The Cognitive Sciences by Sobel Carolyn P.; Li Paul; & Paul Li

Author:Sobel, Carolyn P.; Li, Paul; & Paul Li
Language: eng
Format: epub
Publisher: SAGE Publications
Published: 2017-11-13T14:33:37.865278+00:00

Watson

Jumping ahead to our own time, the second decade of the 21st century, we find real progress in the attempt to endow a machine with human-like intelligence. As mentioned earlier, IBM's project gave two human superstars of the game show Jeopardy! a run for their money—in fact, Watson went home the winner. How did this super machine accomplish such a feat?

Those readers who have watched Jeopardy! are familiar with the game show's format: contestants are given answers in a number of categories and must come up with the correct questions. Furthermore, the categories are not limited to a particular domain (e.g. sports, or recent American fiction). “The key challenge for us was the open-domain question answering…” (Ferrucci, 2011). To complicate matters even more, the question-answer pairs are not straightforward: they involve—at the least—puns, irony, subtleties of meaning, and nonobvious associations. And all of this relies on a huge fund of stored information about all manner of subjects in the human world. It is, as you know if you have attempted to play along with the contestants, a tough task. But a normal human brain, small enough to fit inside a human skull and weighing all of three pounds, contains within it all the mechanisms necessary to accomplish it.

Researchers at IBM, led by principal investigator Dr. David Ferrucci, began in 2007 to design the Watson project, using parallel distributed computing. That is, they employed many different computers operating at the same time, and sharing memory— a lot of memory, a lot of information. The system they required was one that could arrive at and extract the information needed at a faster rate than humans can, while dealing simultaneously with the nuances of language, with slang, regionalisms, and ambiguities such as the intended meaning of the claim: “Last night I shot an elephant in my pajamas.” The ambiguity in this sentence is demonstrated by the potential response “What was the elephant doing in your pajamas?” We are amused by such a response, because we know instantly who's wearing the pajamas and don't expend any thought on the possibility that it is the elephant. Watson would have to be equipped not only to perform all these tasks but also to weigh probabilities in order to judge, for example, who was in fact most likely to be wearing the pajamas. Further, there is the decision about how much to bet on each question; Watson would have to weigh his confidence as well. And press the button (or perform his version of pressing the button) ahead of both the humans competing against him.

Impressed as we are with a machine that is capable of natural language processing, we note that speech recognition, the ability to process spoken language as it is being delivered by the speaker, is missing from Watson's repertoire. When asked why this aspect of human intelligence was omitted from Watson's design, Dr. Ferrucci pointed to the fact that the main effort in the project was directed toward open-domain question-answering, and explained that “…. including speech-recognition challenges would have muddied the experiment.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.