Abstract:
Artificial intelligence is increasingly being introduced into healthcare to address the growing burden of repetitive clinical and operational activities. However, the key question is not whether artificial intelligence can replace clinical work, but which activities can be responsibly automated, where human oversight must remain essential, and how the patient remains at the centre of care.
This presentation explores a clinician- and patient-centric approach to applying artificial intelligence in oncology, with a particular focus on head and neck cancer care. In longitudinal cancer care, healthcare teams repeatedly collect patient information, conduct follow-up conversations, capture symptoms and patient-reported outcomes, assess quality of life, document clinical interactions and identify patients who require further attention. These activities are essential, but many are structured and repetitive.
Artificial intelligence and conversational systems can potentially support these activities by engaging patients through structured conversations, capturing responses, organizing information and generating longitudinal summaries. Voice-based AI can provide an additional layer of patient engagement, particularly for routine follow-up and structured data collection. However, AI should not independently replace clinical judgment or make treatment decisions.
The proposed model is based on a human-in-the-loop approach. AI performs high-volume, repetitive tasks; clinicians review and interpret the information, determine the significance of findings and make clinical decisions; and patients remain active participants rather than simply becoming sources of data.
The presentation will discuss practical applications across follow-up, patient-reported outcomes, quality-of-life assessments, symptom monitoring, clinical documentation and longitudinal patient engagement in head and neck cancer. It will also address implementation challenges including conversational accuracy, inappropriate AI inference, data quality, escalation pathways, patient safety, transparency, traceability and clinician trust.
The objective is to demonstrate that responsible healthcare AI should not be designed around the question, “What can AI replace?” Instead, the focus should be on “What repetitive work can AI take away from clinicians while preserving human judgment, accountability and meaningful patient interaction?”
The proposed balance is simple: automate the repetitive, augment the complex, keep the clinician accountable, and keep the patient at the centre.
