An Empirical Study on Automation Transparency (i.e., seeing-into) of an Automated Decision Aid System for Condition-based Maintenance

EasyChair Preprint 5656

Abstract

Prior studies have shown conflicting results about the impact of information disclosure on human performance– often referred to as transparency (i.e., seeing-into) studies. We conducted an experiment to investigate whether transparency manipulations predicted whether participants could identify whether features and their relative weights of a decision aid guided by a Machine Learning model were consistent with stated best practices for making maintenance decisions. We had insignificant results on state estimation, automation reliance, trust, workload, and self-confidence. This study shows that disclosing information about the decision aid rationale does not necessarily impact operator performance.

Keyphrases: Automation Transparency, Decision aids, Maintenance, human performance

@booklet{EasyChair:5656,
  author    = {Fahimeh Rajabiyazdi and Greg A. Jamieson and David A. Quispe G.},
  title     = {An Empirical Study on Automation Transparency (i.e., seeing-into) of an Automated Decision Aid System for Condition-based Maintenance},
  howpublished = {EasyChair Preprint 5656},
  year      = {EasyChair, 2021}}