Type: dissertation
Abstract: Algorithmic decision-making is increasingly integrated into digital healthcare systems such as symptom checkers and self-triage applications.Although these algorithms can perform comparably to human experts in providing advice, public trust in these algorithmic systems is generally lower than trust in human professionals.Previous research suggests that conversational user interfaces (CUIs), more commonly referred to as chatbots, can increase users' trust compared to traditional graphical interfaces.However, many of these studies compare systems that differ not only in interface modality but also in the underlying decision-making model.For example, form-based symptom checkers usually rely on white-box interpretable models, while chatbot-based symptom checkers typically use black-box models for both communication and diagnosis generation.This study investigates whether trust in interpretable white-box diagnostic models can be increased by presenting users with a CUI rather than a Form UI.A mixed-design experiment was conducted comparing three UI types: a traditional form-based interface, a default uninstructed conversational interface, and a conversational interface instructed to emulate a medical communication style.Across all conditions, the same diagnostic models were used to provide diagnoses for actinic keratosis (AK) and basal cell carcinoma (BCC).Trust in the given diagnosis was measured using an adapted version of the Short Trust in Automation Scale (S-TIAS).In addition, participants' medical care preferences and prior familiarity with chatbots were measured and included as moderating variables.In total, 65 participants completed the study, resulting in 184 valid trials.Statistical analyses showed no significant differences in trust between the Form UI and CUIs.Qualitative reviews of conversation logs indicated that participants in the Medical CUI condition provided longer and more elaborate responses than participants interacting with the Default CUI.However, these behavioral differences did not translate into measurable increases in trust.Furthermore, no evidence was found that medical care preference or chatbot familiarity moderated the relationship between UI type and trust.The findings suggest that conversational modality alone may not be sufficient to increase trust in symptom checkers.Instead, previously observed trust advantages of chatbots may depend on additional factors, such as personalized explanations, confirmation behavior or post-diagnosis discussions.By separating the conversational model from the underlying diagnostic model, these potentially trust-increasing behaviors may have been constrained.This study contributes to a better understanding of how CUI-based symptom checkers should be designed and evaluated when taking trust into consideration.Trust in Dermatological Symptom Checkers: Isolating the Effect of Conversational Interface Design iii iv Trust in Dermatological Symptom Checkers: Isolating the Effect of Conversational Interface Design Preface I started my master's education after more than six years of experience as a software engineer.My motivation was not simply to obtain another degree, but to deepen my understanding of how to design and evaluate better technological products.I wanted to look beyond whether a system works technically, and better understand how people experience, interpret, and trust the systems they use.Now, nearly four years later, this thesis marks the end of that academic chapter.This thesis reflects that same motivation.Throughout the project, I was interested not only in whether a digital health system can provide useful advice, but also in how such advice is perceived by the people using it.Technical performance alone is not enough.A system may be accurate, efficient, or carefully designed, but its value ultimately depends on whether users understand it, trust it, and are willing to act on its advice.This topic therefore felt like a meaningful way to combine my background in software engineering with the human-centered perspective I developed during this master.Conducting this study while continuing my work as a software engineer would not have been possible without the understanding and flexibility of my colleagues at CGI and the Dutch National Police.I cannot count the number of times they covered for me when I had to adjust my schedule at the last minute, and I would like to express my sincere gratitude to them.I would also like to thank my daily supervisor, Chris, whose guidance and feedback throughout the project were invaluable.Finally, I would like to thank my family and friends, especially my little brother, for listening when I needed to vent during moments of low motivation or other struggles.This thesis may have been written by me, but all of you contributed to it in meaningful ways.To the reader, thank you for your interest in this topic.I hope this thesis provides a useful perspective on the role of interface design on trust.Hrvoje Miljak Trust in Dermatological Symptom Checkers: Isolating the Effect of Conversational Interface Design v vi Trust in Dermatological Symptom Checkers: Isolating the Effect of Conversational Interface Design vii
Source: TU/e Research Portal
Author: Miljak, H.
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7203517058