Real world knowledge gathering and investigative tasks are very complex because the problem-solving context is constantly evolving, and the data may be incomplete, unreliable and/or conflicting. We are developing a mixed-initiative reasoning agent that will assist homeland security analysts to choose from and reason about enormous databases of text, imagery, video and webcast. This agent will leverage an AI blackboard system and resource-bounded control mechanisms to support hypothesis tracking and validation in a highly uncertain environment. Interactive visualizations are being used to enable analysts to gather and sift large amounts of evidence and to collaborate with and, where necessary, to control the agent.
The figure shows the TiBor structure. In general, multiple databases will be accessed, each with their own analysis tools. Using a naïve-Bayes classifier, TiBor will decide which databases to access, what products to retrieve, and what level of analysis to perform. The Markov-based decision control then determines choices of visualizations. The blackboard system provides overall supervision and is being designed multilevel system for complex problem solving that may use a variety of approaches. Through the visual interface, the user can intercede in the process at any point.
This work is led by Anita Raja.