Why provenance matters more than volume

In healthcare AI, volume has become a proxy for credibility. The size of a dataset is offered as evidence that a system can be trusted, and the inference passes largely unchallenged.

Large datasets are excellent at addressing statistical problems. They reduce sampling bias and they stabilize estimates. They are much less effective at addressing governance problems - lineage ambiguity, unclear consent, regulatory defensibility - because those are questions about where data came from and under what terms, not about how much of it there is.

The most important question is no longer "how much data does this system have?" It is "if one data source were challenged tomorrow, what would actually break?"

That question has an uncomfortable property: in most systems, nobody can answer it. Sources are combined, transformed and re-transformed until the path from a conclusion back to its origins is no longer traceable. When provenance is unclear, volume amplifies risk rather than mitigating it, because each additional source enlarges the surface that cannot be accounted for.

The practical consequence is a design order. Systems that will face scrutiny need to carry their lineage as a first-class property of execution rather than reconstruct it afterward from logs, because reconstruction is exactly what fails under challenge. In AI-driven decision systems, provenance is a safety feature. Volume is not.