New framework redefines Human-AI teaming for future flight safety

The researchers stress that many traditional Human Factors milestones such as Fitts' List of human-versus-machine capabilities, the Aviation Safety Reporting System, and Crew Resource Management, need urgent revision in light of the growing cognitive parity between AI and human operators. New concepts like operational explainability, cognitive workload mitigation, and trust calibration now dominate the Human Factors agenda.

A new study published in Future Transportation outlines a comprehensive set of human factors requirements to ensure artificial intelligence (AI) can work seamlessly alongside pilots and air traffic controllers in the aviation industry. The research, conducted under the European Union’s Horizon Europe HAIKU project, tackles the pressing challenge of designing human-AI teaming (HAT) (aka, Human-Autonomy Teaming) systems that enhance operational efficiency without compromising safety. As aviation systems integrate increasingly autonomous AI tools, called Intelligent Agents (IAs), the research argues that existing Human Factors (HF) frameworks fall short of ensuring safe and effective human-AI collaboration in cockpit and air traffic management environments.

The study “Human Factors Requirements for Future Aviation Human-Autonomy Teaming Systems: A Preliminary Framework and Evaluation via Use Cases” details a comprehensive 165-point Human Factors requirements set across eight key areas. Developed by the HAIKU project, an EU-funded initiative running from 2023 to 2025, the proposed framework is positioned as a critical safety assurance tool for aviation authorities, system designers, and operational teams preparing for the rapid emergence of AI in civil aviation systems.

What challenges will future AI systems pose for aviation?

The research questions the viability of current safety design paradigms in the face of accelerated AI development. Unlike past automation waves such as the glass cockpit transition of the 1980s, AI integration introduces adaptive, learning systems whose decision-making logic often remains opaque to human operators. Regulators like the European Union Aviation Safety Agency (EASA) and international counterparts have issued preliminary guidance, but the study argues that broader, non-regulatory Human Factors concerns have yet to be fully addressed.

To ground its analysis, the research poses four core questions: what types of AI and teaming characteristics are emerging in aviation; which elements of legacy Human Factors research remain relevant; what gaps exist between current design standards and future AI demands; and whether a new HF requirements framework can operationally support safe system development.

Through this lens, the HAIKU study identifies eight critical domains: Human-Centred Design, Roles and Responsibilities, Sense-Making, Communication, Teamworking, Error and Failure Management, Competencies and Training, and Organisational Readiness. These domains are designed to address systemic risks posed by AI’s autonomy and the evolving nature of pilot-AI and controller-AI interactions.

How does the proposed Human Factors framework address design gaps?

The researchers stress that many traditional Human Factors milestones such as Fitts' List of human-versus-machine capabilities, the Aviation Safety Reporting System, and Crew Resource Management, need urgent revision in light of the growing cognitive parity between AI and human operators. New concepts like operational explainability, cognitive workload mitigation, and trust calibration now dominate the Human Factors agenda.

The proposed framework includes detailed requirements covering whether AI can explain its reasoning in understandable terms, allow human intervention during emergencies, and share situation awareness data in real time. For example, the system must clarify whether the AI has resolved a threat, allow the human to adjust AI parameters, and provide uncertainty estimates alongside decisions. These features are essential in preventing over-reliance on automation, identifying AI failure modes, and ensuring that human operators remain “on the same page” with their AI teammates.

One standout concern involves “Ironies of Automation,” where humans are relegated to monitor complex systems that can fail without notice. As AI becomes more capable, the study warns of role erosion, where humans are left only with unmanageable or ill-timed responsibilities. Addressing this requires transparent design, continuous training, and the preservation of meaningful human roles in AI-led operations.

Are the new Human Factors requirements practical and effective?

To test real-world applicability, the framework was applied to three aviation use cases between December 2023 and January 2025: a single-pilot emergency handling scenario (UC1), an airport diversion recommender (UC2), and an air traffic control assistant (UC4). The evaluation revealed key areas for system refinement, such as improving how AI status is displayed, introducing aural cues for situational awareness, and managing alert interference during critical events.

Pilots and controllers engaged in the simulations provided direct input, leading to design adjustments in explainability features, parameter control interfaces, and training recommendations. The feedback demonstrated that while the AI prototypes showed promise, their success depended heavily on how well human operators understood and trusted the systems.

In UC2, for instance, pilots benefited from AI-generated airport diversion suggestions paired with transparent explainability layers. Meanwhile, in UC4, air traffic controllers noted that the system’s prediction capabilities enhanced their operational awareness, but flagged concerns over control parameter rigidity. Across all scenarios, the need for personalised AI behavior and clear error signaling emerged as priorities.

The preliminary requirements set offers a scalable solution adaptable to different Technology Readiness Levels (TRLs), AI autonomy categories, and operational roles. It also highlights a planned expansion into Urban Air Mobility use cases, where early-stage AI concepts may benefit from the framework’s diagnostic and design guidance.