Lessons Learned in Building a Knowledge Base for a Cognitive Assistant

We present our experience in developing a knowledge base (KB) in the context of an intelligent personal assistant called CALO. The goal of building the intelligent assistant was to aid a knowledge worker in managing information on their desktop, enable the worker in managing time commitments and performing tasks, and assisting during meetings. While we leveraged state-of-the-art best practices in constructing a KB for CALO, we faced unique challenges as the system evolved from its initial conception to deployment and raised the following questions: How well do the traditional KB development methodologies and upper ontologies work in a context in which the goal is a deployed system rather than ontology research? What ontology support is needed for an information management application that leverages statistical machine learning? How can a KB support cognitive task assistance? The primary contribution of our work is in giving a detailed account of successes and failures in addressing these questions and the general lessons that can be drawn for future research and development of knowledge base systems.