Carl Westin

5-2 Automation transparency research 91 5-2-3 Increasing transparency by means of meta-information Importantly, the notion of transparency shifts the focus from presenting raw state- related information, reflecting the current situation, to presenting meta-information that better reflects the human-machine-ecology relationship and guides action and decision-making. Meta-information reflects information qualifiers underlying sys- tem reasoning that facilitate sense-making of information provided. 220, 221 As such, transparency can be considered a measure of the system’s openness in meta- information communicated that entails what the automation is currently doing, which information is being used, how it is being processed, and when and how it is provided. Specifically for decision aids, there is a need to provide meta-information about the system’s rationale underlying its judgments and problem-solving, the cri- teria considered, and uncertainty factored. 191, 218 There are, however, no transparency guidelines that explicitly prescribe which, how much, and in what ways meta-information should be presented. Simply pro- viding more meta-information is undesirable since it can easily overload the op- erator’s cognitive ability to detect and process the information in real time. 222, 223 Such overload is typically characterized by clutter, increased workload, or difficul- ties in finding the significant data, which can result in degraded monitoring and signal/change detection, delayed visual search, confusion, increased confidence in wrong judgments, and increased memory load. 199, 224 As such, there is an apparent need for transparency calibration, to avoid human-automation cooperation break- downs caused by excessive or insufficient access to meta-information. The need for balancing meta-information necessity with meta-information overload in visu- alizations and display design (i.e., the human-computer interface bottleneck 225 ), is especially critical for time critical and information rich environments. 191, 221 5-2-4 Ecological interfaces for visualizing meta-information Research suggests that the ecological interface design (EID) approach lends itself well for facilitating transparency, particularly in the context of supervision and con- trol problem tasks, by means of visualizing relevant meta-information without in- curring drawbacks of information overload. The EID approach strives to identify and visualize relevant relationships and constraints in the work environment that af- ford constructive problem-solving and action. 122 As such, it provides a structured approach for identifying and visualizing meta-information that reflects meaningful relationships in the problem space of interest, rather than simply notifying the op- erator of system health, status or mode changes, which is typical for automation in most current supervisory wok domains. 226 EID maps particularly well onto the domain transparency suggested by Brown, in that it explores the relationship be- tween agents (e.g., human and automation) and the work ecology. 204 Kilgore and