Using an APEX Model to Anticipate Human Error: Analysis of a GPS Navigational Aid.

Mark Van Selst and Michael Freed

NASA Ames Research Center
MS 262-2 Moffett Field CA 94035 USA
{mvselst,mfreed}@mail.arc.nasa.gov

The interface development process can be dramatically improved by predicting design-facilitated human error at an early stage in the design process. The approach we advocate is to SIMULATE the behavior of a human agent carrying out tasks with a well-specified user interface, ANALYZE the simulation for instances of human error, and then REFINE the interface or protocol to minimize predicted error. This approach, incorporated into the APEX modeling architecture (Freed & Remington, 1997), differs from past approaches to human simulation in its emphasis on error rather than (e.g.) learning rate or speed of response. The APEX model consists of two major components: (1) a powerful action selection component capable of simulating behavior in complex, multitasking environments; and (2) a resource architecture which constrains cognitive, perceptual, and motor capabilities to within empirically demonstrated limits. The model successfully mimics human errors (e.g., in air traffic control) arising from interactions between limited human resources and elements of the computer interface whose design fails to anticipate those limits. Our current work employs this approach to analyze the design of a hand-held Global Positioning System (GPS) device used for tactical and navigational decisions in small yacht racing. This analysis demonstrates how human system modeling can be an effective design aid, helping to accelerate the process of refining a product (or procedure).

A simple example: While engaged in a variety of demanding sailing tasks, a helmsman will occasionally take time out to set the boat's GPS device to target the next waypoint on a predefined course. On a "triangle-and-sausage" (Olympic) course [figure], the user sometimes mistakenly selects the wrong waypoint and thus mistakenly starts to repeat a segment of the course.

Our model explains such GPS-facilitated errors as arising from two aspects of the APEX architecture. First, APEX maintains an estimate of future "busyness" called subjective-workload (SW) based on urgency values of pending tasks. A large number of moderately urgent tasks would result in a large SW, as would a small number of highly urgent tasks. Second, the model's action selection component uses SW in several ways to conditionalize action choice. For the present example, its most important effect is to cause action selection to "cut corners" -- i.e. to neglect the least important cognitive and perceptual activities in high workload conditions so that time is available for more crucial operations. This response usually proves adaptive, but can lead to error if neglected operations prove unexpectedly important.

In our GPS example, choosing a waypoint from those stored on the GPS device requires cycling through a course-ordered list of waypoints until a stopping criterion is met. Satisfying the minimum effort criterion involves simply waiting until the current point (the one just passed) is displayed and then stopping at the subsequent waypoint. If more effort is allocated, the user can verify the selected point by recalling from memory whether the indicated portion of the course has been traversed. If so, the user continues cycling through until the stopping criterion is again satisfied. Such verifications are less likely to be performed in high workload conditions, with a consequent increase in the (predicted) likelihood of error.