By SCOTT DEWING
Published: June 2009

Trebek: Alright everyone, here’s today’s Final Jeopardy! clue. IBM is developing a computer system personified by this name to compete on Jeopardy!

Answer: What is Watson?

I’m not making this up. I’m just not that clever and creative. Watson is a real computer system being developed by some very clever and creative computer scientists at IBM. It will compete on the popular game show Jeopardy! as early as next year. And Watson will probably win.

In order to win at Jeopardy!, however, my dear Watson is going to have to be much much more than just elementary. Watson’s going to have to be able to deal with puns, analogies, double entendres and relationships such as location and size then buzz in before its human opponents do. In order to do this, Watson is going to have to process natural language.

While computers have evolved to do many things better than humans, natural language processing is not one of them. That’s why computers never change their behavior when we curse at them. They can’t understand what we’re saying. To get them to do anything, in fact, requires programming them in computer programming languages that are far from anything that could be deemed “natural” by any normal human. That’s why most computer programmers seem a bit awkward to us—they’re actually supernatural beings whose brains are wired quite differently.

Watson is the brainchild of IBM’s DeepQA Project, which, according to IBM, “aims to illustrate how the wide and growing accessibility of natural language content and the integration and advancement of Natural Language Processing, Information Retrieval, Machine Learning, Knowledge Representation and Reasoning, and massively parallel computation can drive open-domain automatic Question Answering technology to a point where it clearly and consistently rivals the best human performance.”

Or to put it another way, IBM seeks to build a computer that is so good at answering questions that it will be able to kick a human’s ass in Jeopardy! If IBM succeeds at this, it will mark a major milestone in the quest to achieve what is arguably the Holy Grail of computer science: artificial intelligence.

We have not yet achieved artificial intelligence. While we have computers that can execute millions of instructions per seconds, we don’t yet have anything that can fully simulate human intelligence. We’ll know when we’ve achieved artificial intelligence.

Computers will be some of the smartest, if not the smartest, people we know. I use the term “people” loosely. Early AI systems won’t have a human-like form to them. That would be impractical and rely on major advances in robotics. Likely, they’ll be avatars or holograms of human forms, or dragons, or whatever you like. But behind all that will be a super computer with massively parallel processors firing away like neurons in a cerebral cortex.

Regardless of outward form, these AI systems will be so intelligent that you will be able to carry on an intelligent conversation with them. In fact, you may not even know you are talking to one of these systems or to a human. If you can no longer distinguish between human intelligence and machine intelligence, would that mean that the machines are able to think? This is the very question that the prominent computer scientist Alan Turing proposed in his 1950 landmark paper Computer Machinery and Intelligence. Because the term “thinking” is difficult to define and is permeated by all kinds of thorny philosophical issues, Turing chose to replace the term with a test. The test, which became famously known as the Turning test, goes as follows: a human judge engages in a natural language conversation with another human and with a machine. Each participant in the test is in isolation. If the human judge is unable to distinguish between the human and the machine, the machine has passed the test and, therefore, is a “thinking” machine. In fairness to the machine, Turing required that the natural language conversations during the test be limited to keyboard input and text output to a monitor.

While today’s computer systems are capable of speech recognition and voice emulation, they can’t pass the Turing test because they are not able to engage in natural language conversation with the kind of depth, breadth, and nuance we human’s would expect.

Watson’s debut on Jeopardy! will not be akin to a Turing test. Watson will not be engaged in free-form conversation. The challenge of winning Jeopardy! is a tough one even for humans so the folks at IBM have their work cut out for them. IBM has a legacy in the machine v. human face-off though. In 1997, IBM’s supercomputer Deep Blue beat world champion chess player Gary Kasparov in a chess match. The “Jeopardy! Challenge”, as it’s being called, is much greater than a chess game, however. Chess has limited pieces with limited moves and a limited (though vast) number of predictable moves. In Jeopardy!, contestants are asked a wide range of questions. Watson will have to first interpret the question, then draw from a vast database of material it has “learned” to come up with the correct answer.

Under the rules of the match that IBM has negotiated with the Jeopardy!, Watson will not have to emulate all human qualities. It will receive questions as electronic text. The human contestants will both see the text of each question and hear it spoken by the show’s host, Alex Trebek. Watson will then respond with a synthesized voice to answer questions and to choose categories.

IBM says that Watson could be ready for competition as early as next year and the media buzz is that it will face-off against Jeopardy! grand champion Ken Jennings who won 74 consecutive matches in 2004.

“I’m fascinated by the question of how an artificial player could do at a quiz game against top competition, whether or not I’m invited to play,” said Jennings in a recent interview. “I genuinely don’t know how it will shake out. But I’ve seen the Terminator movies, so I’m not 100 percent optimistic about mankind’s chances here.”