Executive Summary




This report details a complete, beginning-to-end Cognitive Systems
Engineering (CSE) project tackling the challenges of conducting intelligence
analysis under the condition of data overload. We first reviewed and
synthesized the research base on data overload from multiple complex,
high-consequence settings like nuclear power generation. The product of
this activity was a diagnosis of the challenges of dealing with data
overload in general. Then, leveraging this research base and previous
experience in conducting Cognitive Task Analysis (CTA), we conducted a study
to identify the aspects of this research base that applied to intelligence
analysis as well as unique challenges. We observed expert intelligence
analysts conducting an analysis on a selected unclassified scenario, the
1996 Ariane 501 rocket launch failure, with a baseline set of tools that
supported keyword search, browsing, and word processing in an
investigator-constructed database. From this study, we identified
challenging tasks in intelligence analysis that leave analysts vulnerable to
making inaccurate statements in briefings when they are working in a new
topic area and are under short deadline constraints. In parallel, we
identified limitations of the baseline tools in addressing these
vulnerabilities that pointed to ideas for new design directions. In
addition, the study findings were translated into objective criteria for
evaluating the usefulness of any effort aimed at reducing data overload.

In the final phase of the project, we shifted from an emphasis on problem
definition to an emphasis on developing modular design concepts, or "design
seeds," that could be incorporated into both ongoing and future design
efforts. The design seeds were instantiated as animated fly-through
mockups, or "Animocks," so that feasibility of certain usability
considerations, such as the ability to display data in parallel in an easily
interpretable form, could be explored without being forced into committing
to a particular design. With Animocks, commitment to a particular hardware
infrastructure, visualization instantiation, or combination of design seeds
is lessened and there is greater flexibility to incorporate feedback about
the usefulness of a design concept in addressing data overload. In
addition, the design seeds are conceptually modular and based in challenging
scenarios, which enables generalization of concepts across design projects
and domains. With this strategy, we can better address one of the primary
challenges of a research and development program, which is to develop
research bases that can be translated into fieldable systems in multiple
settings to prevent continuously engaging in individual, one-off design
endeavors without learning how to improve systems over time. We believe
that complementarity of research and design efforts and accompanying
cross-stimulation is the main characteristic of effective Cognitive Systems
Engineering R&D programs.