CDI-Type II: Exploiting Collective Human Knowledge to Understand and Evolve Complex Networked Systems

Massachusetts Institute Of Technology, Cambridge MA

Investigators

Abstract

Large networked computer systems are some of the most complex and sophisticated engineered systems ever constructed. Because of this complexity, our existing computing environment is unreliable, insecure, and difficult to use. One intriguing aspect of this situation is that most problems people encounter in practice are not new --- they have previously been seen and solved by others. Developing techniques that enable the automated analysis of this rich knowledge source is therefore a key step in advancing our understanding of this important class of engineered artifacts. The research effort focuses on developing techniques that automatically process existing repositories of human interactions regarding previously encountered problems and solutions to isolate potential solutions to the problem. It will then use multiple virtual machines to evaluate each of the potential solutions, then use this evaluation to automatically select an candidate. The expected benefits of this approach are three fold. First, developing methods that jointly analyze documents in natural language and formally structured computer information can lead to new insights and a deeper understanding of these important phenomena. Second, this research will yield a new class of techniques for reasoning about computer systems in the presence of uncertainty by leveraging collective knowledge of a user community. Finally, by automatically finding solutions to system problems, these methods hold out the promise of substantially reducing the cost and risks associated with using computers to perform many of the key activities in our society.