Reframing 'dehumanisation': AI and the reality of clinical communication.

Abstract

Warnings that large language models (LLMs) could 'dehumanise' medical decision-making often rest on an asymmetrical comparison: the idealised, attentive healthcare provider versus a clumsy, early-stage artificial intelligence (AI). This framing ignores a more urgent reality: many patients face rushed, jargon-heavy, inconsistent communication, even from skilled professionals. This response to Hildebrand's critique argues that: (1) while he worries patients lose a safeguard against family pressure, in practice, time pressure, uncertainty and fragile dynamics often prevent clinician intervention. Because LLMs are continuously available to translate jargon and provide plain-language explanations in a patient's preferred language, they can reduce reliance on companions from the outset and be designed to flag coercive cues and invite confidential 'pause or reset' moments over time. (2) Appeals to implicit non-verbal cues as safeguards against paternalism misstate their value: when such cues contradict speech they commonly generate confusion and mistrust. By contrast, LLM communication is configurable; patients can make the level of guidance an explicit, revisable choice, enhancing autonomy. (3) Evidence that LLM responses are often rated more empathetic than clinicians disrupts the 'technical AI/empathic human' dichotomy. Moreover, clinical trust is multifaceted, frequently grounded in perceived competence and clarity, not (contra Hildebrand) shared vulnerability. Finally, because consent details are routinely forgotten, an on-demand explainer can improve comprehension and disclosure. While reliability, accountability and privacy remain decisive constraints, measured against real-world practice, careful LLM integration promises more equitable, patient-centred communication-not its erosion.