How we think about this

Medicine runs on guesswork

Almost everyone has a version of the same story, either their own or one belonging to someone close to them. A person gets sick. They see a doctor, and then a specialist, and then another specialist. The tests come back unremarkable. Therapies get tried in sequence, and when one does not work, the next is tried. Months pass, sometimes years. Often they are told the problem is stress, or anxiety, or that there is nothing wrong. Some of them slowly improve. Many simply grow tired of the search and settle for whatever state they have been left in.

We do not think clinicians are at fault. They are handed an impossible task: absorb a volume of knowledge no person could hold, retain all of it, and then recall exactly the relevant part under time pressure, for every patient who comes through the door. Nobody can do that. The expectation itself is the problem.

We sample biology too sparsely

Consider how a continuous sound becomes a digital recording. Everything depends on the sampling rate — how often you measure the wave in order to reconstruct it. Sample often enough and the recording is indistinguishable from the original. Sample too rarely and what comes back is unrecognizable.

Imagine trying to reconstruct a piece of music from one sample every forty-five seconds.

That is close to the resolution at which we observe human biology. A handful of measurements, taken at long intervals, from a system that is densely interconnected, in constant flux, and influenced by nearly everything happening around it. Then we reason about that system as though the picture were complete.

Research runs into a version of the same constraint. A study is typically built to measure the effect of one factor on one other factor — a necessary simplification, and a severe one, given that the subject is a tapestry of systems acting on each other continuously. At the rate we are currently able to study the body, we cannot expect to understand it.

Disease labels are discrete lines across a continuous system

Diagnostic categories are useful compressions. When enough symptoms and biological findings appear together often enough, naming the cluster makes it possible to study, to communicate, and to treat.

But the compression is lossy, and for a great many people it is simply the wrong shape. Health is a spectrum. There is an enormous range between well and dead, and most people living with chronic illness are somewhere inside it, carrying a particular combination of things functioning imperfectly — a combination no single label fully describes.

What follows from a lossy label is a lossy treatment: a therapy generalized across people with roughly similar biology, chosen because the label was the best description available. The alternative is to understand specifically what is going wrong in a specific person, and to intervene accordingly.

A population average is not a person

Clinical research has a measure called number needed to treat: how many people must receive a treatment for one additional person to benefit. A drug that takes heart attacks from four per hundred people down to three per hundred over five years is a real result, and it is correctly described as a twenty-five percent reduction. It also means a hundred people take the drug for one of them to avoid a heart attack.

The statistic is true, and it is nearly silent on the only question a patient is actually asking. It cannot tell you whether you are the one person who benefits or one of the three for whom nothing changes. It cannot tell you whether your biology resembles the people it helped or the people it did not.

This is not an argument against those drugs. They are the best we have been able to build without the ability to answer that question. It is an argument for building the ability to answer it.

The body is a system that can be understood

The body is a physical system. It follows rules, and it spends its life resisting entropy until eventually it cannot.

That framing is not a grim one; it is the hopeful part. A system that follows rules is, at a sufficient level of detail, knowable. If we understand how a process is supposed to work, we can identify where it has departed from that, and then decide whether and how to intervene. Illness is not mystery. It is a process we have not yet observed closely enough.

Why now

Until recently it was reasonable to do what medicine has done: compress an enormously complex system into categories simple enough for a person to hold in mind, and treat according to them. That was the best available option.

That constraint has changed. We can now build models that learn patterns and form representations of systems far too complex and too high-dimensional for anyone to reason about directly. This does not replace biological insight, and it does not make the measurement problem disappear — a model is only ever as good as what it is given to learn from. What it changes is which questions are worth asking, and how much of the body’s complexity we can afford to keep rather than discard.

Which means nearly everything about how we understand human biology deserves to be reconsidered from first principles.

What we are doing about it

We start with diagnostics, because everything downstream is limited by them.

The discipline we want is the one engineering already has. When a complex system fails, the wrong move is to start changing the parts you suspect and watch whether behaviour improves. The right move is to find the root cause first, and then make the smallest, most targeted change that resolves it — the least disruption to everything that was working.

Today a patient’s outcome depends too heavily on their ability to describe what is happening to them, and on which specialist they happen to see. It should depend on what is measurably true about their biology.

We are at the beginning of this. If this is how you see it too — whether you do research, treat patients, build software, or have lived through the problem yourself — we would like to hear from you. Get in touch.