Human Factors Guidance for the Integration of Artificial Intelligence and Machine Learning in FAA Systems

Smith, Philip J.; Roth, Emilie M.; Sarter, Nadine; Atkins, Ella; Evans, Mark; Bihari, Tom; Kim, Heejin · 2025 · ROSA P / United States. Department of Transportation. Federal Aviation Administration

archive: archived pipeline: cataloged verified

Get this paper ↗ (full text — opens at the source; we link to it, we don't host it)

Summary

This report provides human factors (HF) guidance for integrating artificial intelligence (AI) and machine learning (ML) technologies into Federal Aviation Administration (FAA) systems. The primary objective is to support safety-critical operations in air traffic control, traffic flow management, and technical operations through the design of effective decision support tools (DSTs). The document addresses the challenge of ensuring that AI/ML systems augment rather than hinder human performance, emphasizing the need for complementary systems that leverage the distinct strengths of both humans and automation. It targets HF specialists, DST developers, systems engineers, and program managers involved in the research, implementation, and acquisition of FAA systems. The guidance is structured around ten key design areas, including supporting complementary performance, enhancing interpretability, fostering appropriate reliance, preventing cognitive fixation, and managing attention. The authors draw upon established human-automation interaction literature, recent research on human-AI interaction, and industry design principles from major technology firms. The report characterizes AI/ML systems using frameworks such as the NIST AI Risk Management Framework, highlighting traits like transparency, explainability, and resilience. It distinguishes between general human-automation guidance and specific considerations for AI/ML, including emerging technologies like Generative AI and Large Language Models. The guidance applies across the FAA Acquisition Management System process, from initial service analysis to solution implementation, urging early engagement of HF specialists to inform technology selection and interface design. Key findings and recommendations emphasize that AI/ML systems and humans are fallible in different ways, necessitating designs that foster synergistic performance. The report warns that without careful HF design, users may reject correct AI suggestions or accept incorrect ones, and that concurrent presentation of AI conclusions can impair users' ability to develop accurate mental models. To mitigate these risks, the guidance recommends designing DSTs to provide verification-focused explanations that allow users to independently evaluate AI outputs. This includes displaying input data, relevant context not used by the algorithm, and evidence for and against recommendations. The report also addresses the trade-off between model interpretability and performance, noting that while complex models like deep neural networks are difficult to explain fully, designers must still provide sufficient information for users to assess the appropriateness of AI conclusions. The significance of this work lies in its provision of actionable, non-mandatory recommendations to ensure AI/ML integration in FAA systems is useful, usable, and effectively utilized. By focusing on complementary performance, the guidance aims to prevent negative outcomes such as over-reliance, under-reliance, and cognitive biasing. It underscores the importance of considering task-specific constraints, such as time stress and workload, when applying these recommendations. Ultimately, the report seeks to establish a foundation for trustworthy and effective human-AI collaboration in aviation, ensuring that automation supports rather than replaces critical human expertise in safety-critical environments.

Key finding

Effective integration of AI/ML into FAA systems requires design principles that foster complementary human-machine performance, enhance interpretability, and support appropriate user reliance to mitigate cognitive biases and fixation.

Methodology

review

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via bulk_ingest_rosap on 2026-05-23 (6 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• automation

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Applied Guidance: design guidelines