task) well over USD$1,000,000. After completion of the project, IBM intends to market this system, say as a tool for user support personnel. Potential uses range from medicine to online government. Some applications, particularly medical applications, also entail various ethical problems.
This development underscores the tremendous advances that have occurred recently in the field of artificial intelligence, or, as it is more commonly now known, “machine learning.” The key breakthrough in the field occurred about 15 years ago, when researchers switched to probabilistic (i.e., Bayseian) schemes instead of the strict rule-based schemes that had been attempted in the past. Researchers began using statistical schemes to analyze large volumes of documents, many of them available on the Internet. Their algorithms process any topic and automatically “learn” what words are statistically most closely associated with it. These ideas parallel developments in neurobiology and even neurophilosophy (see [Churchland]). Unlike progress in computer chess, Watson may even tell us something about human cognition.
IBM’s “Watson” system is a premier example of this new technology. IBM project manager David Ferrucci is himself an artificial-intelligence researcher specializing in question-answering systems. When Ferrucci approached IBM’s management about taking on Jeopardy!, initially he was told to forget it. But he persisted and was finally given the go-ahead to start a development effort.
The initial results were “dismal,” according to Ferrucci, but after a year, Watson’s performance had moved halfway up to the “winner’s cloud,” namely the level of effectiveness achieved by many Jeopardy! champions (some of whom have been invited to come to IBM’s research lab to compete against the system, as part of the development effort). In 2008, when Watson had achieved a level roughly comparable with the “winner’s cloud,” IBM approached Harry Friedman, the executive producer of Jeopardy! with the possibility of having the “Watson” system compete on the air with some former Jeopardy champions. The competition is scheduled for Fall 2010.
Here are some examples of some of the question that the Watson system is currently capable of handling (faster than human contestants):
- [Category: “Postcards From the Edge”]: “Toured the Burj in this U.A.E. city. They say it’s the tallest tower in the world; looked over the ledge and lost my lunch.” [Answer: “What is Dubai?”]
- “He was presidentially pardoned on Sept. 8, 1974.” [Answer: “Who is Nixon?”]
- “In 1594 he took a job as a tax collector in Andalusia.” [Answer: “Who is Cervantes?”]
- “Thousands of prisoners in the Philippines re-enacted the moves of the video of this Michael Jackson hit.” [Answer: “What is ‘Thriller’?”]
- [Category: “All-Eddie Before & After”]: “A ‘Green Acres’ star goes existential (& French) as the author of ‘The Fall.’” [Answer: “Who is Eddie Albert Camus?”]
But many questions that are simple for good human competitors still flummox Watson. Additional information may be found in an excellent, in-depth article that appeared recently in the New York Times [Thompson2010], from which the information above was summarized.
References:
- [Churchland2008] Paul Churchland, Neurophilosophy at Work, Cambridge University Press, UK, 2007.
- [Thompson2010] Clive Thompson, “What Is I.B.M.’s Watson?”, New York Times, 14 Jun 2010, available at Online article.