Artificial Intelligence, the Clinician–patient Relationship and the Democratisation of Health Care: A Critical Narrative Review
Mitesh Mohan Hood *
WPP Production, India.
*Author to whom correspondence should be addressed.
Abstract
Artificial intelligence has moved from retrospective evaluation into everyday clinical settings, and two claims now dominate professional and policy discussion. The first holds that these systems will return relational time and attention to the consultation by absorbing administrative labour. The second holds that they will democratise health care by widening access to expertise that has historically been rationed by geography, workforce supply and cost. Both claims carry considerable normative weight, and both are being used to justify rapid procurement decisions, yet the evidence supporting them originates in separate research traditions and has rarely been appraised together. This critical narrative review examines the strength, consistency and limitations of that evidence, treating relational quality and distributive justice as interdependent rather than as parallel concerns. Literature was identified through structured searching of biomedical and multidisciplinary indexes, supplemented by backward and forward citation tracking and by authoritative institutional guidance. The synthesis is organised around five analytical problems: the reallocation of clinician attention through ambient documentation and drafted correspondence; the evidentiary status of machine-generated empathy; the redistribution of epistemic authority and its consequences for trust; cognitive dependence, skill erosion and the location of responsibility; and the conditions under which broader access translates into narrower inequity. The available evidence indicates that documentation technologies reliably improve clinician-reported workload and burnout, that objective efficiency gains are far less consistent, and that direct measurement of patient-experienced relational quality remains conspicuously scarce. Comparative studies favouring machine-generated communication rest predominantly on asynchronous text and clinician-rater judgements rather than on encounters with real patients. Evidence bearing on democratisation is weakest precisely where the claim is strongest, since deployment, evaluation and training data remain concentrated in high-income and academic settings. Confidence in current conclusions is limited by short observation periods, single-site designs and reliance on self-report. Priorities are proposed for evaluation that measures relational and distributive outcomes directly.
Keywords: Ambient documentation, clinician–patient relationship, digital health equity, generative artificial intelligence, health-care access, large language models, trust in technology