Suher Salim
Research Mentor: Paul Green
Mentor Department: University of Michigan Transportation Research Institute, Other
Author(s): Suher Salim, Paul Green
Session: Session 5 (2:00 PM – 2:50 PM)
Presentation Type: Poster 132
Abstract
Human factors engineers and ergonomists are increasingly encountering AI-generated outputs in their professional workflows, yet little guidance exists on how to effectively communicate with large language models (LLMs) to produce reliable, domain-appropriate results. This paper addresses three research questions: (RQ1) What prompt engineering techniques are most effective for human factors tasks? (RQ2) Do those recommendations hold when applied across popular AI tools? (RQ3) Which AI tools and prompt structures are best suited to the specific task categories that human factors professionals routinely perform? To address RQ1, a systematic literature review was conducted using Google Scholar, ACM Digital Library, IEEE Xplore, arXiv, and supplementary industry sources. Search terms spanning prompt engineering methodology and human factors applications were used, with AI-assisted screening verified by the student researcher. The review identified structured Role-Context-Task-Format prompting, few-shot prompting with domain examples, and chain-of-thought (CoT) reasoning as the highest-efficacy techniques for human factors work. To address RQ2 and RQ3, a series of applied case studies will evaluate these three prompt structures across the 27 task categories defined by O*NET for Human Factors Engineers and Ergonomists (SOC 17-2112.01), using an air traffic control scenario as a primary test case. A participant study at the University of Michigan will then invite human factors engineers to rate AI-generated outputs produced under each prompt condition for accuracy, completeness, and usability. Findings are expected to show that AI tools offer meaningful productivity gains across a broad range of human factors task categories, particularly in report writing, literature synthesis, incident analysis, and data interpretation, while also confirming that virtually all AI outputs will require expert human review before operational use. AI is not expected to replace human factors engineers, but rather to substantially reduce the burden of time-intensive, data-heavy processes. Additionally, different prompt structures are expected to prove more effective for different task types, providing HF practitioners with actionable, task-specific prompting guidance.
