The Human Function World Model: Modeling the Whole Person Through Human Function Across Tasks, States, Actions, and Longitudinal Change

Artificial intelligence systems for health increasingly model patient states, trajectories, and intervention-conditioned change, yet a persistent whole-person representation remains difficult when evidence spans medicine, rehabilitation, exercise, nutrition, daily activity, and internal physiological demands. This paper presents the Human Function World Model (HFWM), a conceptual and formal AI framework that uses human function as the organizing coordinate for representing the same person across tasks, states, actions, and longitudinal change. Human function retains the UOHF root definition: the body's capacity to be appropriately engaged to meet internal and external demands. HFWM separates Task, Demand, Required Capacity, Current Capacity, the Actual Engagement Process, observed performance/cost/boundary/recovery, evidence, judgment, governed action, and model update. It explicitly distinguishes bodily state from the evidence-conditioned model state, and capacity from actual engagement. Disease, injury, infection, surgery, and endogenous change may trigger governed internal tasks but are not themselves tasks. The current public capacity coordinate system contains 18 core and 104 specific capacities, while governed Functional Bridge relations connect capacities to anatomical and physiological reference knowledge without treating general knowledge as person-specific evidence. Relative to learned AI world models, longitudinal EHR world models, medical world models, physiological world models, ICF, the NIH Whole Person Reference Physiome, and Virtual Human Twins, HFWM's proposed contribution is the explicit demand-capacity-engagement organization of whole-person state and action. Internal requirements-traceability analysis and three detailed worked competency cases demonstrate architectural consistency across infection, post-discharge restoration, and new progressive weakness. These cases are illustrative rather than empirical validation. HFWM currently specifies a representational, evidential, governance, and transition architecture; calibrated prediction, causal counterfactual simulation, and outcome superiority remain empirical research targets.