Precision and Approximation in Digitisation and Transformation of the Individual: Balancing Accuracy and Well-Being in AI-Driven Digital Systems

Abstract

Chronic diseases are largely caused by unhealthy lifestyle choices and behaviours. Early diagnosis and transformative management of chronic diseases are vital for the well-being of the global population. Unfortunately, data regarding the lifestyle choices and behaviours of individuals are sparse, fragmented, or nonexistent. These problems motivated the question of whether we can use both rough and precise data on individuals in a complementary fashion to diagnose and manage chronic diseases, ultimately leading to the well-being and transformation of the individual. We develop a holistic Measure, Model, Manage framework and an AI-driven granularity adaptation framework that learns interpretable mappings between rough self-reported lifestyle data and precise clinical indicators. Using both publicly available datasets and AI-generated synthetic datasets, we compare the robustness of models across varying input granularities. We demonstrate that chronic disease risk can be accurately predicted using not only high-precision biometric inputs but also rough, qualitative data.