An automated method to detect potential mode confusions

Abstract

Mode confusions are a type of "automation surprise"-circumstances where an automated system behaves differently than its operator expects. It is generally accepted that operators develop "mental models" for the behavior of automated systems and use these to guide their interaction with the systems concerned, so that an automation surprise results when the actual system behavior diverges from its operator's mental model. Complex systems are often structured into "modes" (for example, an autopilot might have different modes for altitude capture, altitude hold, and so on), and their behavior can change significantly across different modes. "Mode confusion" arises when the system is in a different mode than that assumed by its operator; this is a rich source of automation surprises, since the operator may interact with the system according to a mental model that is inappropriate for its actual mode. Mode confusions have been implicated in several recent crashes and other incidents, and are a growing source of concern in modern automated cockpits. If we accept that mode confusions are due to a mismatch between the actual behavior of a system and the mental model of its operator, then one way to look for potential mode confusions is to compare the design of the actual system against a mental model. There are two challenges here: how to get hold of a mental model, and how to do the comparison. Through observation, questionnaires, and other techniques, psychologists have been able to elicit the mental models of individual operators (typically pilots). However, comparison between a design and the mental model of a specific individual will provide only very specific information; we are interested in whether a design is prone to mode confusions, and for this purpose it is more useful to compare the design against a generic mental model rather than that of an individual.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 47 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.