- The paper develops a novel MDMT instrument that expands trust measurement from two to five dimensions, incorporating capable, reliable, ethical, sincere, and benevolent traits.
- It details the instrument’s evolution through rigorous empirical studies and revisions, demonstrating strong internal consistency and flexible subscale application in human-robot contexts.
- Findings highlight the MDMT’s potential to set validation benchmarks in human-robot interaction research and encourage continual refinement based on selective subscale analysis.
Background
The development of the Multi-Dimensional Measure of Trust (MDMT) represents an effort to provide a comprehensive, intuitive, and theoretically grounded self-report instrument to assess perceived trustworthiness across different agents, including humans, robots, animals, and AI. Pre-existing measures of trust were found to have several limitations, such as being overly time-consuming, not tailored for interactions specific to humans and robots, or lacking in theoretical framework and psychometric robustness. In response, the MDMT was created to address these issues and has since undergone multiple revisions and validations.
The first version of the MDMT emerged from empirical studies aimed at identifying distinct dimensions of trustworthiness. Initial research assumed the existence of only two main dimensions—capacity trust and moral-personal trust. Surprisingly, analysis of study data revealed a four-dimensional structure comprising Capable, Reliable, Ethical, and Sincere. This version allowed respondents to rate trust across these dimensions or opt out if the attribute seemed inapplicable to the agent in question. This feature, known as "Does Not Fit," improved the interpretability of trust ratings, providing a mechanism for acknowledging areas where trust dimensions may not be relevant to simpler robotic agents.
Further research and a comprehensive review of existing literature suggested the addition of a fifth dimension, Benevolence. A new study presenting 41 trust-related terms to participants supported the expansion to this five-dimensional model, demonstrating clear links between the newly identified attributes and the five established dimensions. As a result, the MDMT was revised to version 2, which includes four items for each of the five dimensions, resulting in a 20-item instrument.
Strengths, Limitations, and Future Direction
The MDMT boasts strong internal consistency and the flexibility to use certain subscales selectively, based on the context and type of "trustee." While the measure shows promise, its dimensions often exhibit high inter-correlations, potentially indicating an overlap in the conceptual space of trust. Ongoing analysis and the responses to specific information suggest subscales can be differentiated in response to particular contexts, even when correlations are high.
Future research will aim to establish validation benchmarks within the human-robot interaction (HRI) field while refining the MDMT to ascertain its definitive validity and reliability. The research collective is encouraged to continually utilize MDMT, contributing to its ongoing refinement and defining the scope of its applicability in diverse trust-related scenarios.
