Manually tracking all the published science relevant to a scientist's research is an impossible task, but it is predicted that computers capable of generating many helpful hypotheses with little human input will emerge within a decade. New computational tools can broaden the range of concepts and relations used for producing automated hypotheses by tapping a greater portion of the massive archive of published science, and by synthesizing new higher- and lower-order concepts and relations from the existing body of knowledge. Researchers can productively trim the multitude of low-quality hypotheses generated by a bigger pool of concepts and relations by employing a selection process that draws on insights into the social, cultural, and cognitive creation of science.
The number of possible hypotheses could be vastly enlarged if researchers could computationally map concepts across different scientific communities' distinct languages. This would flag parallels in theories from different domains, as well as changes in meaning with time and multiple meanings. These distinctions could be computationally mined to uncover unique conceptual connections. By assigning priority to hypotheses containing concepts spanning existing scientific theories, cultures, and languages, investigators could profitably concentrate on the most novel.
View Full Article
Abstracts Copyright © 2010 Information Inc., Bethesda, Maryland, USA