Investigating Human Trust in AI- A Case Study of Human-AI Collaboration on a Speech-Based Data Analytics Task
Abstract
This research will investigate trust in artificial intelligence (AI) during a human-in-the-loop collaborative speech-based data analytics task (DAT). The increasing use of AI in the military gives rise to questions about how much AI systems are and can be trusted by their users. There is evidence suggesting that the dimensions of trust in automation and trust in AI are common. However, despite the similarities, it is unclear whether trust within human-AI collaboration actually has the same dynamic as in the human-machine context (e.g., vehicles), therefore the proposed work will provide empirical evidence and yield an improved understanding as to how human trust is established, maintained, and repaired when teaming up with AI systems. The proposed work will examine dimensions of trust during a human-AI collaboration paradigm that involves a speech-based DAT, in which human users will be called to collaborate with an AI system in detecting deceptive-truthful speech, a challenging DAT of high relevance to AFRL and the military. The objectives of the proposed work are to- (1) Investigate dimensions of trust in human-AI collaborative DAT- Trust calibration, resolution and specificity will be examined via annotators’ self-reported, behavioral, and neural measures; (2) Identify human and system-related factors of trust in AI- The annotators traits (i.e., expertise, personality, trust propensity) and the performance and characteristics of the AI system will be studied as potential factors that affect the way human annotators establish and maintain their trust in the AI; and (3) Build an evidence-based model of human trust in AI and its effect on human-AI teaming outcomes- Results yielding from the previous objectives will be integrated into a unified empirical model of human trust in the AI and its effect on the overall DAT performance. The anticipated outcomes from this research include- (1) new knowledge about human factors of trust and AI design elements that affect trust in AI; (2) a new empirical model of human trust in AI; (3) curated data, including self-reports, neural measures, and outcomes, from human-AI collaboration during a speech-based DAT; and (4) detailed report of this research and outcomes.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 07, 2023
- Source ID
- FA95502210010
Entities
People
- Theodora Chaspari
Organizations
- Air Force Office of Scientific Research
- Texas A&M University
- United States Air Force