Can AI help us make difficult medical decisions?

Patients, caregivers, and clinicians often struggle to translate personal priorities into concrete treatment decisions—especially when choices involve multiple trade-offs under uncertainty. People may know what matters to them, but integrating these values with complex medical information can exceed normal cognitive capacity

We are developing P4 (Personalised Patient Preference Predictor), an LLM-based “patient co-pilot” designed to support this integration. Rather than treating preferences as fixed, P4 learns what matters to an individual through structured, scenario-based interactions. It captures how priorities—such as longevity, comfort, independence, and burden on others—shift across contexts, and applies this to new medical scenarios to suggest aligned treatment choices.

Importantly, P4 is not just a prediction system. It makes trade-offs explicit, shows which values drive each recommendation, and surfaces tensions or inconsistencies. This reflects a shift from simply providing information to supporting deliberation as it helps users apply what they value to real decisions over time.

In an early prototype, we observed 86.7% agreement between participants and the system. We are now extending this as a proof-of-concept with the aim of helping solo agers, easing caregiver burden, and supporting stakeholders to make more values-concordant decisions in high-stakes contexts.

References:
Earp, B. D., van Veenendaal, T., Porsdam Mann, S., & Savulescu, J. (2025). Digital psychological twins in medicine: addressing risks to human relationships. In Artificial intelligence and the future of human relations: eastern and western perspectives (pp. 239-257). Singapore: Springer Nature Singapore.

Sim, K. Y. H., Foong, P. S., Zhao, C., Quek, M. Y. N., Mehta, S. S., & Choo, K. T. W. (2026, April). Words to Describe What I’m Feeling: Exploring the Potential of AI Agents for High Subjectivity Decisions in Advance Care Planning. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (pp. 1-34).