Beyond Algorithms: Preserving the Human Core of Palliative Care in the Age of Artificial Intelligence

Artificial intelligence (AI) is rapidly being positioned as the next frontier in healthcare. Predictive models, automated documentation and decision-support systems are increasingly being promoted as solutions to longstanding clinical challenges. In palliative care, however, this enthusiasm warrants careful scrutiny. This is a discipline where uncertainty is not a flaw but a reality and where care is defined not by algorithms, but by relationships. In this context, the challenge is not merely adopting AI, but ensuring that it does not displace the human core of palliative care.

AI has clear potential. It can assist in early identification of symptom burden, risk stratification and prognostication. Machine learning models have demonstrated the ability to detect deterioration earlier than traditional clinical assessment, while natural language processing tools can identify distress signals embedded in routine documentation.^[1,2] In overburdened systems, particularly in low-resource settings, such tools may appear to offer scalability and efficiency.

Yet, there is a growing risk of solution – the belief that complex clinical problems can be solved primarily through technology. Prognostication, for example, is increasingly being framed as a problem of predictive accuracy. However, in palliative care, the challenge is not merely predicting survival, but communicating uncertainty, aligning care with patient values and supporting decision-making in emotionally charged contexts. An accurate prediction that is poorly communicated is not progress; it is harm in another form.^[3]

In the Indian setting, this tension becomes even more pronounced. Most AI models are developed using datasets from high-income countries, with limited representation of the sociocultural and clinical realities seen in India. Using such models without validation in the local context may inadvertently perpetuate existing disparities. There is a danger that AI, rather than democratising care, may create a new form of inequity – where decisions are influenced by tools that do not fully understand the populations they serve. At the same time, the narrative that ‘more technology equals better care’ deserves to be challenged. Experience from Indian palliative care settings suggests that appropriate technology often outperforms advanced technology. Point-of-care ultrasound (POCUS), for instance, has emerged as a practical, bedside tool that enables rapid identification of reversible causes of dyspnoea and ascites, leading to timely interventions and improved clinical decision-making.^[4,5]Its value lies not in complexity, but in clinical immediacy, accessibility and relevance.

The contrast is instructive. While AI promises sophistication, tools like POCUS demonstrate impact through simplicity and integration into routine care. The future of palliative medicine does not lie in choosing one over the other, but in recognising that technology must serve clinical purpose, not define it.

There is also a subtler risk – the gradual erosion of clinical intuition. As decision-support systems become more embedded, clinicians may begin to defer to algorithmic outputs, particularly in high-pressure environments. In palliative care, where nuance and individualisation are central, such shifts could fundamentally alter the nature of care. The concern is not that AI will replace clinicians, but that it may quietly reshape how clinicians think.

In conclusion, AI in palliative care must be approached with measured scepticism and clear intent. Innovation is necessary, but not sufficient. The true test of any technology in this field is not its predictive accuracy or computational power, but whether it enhances patient-centred care without diluting its human core. Progress in palliative medicine will not be defined by how advanced our tools become but by how well we preserve the values that underpin our practice.