Application of Brunswikian Ideas to Human-Automation Interaction
Reflecting over the past year, I realize I have spent a great deal of time pondering and describing interactions: among theories, data, people, methods, technologies, and so on.
When Hammond and Stewart put out a call for nominations for Brunswik's papers to be included in the The Essential Brunswik, I nominated a paper not written by Brunswik. I suggested that James Gibson's review of Brunswik's 1956 book ("Survival in a World of Probable Objects") should be included, and was appropriately punished by the editors by being asked to write an accompanying chapter: "Gibson's Review of Brunswik."
You are not likely to learn anything new there about either of these great theorists; instead, my goal was to use the benefit of hindsight to try to separate the real from the merely apparent in their disagreements at the time. It is a great understatement to say that this was not as easy (or as brief) a task as I had first thought. I also wrote a second chapter for The Essential Brunswik, a piece on Brunswik's implications for "Human Factors."
Two of my students, Rich Strauss and Ellen Bass are currently engaged in dissertation research applying Brunswikian ideas to human-automation interaction issues. We are relying heavily on the extension of the lens model equation by Stewart and his colleagues to also include base-rate and regression-to-the-mean effects.
Strauss is evaluating the suitability of this model in a "coming to periscope depth" submarine task. Bass is using the approach in a triple-system arrangement to analyze the interaction between pilots, cockpit alerting automation, and the task environment. The goal here is to come up with some new methods for coupling people and alerting automation for the forthcoming "free-flight" environment: Currently, pilots are faced with overly conservative alerts (automation designers can be the subject of human error lawsuits as well!), yet they are often blamed when they perform the necessary task of separating false alarms from true alerts.
I am also working on a paper with my former student Ling Rothrock on inferring simple, noncompensatory judgment heuristics from behavioral data. In the dynamic command and control situation giving rise to this work, time pressure appeared to prompt our laboratory participants to develop simple if/and/not/or/then heuristics as coping strategies.
We did a lens model analysis of this data (a paper with another ex-student Ann Bisantz currently in review), but the results were less than encouraging: Participants reported simple heuristics during debriefing, and we found these rules to be consistent with their judgment data.
The technique Rothrock and I are now writing up consists of using a genetic algorithm for representing, generating, mutating, and crossing rules, and a multi-objective optimization algorithm to create rule fitness measures (maximizing a combination of rule completeness, parsimony, and concreteness). We applied the technique to inferring rules describing high and low performers, with very encouraging results.
Two additional writing projects are currently underway (also with Strauss). The first is a chapter on an ecological approach to human error in health care, for a forthcoming volume-Human Factors Interventions for the Health Care of Older Adults (Erlbaum); we are grateful in this regard to Gérard Chasseigne for sending us his aging-related papers. The second is a review (for Applied Cognitive Psychology) of the recent volume (edited by Juslin and Montgomery): Judgment and Decision Making: Neo-Brunswikian and Process-Tracing Approaches.