Exploring Opportunities in Usable Hazard Analysis Processes for AI Engineering
Abstract
Embedding artificial intelligence into systems introduces significant challenges to modern engineering practices. Hazard analysis tools and processes have not yet been adequately adapted to the new paradigm. This paper describes initial research and findings regarding current practices in AI-related hazard analysis and on the tools used to conduct this work. Our goal with this initial research is to better understand the needs of practitioners and the emerging challenges of considering hazards and risks for AI-enabled products and services. Our primary research question is: Can we develop new structured thinking methods and systems engineering tools to support effective and engaging ways for preemptively considering failure modes in AI systems? The preliminary findings from our review of the literature and interviews with practitioners highlight various challenges around integrating hazard analysis into modern AI development processes and suggest opportunities for exploration of usable, human-centered hazard analysis tools.
Cite
@article{arxiv.2203.15628,
title = {Exploring Opportunities in Usable Hazard Analysis Processes for AI Engineering},
author = {Nikolas Martelaro and Carol J. Smith and Tamara Zilovic},
journal= {arXiv preprint arXiv:2203.15628},
year = {2022}
}
Comments
8 pages, Presented at 2022 AAAI Spring Symposium Series Workshop on AI Engineering: Creating Scalable, Human-Centered and Robust AI Systems