At TiEcon 2011 keynote session (www.tiecon.org), Dr. David Ferrucci discussed the phenomenal challenges of building Watson, an artificial intelligence computer system that competed with human subjects in the quiz show Jeopardy and consistently outperformed the human competitors. Compared to the Deep Blue, the chess-playing computer, developed in 1997, the Watson Deep QA project was infinitely more challenging. Chess is finite, mathematically well defined with limited number of moves. Computer can be really good at mathematically challenges and can find information very quickly. Jeopardy however is a significantly more complex challenge. Not only Watson had to navigate through frequently convoluted human speech which is infinitely complex, contextual, idiomatic and full of puns and expressions that alter the meaning of the words, but additionally Watson had to demonstrate the level of confidence it had in the answer it came up with and it had to do all that in a matter of seconds. Watson got better at the game and finally came up with answers at the average speed of 3 seconds. Watson had access to over 200 million pages of structured and unstructured content consuming several terabytes of storage. While Watson outperformed its human opponents most of the time, it had trouble responding to several categories. Ferruci shared several examples where Watson was completely off the mark and that gave impressive evidence of how infinitely complex human speech is, where meaning of the same words is often determined or altered based on the order of the words, the sentence structure, placement of commas and exclamation marks.
At TiEcon, this was a breakthrough keynote, considering that most keynotes focus on the business aspect. Ferrucci’s presentation was technical and fascinating and kept the audience riveted, soaking up the impressive information.
Ferrucci discussed the implications of this project in several other domains, with the members of the press. It is possible that Watson-based technology, or services built upon it, might appear and might have implications in a wide range of industries. Application in disease and healthcare diagnosis for instance, can be an invaluable tool to aid the doctors rarely have enough time with each patient to consider all possible data. Also unlike many previous projects, Watson shows considerable similarity of thinking with the human mind. For instance, Watson is programmed for uncertainty. If is often not completely confident then Watson gives answers with varying levels of confidence, making it easier to determine how much weight to give to the answer given. Truly fascinating project.