I bring you a weekly bite-sized chunk of the science behind helicopter human factors and CRM in practice, simplifying the complex and distilling a helicopter related study into a summary of less than 500 words.
TITLE:
Seeing inside the black box: Effects of algorithmic knowledge on system trustworthiness in aviation.
WHAT?
An experimental study investigating whether providing pilots with greater information about how an AI cockpit assistant works influences their trust in the system, and how that trust develops through practical experience.
WHERE?
German Aerospace Center (DLR).
WHEN?
Published online on 20 October 2026 in the journal Ergonomics.
WHY?
Trust dynamics are one of the key challenges for developing and using AI in an aviation context. Unlike conventional automation, AI systems may generate recommendations through processes that pilots cannot readily understand or verify and this impacts trust.
The researchers wanted to establish how and to what extent understanding an AI system’s underlying processes before using it would influence both initial trust and how that trust develops through experience and use.
HOW?
21 commercial pilots participated in a randomised experiment. One group received detailed explanations of an AI assistant’s algorithms, capabilities, limitations and accuracy before using it. The other only received the same information about the AI after completing the experiment.
Pilots first completed a guided tutorial, followed by a simulated flight involving three AI-supported tasks: interpreting ATC taxi instructions, recommending weather diversions and predicting arrival routes and runways.
Researchers then measured perceived reliability, predictability, usability and willingness to use the assistant through questionnaires, and compared responses after the tutorial and flight.
FINDINGS:
- Pilots who received the explanations of the AI algorithms and greater information beforehand reported significantly higher trust following both the tutorial and simulated flight.
- Pilots generally expressed greater trust in the weather-diversion assistant than in the taxi or approach functions, showing that trust varied according to the application and purpose of the AI.
- Pilots with more flight hours tended to report higher trust ratings following the simulated flight. It is possible that experienced pilots possess more developed mental models of aircraft systems and operational processes that could allow them to interpret explanations of an AI assistant more effectively and evaluate its behaviour against their existing knowledge.
- Pilots were less trusting of functions they considered riskier. A pilot might regard an AI recommendation as potentially dangerous because of the consequences of an error, even if the system performs reliably.
SO WHAT?
The findings suggest that pilots’ trust in AI may depend on more than simply how well the system performs. Their existing expertise, understanding of the technology and perception of operational consequences all appear relevant.
Pilots’ level of understanding about how automation works can influence their trust even before they have accumulated substantial experience using it. This means there is training value in explaining of systems’ underlying logic, data sources, capabilities and limitations before using them, rather than simply concentrating on operating procedures.
The study also highlights that trust is application-specific: pilots may trust one function of an AI assistant considerably more than another and this can be dependent on the risks of the system failing.
REFERENCE:
Hamann, A., Heise, A., Lenz, H., & Gampe, A. (2026). Seeing inside the black box: Effects of algorithmic knowledge on system trustworthiness in aviation. Ergonomics. Advance online publication. https://doi.org/10.1080/00140139.2026.2738038
