One of many guiding factors in determining what technology to apply is the nature of the information we have available on which to act. Consideration must be given to the differences between the types and quality of information available to assess a given situation. The process of assessing and accounting for the quality of available data requires different approaches depending on both the quantity and type of uncertainty. For example, the reliability of diverse sources of human intelligence is typically harder to characterize than other forms of information such as electronically collected sensor data. Each of these characteristics leads us towards a different solution based on the technology that is best suited to acting on a particular kind of information. Many complex real world problems can not be effectively solved using a single approach in isolation, but require a combination of technologies and models.
Heterogeneous reasoning systems require the ability to exchange uncertain information between software elements based on different technologies, with different uncertainty models and which were potentially developed independently. While there has been a significant amount of research involving modeling uncertainty it typically focuses on issues related to a single uncertainty model or reasoning technology. Although it is an increasingly important problem, relatively little progress has been made with regards to fusing uncertain information across reasoning technologies.
Accounting for uncertainty in intelligence information is complicated for several reasons. The first is due to the numerous different models used by the various reasoning technologies and the lack of a widely accepted grand unified theory of uncertainty. Second, is the fact that certain technologies, such as neural networks, although appropriate for addressing some forms of processing involving uncertainty, are not typically structured to produce values representing the type and quantity of uncertainty present in their output. Finally, the nature of the input information from sources such as human intelligence presents especially difficult challenges relating to quantifying the uncertainty present.