Ethics of AI-Driven Decision-Making in Healthcare Interfaces

Thesis Abstract

The pervasive integration of AI-driven decision-support systems into healthcare interfaces raises critical ethical concerns surrounding patient autonomy, accountability, transparency, and trust, particularly as clinicians increasingly rely on algorithmic recommendations in high-stakes medical decisions. This study investigates how AI-driven decision-making in healthcare interfaces influences ethical percepts and professional practice, and how normative theories can guide the responsible design and deployment of such systems. The aims are to (a) examine clinicians’ and patients’ ethical evaluations of AI-assisted decisions, (b) identify factors that modulate trust, acceptability, and perceived accountability, and (c) develop a normative framework for ethical interface design grounded in established theories of moral agency and technomoral change. Specific objectives include (1) mapping the dimensions of ethics salient to AI-enabled interfaces (autonomy, explanation, fairness, safety) as experienced by end-users; (2) assessing how transparency and explainability features affect perceived legitimacy and clinical decision-making performance; (3) evaluating the impact of AI assistance on perceived professional accountability and liability concerns; (4) testing a theoretical model linking trust, perceived fairness, and user acceptance to intention to use AI-assisted interfaces in patient care; and (5) proposing design guidelines anchored in virtue ethics, principlism, and the theory of technomoral accommodation. The study employs a mixed-methods design comprising two sequential phases. Phase I uses a cross-sectional survey of 260 clinicians (physicians and advanced practice clinicians) and 260 patients across three tertiary hospitals to quantify variables related to trust, perceived transparency, accountability, and willingness to rely on AI recommendations, measured through validated scales adapted from prior work in medical informatics and ethics. Phase II involves 40 in-depth semi-structured interviews (20 clinicians, 20 patients) and four focus groups with 6–8 participants each to elicit nuanced interpretations of ethical concerns, experiences with explainable AI features, and scenarios illustrating decision-making under AI assistance. Data collection instruments include a clinician-patient vignette-based questionnaire, an explainability preference inventory, and interview/focus group guides informed by moral philosophy and human-computer interaction theory. Validity and reliability are enhanced through pilot testing (n=30), triangulation across methods, and member checking. Quantitative analyses employ structural equation modeling (SEM) to test the proposed model linking explainability, transparency, and trust to acceptance of AI-enabled decisions, with multi-group analysis to detect differences between clinician and patient perspectives. Regression analyses will identify predictors of perceived accountability and willingness to rely on AI recommendations. Qualitative data will undergo reflexive thematic analysis, following Braun and Clarke’s framework, with coding performed by two independent researchers and adjudicated through consensus. A deductive approach will integrate findings with theories of moral agency (Miller), principlism (Beauchamp and Childress), virtue ethics (MacIntyre), and technomoral change (Verbeek), producing a robust normative framework. Expected findings anticipate that higher levels of explainability and patient-facing justifications will bolster trust and perceived legitimacy, but may also raise concerns about the accuracy-variance trade-off and liability ambiguity. It is hypothesized that explicit attribution of decision responsibility (who bears accountability) and transparent disclosure of AI limitations will mitigate perceived risk and improve acceptance among patients and clinicians. The study will identify contextual factors (clinical domain, severity of condition, and health literacy) that modulate ethical acceptability and system use. The contribution to knowledge includes (i) empirical illumination of how ethical considerations shape the adoption of AI in healthcare interfaces across stakeholder groups; (ii) a normative framework integrating virtue ethics, principlism, and technomoral change to inform ethical interface design and governance; and (iii) actionable design guidelines for developers and healthcare organizations emphasizing explainability, accountability attribution, and user-centered transparency. The findings will inform policy recommendations on consent processes, liability frameworks, and professional training for AI-enabled care. The main conclusion is that ethically robust AI-driven decision-making in healthcare interfaces necessitates a dual focus on technical explainability and coherent moral governance to sustain clinician stewardship, patient autonomy, and trust in algorithm-assisted medical practice. Recommended strategies include embedding context-sensitive explanations, explicit accountability mappings, ongoing ethical audits, and interdisciplinary collaboration in the development cycle.