From AI Trustworthiness to Perceived User Trust in High-Risk AI: An Expert-Derived Conceptual Model

Puntis Palazzolo, Bernd Carsten Stahl, Helena Webb
Proceedings of the Fourth UK AI Conference 2026, PMLR 348:116-133, 2026.

Abstract

Artificial intelligence (AI) systems are increasingly used in high-risk domains where their decisions can significantly affect people’s lives. While trustworthy AI frameworks emphasize attributes such as accuracy, fairness, transparency, and reliability, these attributes do not necessarily translate into user trust. This study examines expert perspectives on this relationship through 14 interviews across four high-risk industries and six countries. Thematic analysis identified interconnected technical, human, organizational, and contextual factors that experts believe influence perceived trust. Participants associated trust with system performance and reliability, but also emphasized usefulness, contextual relevance, transparency, accountability, human agency, user control, and organizational reputation. Building on these findings, we propose an expert-derived conceptual framework that extends the established distinction between AI trustworthiness and human trust by explaining how the two may become connected. Rather than assuming that trustworthy system characteristics directly generate trust, our framework proposes that trustworthiness attributes are translated into perceived trust through users’ interpretation and experience of the system. This interpretive stage is shaped by two distinct sets of factors: user dispositional factors, such as demographics, risk tolerance, AI literacy, prior AI experience, and propensity to trust, and contextual influences, such as organizational reputation, social and domain norms, and regulatory context. Perceived trust in turn influences reliance and system use, while remaining distinct from them; because reliance is directly observable, experts often infer trust from it, even though continued use is an ambiguous signal that may instead reflect convenience, necessity, or a lack of alternatives. This framework represents experts’ perspectives rather than direct evidence of users’ trust experiences and provides a foundation for future empirical research with end-users.

Cite this Paper

BibTeX
@InProceedings{pmlr-v348-palazzolo26a, title = {From AI Trustworthiness to Perceived User Trust in High-Risk AI: An Expert-Derived Conceptual Model}, author = {Palazzolo, Puntis and Stahl, Bernd Carsten and Webb, Helena}, booktitle = {Proceedings of the Fourth UK AI Conference 2026}, pages = {116--133}, year = {2026}, editor = {Benford, Alistair and Büyükateş, Baturalp and Cabrera, Christian and Kiden, Sarah and Salili-James, Arianna and Zakka, Vincent and Zhou, Feng}, volume = {348}, series = {Proceedings of Machine Learning Research}, month = {29--30 Sep}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v348/main/assets/palazzolo26a/palazzolo26a.pdf}, url = {https://proceedings.mlr.press/v348/palazzolo26a.html}, abstract = {Artificial intelligence (AI) systems are increasingly used in high-risk domains where their decisions can significantly affect people’s lives. While trustworthy AI frameworks emphasize attributes such as accuracy, fairness, transparency, and reliability, these attributes do not necessarily translate into user trust. This study examines expert perspectives on this relationship through 14 interviews across four high-risk industries and six countries. Thematic analysis identified interconnected technical, human, organizational, and contextual factors that experts believe influence perceived trust. Participants associated trust with system performance and reliability, but also emphasized usefulness, contextual relevance, transparency, accountability, human agency, user control, and organizational reputation. Building on these findings, we propose an expert-derived conceptual framework that extends the established distinction between AI trustworthiness and human trust by explaining how the two may become connected. Rather than assuming that trustworthy system characteristics directly generate trust, our framework proposes that trustworthiness attributes are translated into perceived trust through users’ interpretation and experience of the system. This interpretive stage is shaped by two distinct sets of factors: user dispositional factors, such as demographics, risk tolerance, AI literacy, prior AI experience, and propensity to trust, and contextual influences, such as organizational reputation, social and domain norms, and regulatory context. Perceived trust in turn influences reliance and system use, while remaining distinct from them; because reliance is directly observable, experts often infer trust from it, even though continued use is an ambiguous signal that may instead reflect convenience, necessity, or a lack of alternatives. This framework represents experts’ perspectives rather than direct evidence of users’ trust experiences and provides a foundation for future empirical research with end-users.} }
Endnote
%0 Conference Paper %T From AI Trustworthiness to Perceived User Trust in High-Risk AI: An Expert-Derived Conceptual Model %A Puntis Palazzolo %A Bernd Carsten Stahl %A Helena Webb %B Proceedings of the Fourth UK AI Conference 2026 %C Proceedings of Machine Learning Research %D 2026 %E Alistair Benford %E Baturalp Büyükateş %E Christian Cabrera %E Sarah Kiden %E Arianna Salili-James %E Vincent Zakka %E Feng Zhou %F pmlr-v348-palazzolo26a %I PMLR %P 116--133 %U https://proceedings.mlr.press/v348/palazzolo26a.html %V 348 %X Artificial intelligence (AI) systems are increasingly used in high-risk domains where their decisions can significantly affect people’s lives. While trustworthy AI frameworks emphasize attributes such as accuracy, fairness, transparency, and reliability, these attributes do not necessarily translate into user trust. This study examines expert perspectives on this relationship through 14 interviews across four high-risk industries and six countries. Thematic analysis identified interconnected technical, human, organizational, and contextual factors that experts believe influence perceived trust. Participants associated trust with system performance and reliability, but also emphasized usefulness, contextual relevance, transparency, accountability, human agency, user control, and organizational reputation. Building on these findings, we propose an expert-derived conceptual framework that extends the established distinction between AI trustworthiness and human trust by explaining how the two may become connected. Rather than assuming that trustworthy system characteristics directly generate trust, our framework proposes that trustworthiness attributes are translated into perceived trust through users’ interpretation and experience of the system. This interpretive stage is shaped by two distinct sets of factors: user dispositional factors, such as demographics, risk tolerance, AI literacy, prior AI experience, and propensity to trust, and contextual influences, such as organizational reputation, social and domain norms, and regulatory context. Perceived trust in turn influences reliance and system use, while remaining distinct from them; because reliance is directly observable, experts often infer trust from it, even though continued use is an ambiguous signal that may instead reflect convenience, necessity, or a lack of alternatives. This framework represents experts’ perspectives rather than direct evidence of users’ trust experiences and provides a foundation for future empirical research with end-users.
APA
Palazzolo, P., Stahl, B.C. & Webb, H.. (2026). From AI Trustworthiness to Perceived User Trust in High-Risk AI: An Expert-Derived Conceptual Model. Proceedings of the Fourth UK AI Conference 2026, in Proceedings of Machine Learning Research 348:116-133 Available from https://proceedings.mlr.press/v348/palazzolo26a.html.

Related Material