A Longitudinal Study of Trust Calibration Methods with Individual Differences

Abstract

This report summarizes our major research activities, study results and research accomplishments out of the trust calibration project in the past years. This is also the final report of the project. We have conducted different experiments on trust examination with varied system accuracy, and human trust in predictive decision making. From the study we have revealed that: 1) AI performance, in particular system errors, has a huge implication on human's trust adjustment and decision making. 2) Humans are able to perceive the detailed system performance not only at system level but also at subsystem level using systematic sampling strategies. 3) Trust knowledge acquired by human could be used to guide their future decision making. 4) The influence information of training data points (functioned as fact-checking) on predictions can benefit trust. 5) Personality traits affect trust in predictive decision making differently under different cognitive load levels. Our on-going work is focusing on refined quantification of human trust, and its implication on perception and decision making in the human-machine collaboration context.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jun 21, 2022
Accession Number
AD1175601

Entities

People

  • Fang Chen

Organizations

  • Commonwealth Scientific and Industrial Research Organisation

Tags

Communities of Interest

  • Autonomy
  • Human Systems

DTIC Thesaurus Topics

  • Accuracy
  • Algorithms
  • Artificial Intelligence
  • Automation
  • Control Systems
  • Dynamics
  • Errors
  • Experimental Design
  • Human-Machine Interaction
  • Human-Machine Interfaces
  • Human-Machine Systems
  • Information Science
  • Perception
  • Personality
  • Psychology
  • Training
  • Uncertainty

Fields of Study

  • Computer science

Readers

  • Instructional Design and Training Evaluation.
  • Team-Based Human-Centered Cognitive Task Decision Making and Information Performance.