Between map and maze: reframing trust in healthcare AI

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine
• Trust Studies

Background:

• Artificial intelligence (AI) integration in healthcare promises enhanced diagnostics and decision-making.
• The concept of trust is central to AI adoption but lacks systematic analysis.
• Existing literature often treats trust as a designable attribute or behavioral calibration.

Purpose of the Study:

• To systematically review and analyze conceptualizations of trust, trustworthiness, distrust, and mistrust in AI healthcare literature.
• To identify and categorize different framings of trust in the context of healthcare AI.
• To explore the implications of these conceptualizations for understanding AI integration in clinical practice.

Main Methods:

• An interdisciplinary scoping review of 82 publications from 2015-2025.
• Literature search conducted across six major databases: Web of Science, Scopus, PubMed, PhilPapers, SocINDEX, and ACM Digital Library.
• Analysis focused on how trust is defined, measured, problematized, and conceptualized across different disciplines.

Main Results:

• Identified five distinct conceptualizations of trust: principles, attitude/belief, binary, structural mechanism, and relational process.
• Trust is often framed as an engineering problem or a behavior to be calibrated, detached from clinical realities.
• Distrust, mistrust, overtrust, and undertrust are undertheorized, viewed as obstacles rather than analytical phenomena.

Conclusions:

• Trust in healthcare AI is a dynamic negotiation, not a stable prerequisite for adoption.
• Diverging conceptualizations reveal power dynamics, uncertainties, and institutional dependencies.
• Analyzing distrust and mistrust offers a more reflexive understanding of AI's contested role in healthcare.