Our bodies remember everything.
I've always suspected that our daily choices — what we eat, how we move, how we live — shape our health in ways that go far deeper than we realise. It's guided how I try to live, even when I couldn't fully comprehend why.
Today I read a study in Nature that made me sit up and pay attention. Scientists have taught an AI called Delphi-2M to read the health stories of over 400,000 people — and what it discovered is remarkable.
The AI learned something we've long intuited but never quite proven: our bodies remember everything. Not just dramatically, but quietly, persistently. A health event today doesn't just affect you today — it ripples forward for decades, connecting to seemingly unrelated conditions in ways we're only beginning to understand.
Here's what's fascinating: they trained this AI primarily on disease patterns, with just basic lifestyle factors like BMI, smoking, and drinking. Yet it still learned to predict health outcomes across 1,000+ different diseases with remarkable accuracy. When they tested it on 1.9 million people in Denmark — people the AI had never "seen" before — it still worked.
The AI essentially discovered something profound: human health isn't a collection of separate problems to solve, but an interconnected web where everything influences everything else. Some effects fade quickly — others, like cancer, cast shadows that last for years.
It's not at a stage where it can be clinically deployed. There are biases to sort out, regulations to navigate. But such studies move toward something profound: truly personalised prevention. Instead of generic advice, we might soon understand exactly how each person's choices compound over decades.
For me, who's always believed nutrition and lifestyle matter deeply, watching AI discover these hidden connections purely from data feels like validation and revelation combined.
We're approaching a world where preventing disease could become as precise as treating it.
Full study: Nature — Learning the natural history of human disease with generative transformers →
Originally published on LinkedIn · View discussion →