AI in Epidemiology: Where It Works, Where It Doesn't
From protein folding to policy: why AI's impact on health follows a U-shaped curve.
Over two weeks I researched where AI actually delivers value in biology and public health — and where it doesn't. What follows is a landscape analysis across three domains (molecular biology, clinical diagnostics, and population health), an interactive simulator I built with Claude to model H. pylori intervention tradeoffs, and a process appendix documenting how I collaborated with AI agents throughout.
Try the simulator — pick a population, adjust three intervention levers, and watch how cancer cases, deaths, and DALYs change. You don't need to understand the biology to see the tradeoffs.
Part 1: The Landscape
AI's track record across biology and health follows a U-shaped curve: it excels where data is standardised and high-volume — protein folding, medical imaging, outbreak surveillance — and bottoms out where human judgement, political negotiation, and behavioural complexity take over. This section maps that curve across molecular biology, clinical diagnostics, and population health, and argues for where epidemiology + AI literacy is the combination that matters.
Part 2: AI-Augmented Capability Demo
I built an interactive H. pylori intervention simulator with Claude — a tool I could not have built on my own. Three moments during the build revealed where domain judgement was the critical input: dropping a decorative model structure, catching a screening artefact disguised as biology, and discovering that the code had drifted from the outcome tree.
Part 3: Process Appendix
How I collaborated with AI agents throughout this project — the working method, the prompting techniques that made outputs auditable rather than trustable, and the three categories of intervention where domain knowledge shaped what the AI produced.