GHOST: experimenting conflicts countermeasures in the pilot's activity

Frédéric Dehais, Catherine Tessier, Laurent ChaudronIJCAI 2003

Abstract

The GHOST project introduces a novel AI-driven framework designed to detect and mitigate "perseveration syndrome" in aircraft pilots. This dangerous cognitive state occurs when a pilot fixates on a failing goal while ignoring critical onboard warnings. By distinguishing mission-critical conflicts from simple inconsistencies and employing "shock" countermeasures—such as the temporary removal of fixated displays—the GHOST system successfully nudges pilots to abandon dangerous maneuvers and choose safer mission alternatives.

Introduction: The Challenge of Perseveration Syndrome

In aviation, major accidents often stem from a breakdown in how pilots process information rather than a lack of data. Under high-stress conditions, such as engine failure or sudden loss of visibility, pilots may fall victim to perseveration syndrome. In this state, they become cognitively "locked in" on their original goal, focusing exclusively on a single instrument while ignoring critical alarms.

Figure 3: Spatial representation of perseveration syndrome, where a pilot repeatedly circles a target despite mission failure.

Existing cockpit technologies often exacerbate the problem by simply adding more alarms, which a perseverating pilot is likely to ignore. The GHOST project addresses this by computationally distinguishing between harmless inconsistencies and mission-threatening "conflicts," then applying interventions to break the pilot's tunnel vision.

The GHOST Methodology

The researchers developed the GHOST (experimenting conflicts countermeasures in the pilot's activity) environment, which utilizes a three-step approach to manage pilot attention:

  1. Activity Tracking: The system reconstructs pilot actions by comparing real-time flight parameters against a "Reference Model" based on Propositional Attitudes (PAs), which represent the pilot's goals and constraints.
  2. Conflict Detection: Unlike standard logic that flags all errors, GHOST uses a two-step filter. It identifies incoherent data sets and determines if they involve Crucial PAs—constraints vital for safety. If an inconsistency threatens the mission, it is flagged as a conflict.
  3. Countermeasures: Based on the theory that pilots ignore added information during stress, the system employs "shock" countermeasures. It temporarily removes the information the pilot is fixated on (e.g., blacking out a primary display) for a few seconds to reset the cognitive loop, then replaces it with an explicit solution (e.g., "Fly back to Blagnac").
Figure 6: The GHOST system architecture, illustrating the loop from flight simulation to conflict detection and countermeasure deployment.

Experimental Evaluation and Results

The study involved human-in-the-loop simulations with 21 pilots, ranging from novices to experienced military personnel. Using a custom setup with Flightgear and a "Wizard of Oz" interface, researchers observed pilot reactions to impossible landing scenarios.

The effectiveness of the GHOST countermeasures was significant:

  • Control Group (No Countermeasures): Out of 7 pilots, zero made the correct decision to abort the landing. All 7 persevered, resulting in 4 simulated crashes.
  • Experimental Group (With Countermeasures): Out of 12 pilots, 9 (75%) successfully changed their minds and flew to safety after the system "shocked" them out of their perseveration.

Qualitative feedback suggested that the temporary removal of a display was less stressful than traditional auditory alarms because it allowed pilots to immediately re-evaluate the core problem.

Conclusion and Future Work

The GHOST project demonstrates that "shocking" a pilot's attention can be a powerful tool for flight safety. While the current study used a human operator to trigger countermeasures, future work aims to automate the system using Kalmansymbo(aero), a Petri net-based situation tracker. This will enable autonomous conflict detection and intervention, ensuring that pilots receive specific, actionable instructions when cognitive lock-in occurs.

Summary updated

BibTeX
@inproceedings{dehais2003ghost,
  author = {Frédéric Dehais and Catherine Tessier and Laurent Chaudron},
  title = {{GHOST: experimenting conflicts countermeasures in the pilot's activity}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2003},
}
Figure