Integrated Cognitive-neuroscience Architectures for Understanding Sensemaking (ICArUS): A Computational Basis for ICArUS Challenge Problem Design.

摘要

The IARPA (Intelligence Advanced Research Projects Activity) program ICArUS (Integrated Cognitive-neuroscience Architectures for Understanding Sensemaking) requires "sensemaking" challenge problems as described in the BAA (Broad Agency Announcement, 2010). These problems are needed for conducting experiments with human participants, and for assessing neural-computational models of human performance in Test and Evaluation (T&E). Previously, researchers have described the cognitive challenges of sensemaking only informally using conceptual notions like "framing" and "re-framing", which are not sufficient to support T&E in accordance with the BAA. To overcome this limitation, a Bayesian-computational model of sensemaking was developed by dissecting a prototypical example of intelligence analysis, and by defining eight discrete steps in a processing cycle dubbed the Octaloop. This document details the Octaloop model and describes how it was used as a computational basis for design of ICArUS Phase 1 and Phase 2 challenge problems. Other uses of the Octaloop beyond ICArUS experiments are also identified. These uses include structured analytic techniques, training of critical thinking skills, and automated tools for improving the effectiveness of intelligence analysis.