How can institutions verify each other’s work without seeing each other’s patients?
This is the problem at the centre of Asuu Network. The papers propose an approach. It has not yet been tested in a hospital.
Open question. Not yet tested.
On this page
Why it is hard.
Patient information cannot simply be made public for inspection. Sharing between institutions needs an appropriate legal basis, permissions, and safeguards. Asuu asks whether a defined claim about model use and authorisation could be checked with less disclosure, and what its privacy limits would be.
This is the problem at the centre of Asuu Network, and the reason the Network was the first thing we began.
What has been tried.
Today the answer is an auditor who visits and is trusted. Research groups train models across hospitals without moving the data, which solves a different problem. The papers propose a shared ledger that carries only fingerprints and mathematical proofs, while every patient's record stays inside the institution that holds it. It is a design on paper. No hospital has run it.
A direction we would test.
We would begin smaller than the papers do. Take one workflow, using invented patients, and ask an outside auditor to reconstruct what happened from the record alone. Then ask the same auditor to do it from an ordinary log that the hospital signed and kept itself, which is the simplest alternative anyone could build.
The comparison is the experiment. A shared record should earn its extra cost in ways a local log cannot: proof that the log was not rewritten after the event, and the ability to compare one institution with another. We would measure the hours the auditor spends, and the hours the hospital's staff spend feeding the system.
How we would know we were wrong.
If the signed local log does the job as well, then for this purpose the Network is unnecessary, and we should say so in public and build something else. That outcome is possible. It is why the evaluation comes before everything that follows it.
Who could solve it.
An auditor who has had to establish what a system did after the fact. The person in a hospital who is responsible for its clinical AI, or for its information systems. A cryptographer who can tell us where the papers' proofs are weakest. And one institution willing to host a small test on invented data.
Sources and scope
This page describes a proposed design and its open questions. The references inform the discussion; they do not validate an Asuu service.
References may be in English.
Where this connects.
Solve one with us.
Tell us which problem you work on, and what we have got wrong about it.