Trusting robots: a relational trust definition based on human intentionality

Robots have moved beyond factory automation into collaborative, assistive, and domestic contexts, leading to increasingly complex human-robot interactions where trust is pivotal for effective delegation and reduced supervisory demands. Despite its importance, there is no consensus on a definition of trust in HRI, and studies often conflate trust with related constructs such as trustworthiness of the trustee and the trustor’s disposition to trust. This lack of conceptual clarity hampers theory development and valid measurement. The paper proposes adapting the interpersonal trust model of Mayer et al. (1995) to HRI while centering the trustor’s intentionality. It defines trust as an asymmetric relation from the human trustor to the robot, distinguishing it clearly from antecedents (disposition to trust, perceived trustworthiness) and consequences (risk-taking behavior). The work aims to improve conceptual clarity, guide appropriate measurement and manipulation of trust in empirical studies, and provide a definition applicable to all types of trustees (human, non-human, artificial).

The paper reviews diverse definitions of trust across disciplines, noting common inclusion of a trustor and a trustee but differing emphases on trustor-focused versus trustee-focused approaches. It highlights frequent conflation of trust with perceived trustworthiness (e.g., capability, benevolence, integrity) and with trust disposition, both of which are antecedents rather than trust itself per Mayer et al. (1995). Prior HMI/HRI work (e.g., Lee and See, 2004; Hoff and Bashir, 2015; Malle and Ullman, 2021) often treats trust as involving vulnerability and expectations. Anthropomorphism is insufficient to justify transferring interpersonal definitions to robots due to variability in anthropomorphizing across contexts and robot designs. The authors introduce a phenomenological notion of intentionality (aboutness) to explain the asymmetry and directedness of trust from trustor to trustee, enabling transfer of interpersonal trust concepts to HRI without requiring reciprocal trust or human-like agency in robots. Prior relational accounts (e.g., Taddeo, 2010; Coeckelbergh, 2012; Malle and Ullman, 2021) support viewing trust as occurring within a relation, but the present work specifies trust as a particular asymmetric way of relating, distinct from attitudes or evaluations of trustworthiness.

This is a conceptual/theoretical paper. The authors adapt Mayer et al.’s (1995) interpersonal trust model to HRI by grounding trust in the human trustor’s phenomenological intentionality. They argue trust is an asymmetric relation directed from trustor to trustee and articulate necessary properties that characterize this way of relating: dependence, risk, vulnerability, positive expectations, and free choice. They distinguish trust from antecedents (trustor’s disposition, trustee’s trustworthiness) and from behavioral consequences (risk-taking). The approach justifies applicability to all robots and artificial entities without assuming anthropomorphism or reciprocal trust. The authors then analyze how each property manifests in HRI contexts (e.g., surgical assistance, manufacturing, social interaction scenarios) and discuss implications for unidimensionality of trust and for future measurement development. No empirical data were collected or analyzed.

• Proposed definition: Trust is the way a trustor relates to another entity (the trustee). This way of relating involves being dependent and vulnerable but having positive expectations about the activities of the other. The trustor trusts under circumstances of risk and chooses freely whether to trust the other.
• Trust is an asymmetric relation grounded in the trustor’s intentionality; it does not require the trustee to reciprocate trust or possess human-like agency.
• Five necessary properties of trust: dependence, risk (stemming from uncertainty and lack of control), vulnerability (inherent in the trust relation, not merely willingness), positive expectations about the trustee, and free choice to engage in trust.
• Clear separation of trust from its antecedents: (a) trustor’s disposition to trust and (b) trustee’s perceived trustworthiness (e.g., capability, benevolence, integrity; performance, process, purpose) influence but are not trust.
• Trust is unidimensional; affective and cognitive aspects are better considered as multidimensional antecedents rather than dimensions of trust itself.
• Applicability across HRI: the five properties can characterize human-robot relations in diverse contexts (industrial cobots, social robots, teleoperation, domestic service), including non-social relations.
• Implications for measurement: the definition provides a basis to develop valid instruments that assess the trustor’s trust under the specified properties, avoiding conflation with trustworthiness or disposition.

By centering trust in the trustor’s intentionality and defining it as an asymmetric relation, the paper addresses the conceptual ambiguity in HRI where trust is often conflated with trustworthiness or disposition. The proposed definition allows transferring key distinctions from Mayer et al. (1995) to HRI, clarifying antecedents versus trust itself and separating trust from behavioral outcomes. This clarity supports better theory building and more valid operationalization in empirical research. The account demonstrates that trust in robots does not require anthropomorphism or social relations and applies to various artificial entities (automation, AI), focusing on how humans relate to them under dependence, risk, vulnerability, positive expectations, and free choice. It emphasizes trust’s unidimensionality yet accommodates multidimensional antecedents, offering a coherent framework for modeling trust dynamics over time and informing measurement development.

The paper advances a relational, trustor-focused definition of trust in robots: an asymmetric way of relating characterized by dependence, risk, vulnerability, positive expectations, and free choice. This definition separates trust from trustworthiness and disposition, treats trust as unidimensional, and generalizes to all robots and artificial entities without requiring anthropomorphism. Contributions include: (1) transferring key aspects of an established interpersonal trust model to HRI, (2) enhancing conceptual clarity for research and practice, (3) guiding development of valid trust measures and manipulations, and (4) applicability across entity types and time points. Future research should empirically test the suitability of this definition for initial and evolving trust across HRI contexts and translate the conceptualization into reliable, valid instruments that assess trust under the specified properties.

The work is conceptual and does not include empirical data or a measurement instrument; the proposed definition requires empirical validation across contexts and over time. The definition itself is not an operationalization; additional work is needed to develop and validate tools that assess trust as defined. Data availability notes that no datasets were generated or analyzed.